Re: uclibc mips ld.so and undefined symbols with nonzero symbol table entry st_value

On Mon, May 10, 2004 at 11:39:41PM -0400, Bradley D. LaRonde wrote:

----- Original Message ----- 
From: "Richard Sandiford" <redacted>
To: "Bradley D. LaRonde" <redacted>
Cc: <redacted>; <redacted>
Sent: Monday, May 10, 2004 4:41 PM
Subject: Re: uclibc mips ld.so and undefined symbols with nonzero symbol
table entry st_value

quoted

"Bradley D. LaRonde" [off-list ref] writes:

quoted

I read this in the spec:

    All externally visible symbols, both defined and undefined,
    must be hashed into the hash table.

Should libpthread's malloc stub be added to the hash table?

Yes.

quoted

I guess not, but I think that might be happening (haven't verified),
and libdl finding it in there and thinking it is the real deal, not
realizing it is just a stub.

If you have an undefined function symbol with st_value != 0, then
that st_value must be for a stub.  That's how the loader can (and is
supposed to) tell the difference.

It's probably a good idea to look at how glibc handles this.

uClibc/ldso/ldso/mips/elfinterp.c around line 288 looks like this:


    /* Relocate the global GOT entries for the object */
    while(i--) {
      if (sym->st_shndx == SHN_UNDEF) {
        if (ELF32_ST_TYPE(sym->st_info) == STT_FUNC && sym->st_value)
          *got_entry = sym->st_value + (unsigned long) tpnt->loadaddr;
        else {
          *got_entry = (unsigned long) _dl_find_hash(strtab +
             sym->st_name, tpnt->symbol_scope, ELF_RTYPE_CLASS_COPY);
        }
     }


If I change that ELF_RTYPE_CLASS_COPY to ELF_RTYPE_CLASS_PLT to tell
_dl_find_hash to ignore stubs when resolving undefined functions without
stubs, the dlopen tests all pass.  dlopen gets a pointer to the libc.so
malloc instead of a pointer to the libpthread malloc stub.  Yay!  :-)

Does that look like the correct fix?

Probably, since MIPS doesn't have a copy reloc.

-- 
Daniel Jacobowitz