|
Description
|
Well, the first problem with lx_librtld_db is that in most cases it
is not loaded. this is because when librtld_db tries to determine if
the current target is branded, it calls ps_pbrandname() which fails
because of:
6599699 libproc sometimes fails to determine a processes brand
but with this bug fixed, librtld_db does load the lx_librtld_db brand
plugin library. It then tries calling into the plugin library which then
fails with the following error message:
lx_ldb_client_init: couldn't read ehdr
If we use pargs -x to lookup the phdr lx brand aux vector and then
use mdb to print it out we usually see something like this:
---8<---
> fef60034::print Elf32_Phdr
{
p_type = 0x6
p_offset = 0x34
p_vaddr = 0x34
p_paddr = 0x34
p_filesz = 0xe0
p_memsz = 0xe0
p_flags = 0x5
p_align = 0x4
}
---8<---
This means that lx_librtld_db`lx_ldb_client_init() tries to read
from p_vaddr - p_offset, which is usually 0. Hence the failure
and error message.
On live processes, libproc will fail back to using /proc and we'll
manage to get a limited number of symbols. (local symbols are not
accessible via /proc.) but on core files we're left without any
linux symbols at all.
The problem is that the code above is not translating elf header
offsets into memory addresses correctly. it needs to take into
account the addresses that objects were loaded at instead of
just the offsets in the headers themselves.
|
