共享库中R_386_32重定位机制咨询：跨库全局变量引用场景

Let’s walk through exactly how the dynamic linker handles relocation for libbigshr.so and its dependency on libfunlib.so—this is core shared library linking behavior.

1. Initial Library Loading #

When your program (or another library) tries to load libbigshr.so, the dynamic linker (e.g., ld-linux.so on Linux) first parses its dynamic section and discovers it depends on libfunlib.so. The linker:

• Loads libfunlib.so into the process’s virtual address space (at a random base address thanks to ASLR, Address Space Layout Randomization).
• Resolves libfunlib.so’s own relocations first (if any) so its symbols—including the global variable foo—have valid memory addresses.

2. Symbol Resolution for foo #

Next, the linker turns its attention to libbigshr.so’s .rel.dyn section, which lists all data symbols that need relocation. Let’s break down each entry in your .rel.dyn output:

• Offset: The memory offset within libbigshr.so’s loaded image where the reference to foo lives (e.g., 000005a5).
• Info: A 32-bit value split into two parts: the upper 16 bits are the index of foo in libbigshr.so’s dynamic symbol table (.dynsym), and the lower 16 bits specify the relocation type (01 = R_386_32).
• Type (R_386_32): This tells the linker to write the 32-bit absolute address of foo directly into the memory location at the given offset.
• Sym.Value: 00000000 here is a placeholder—this was the "fake" address used at link time, since foo wasn’t defined in libbigshr.so.
• Sym.Name: foo, the symbol the linker needs to find.

The linker looks up foo in the dependency chain: it checks libfunlib.so’s dynamic symbol table (.dynsym) and retrieves the actual virtual address of foo (now set because libfunlib.so was loaded).

3. Executing the Relocations #

For each of the three foo entries in .rel.dyn:

1. Calculate the full memory address to modify: libbigshr.so’s loaded base address + the entry’s offset (e.g., base + 0x5a5).
2. Write the actual 32-bit address of foo (from libfunlib.so) into that memory location.
3. Repeat this for 0x5ab and 0x5c7—each offset corresponds to a separate place in libbigshr.so where foo is referenced (e.g., three different variable assignments or code lines that use foo).

Key Notes #

• The three entries mean libbigshr.so has three distinct direct references to foo’s absolute address—each requires its own relocation.
• .rel.dyn is used for data symbols (like global variables) because these are typically resolved immediately at load time, unlike function symbols which often use .rel.plt for lazy (on-first-call) binding.
• ASLR ensures that each library loads at a unique address every time the program runs, making dynamic relocation mandatory—static linking wouldn’t need this because addresses are fixed at compile time.

内容的提问来源于stack exchange，提问作者ultimate cause