COMS 4995 Advanced Systems Programming

Linking

ELF Overview

• Review of 15-elf-intro
• ELF Executable Layout
  • Ehdr
    • e_shoff
    • e_shstrndx
    • HW 5 Part 1: gen-exe
  • Phdr table
    • Segment 0 contains Ehdr & Phdr table
    • A segment contains one or more sections
      • readelf -a ./main (in code/13) shows .data & .bss sections in the same segment
  • .symtab & .strtab
    • st_name
    • st_value
    • st_shndx
  • Shdr table & .shstrtab
    • sh_name`
    • sh_size
    • sh_offset

Linking an Assembler-Generated Object File

• hello program (code/16/hello)
  • code walkthrough
  • ld-generated executable: readelf -a hello
  • cld-generated executable: readelf -a a.out
    • compare final cld linker to ld: pretty much the same
    • compare final cld linker to part 2 cld: missing strings & symbols
  • Alignment
    • Segments start addresses must be aligned to page boundary
    • cld aligns .symtab start offset to page boundary for simplicity

Linking a gcc-Generated Object File

my_libc library

• Linking GNU C stdlib (glibc) is very complex (regardless of static vs. dynamic linking)
• We provide our own minimal C library that contains:
  • Startup routine that calls main()
  • write() & exit() syscalls built on top of our own syscall() wrapper
• echo1 program walkthrough (code/16/echo1)

Single Object File Relocation

• readelf -a echo.o
  • relocation entries
  • Elf64_Rela
• objdump -d echo.o
  • Inspect placeholder call instruction to my_strlen()
• Calculate PC-relative encoding for operand to call instruction
  • Verify relocation result by comparing with objdump -d echo to see completed relocation

Linking Multiple Object Files

Each object file will have a mixture of symbol definitions and external symbol references in its symbol table. External symbol references refer to symbols define in some other file and appear as undefined symbols in the referencer’s symbol table. The linker’s job is to merge the input .text sections and symbols into a single executable. Here’s an outline of the steps to be performed:

• Concatenate the input .text sections into a single output .text section
• Collect all symbols from multiple input symbol tables and generate an output symbol table, the union of the input symbol tables
• Create a mapping from input symbols to output symbols to use for relocation later. Multiple references to the same symbol should map to the same output symbol.
  • Linker emits error if there are multiple definitions of the same symbol
  • Linker emits error if there are undefined symbols after processing all input object files
• Generate a new output string table for the output symbol table
• Adjust the following fields in the output symbol:
  • st_name: Refer to the newly generated string table
  • st_value: Refer to the memory address of the symbol in the output .text section, which is the concatenation of all the input .text sections

---

Last updated: 2024-04-03