On Fri, Jul 22, 2011 at 10:36 AM, Kevin Cernekee <[email protected]> wrote:
> On Thu, Jul 21, 2011 at 2:52 AM, manish kumar <[email protected]>
> wrote:
>> if(!libc_malloc)
>> libc_malloc = (void *(*)(size_t))dlsym(RTLD_NEXT, "malloc");
>> dlerror();
>
> You probably want braces here, but that is not a likely cause of the
> reported problem.
I fixed it. :-)
> I have seen cases where uClibc dlsym(RTLD_NEXT, ...) will find an
> undefined symbol and return an unusable pointer. I suspect that this
> is not the behavior intended by the authors. For example, on an
> application linked with these libraries:
>
> $ readelf -s libuClibc-0.9.32.so | grep -E "\<malloc\>"
> 1225: 00036c90 2488 FUNC GLOBAL DEFAULT 7 malloc
>
> $ readelf -s libfoo.so | grep malloc
> 74: 00003140 0 FUNC GLOBAL DEFAULT UND malloc
>
> dlsym(RTLD_NEXT, ...) might find the libfoo entry instead of the
> libuClibc entry. The returned pointer will not really reach malloc().
>
> I noticed that when dlsym() calls _dl_find_hash(), it uses type_class
> 0. If it used type_class ELF_RTYPE_CLASS_DLSYM that would make more
> sense to me (although it would not fix this problem), because nobody
> else seems to pass in ELF_RTYPE_CLASS_DLSYM. If it used type_class
> ELF_RTYPE_CLASS_PLT I do suspect it would fix your problem, but it
> might break other things like data references. check_match() will not
> match an undefined symbol, as long as (type_class & 1) == 1.
>
> Would it be safe to pass in a bitmask of the desired types? ala:
>
> diff --git a/ldso/libdl/libdl.c b/ldso/libdl/libdl.c
> index 52c77b0..e256f86 100644
> --- a/ldso/libdl/libdl.c
> +++ b/ldso/libdl/libdl.c
> @@ -667,7 +667,7 @@ void *dlsym(void *vhandle, const char *name)
> tpnt = NULL;
> if (handle == _dl_symbol_tables)
> tpnt = handle->dyn; /* Only search RTLD_GLOBAL objs if
> global object */
> - ret = _dl_find_hash(name2, handle, tpnt, 0, &sym_ref);
> + ret = _dl_find_hash(name2, handle, tpnt, ELF_RTYPE_CLASS_PLT |
> ELF_RTYPE_CLASS_COPY, &sym_ref);
>
> #if defined(USE_TLS) && USE_TLS && defined SHARED
> if (sym_ref.tpnt) {
>
Well, this fixes the crash. But I don't know if this is the right fix.
If there is a bug in uClibc, there should be an official release. Do
you think I should report this as BUG in the community?
And are there any side-effects of this fix? Frankly speaking, I am not
a uClibc expert.
_______________________________________________
uClibc mailing list
[email protected]
http://lists.busybox.net/mailman/listinfo/uclibc
