Re: [PATCH v2 01/20] vmcore: refer to e_phoff member explicitly

[Vivek Goyal <vgoyal@redhat.com>]

On Mon, Mar 11, 2013 at 09:31:41AM +0900, HATAYAMA Daisuke wrote:
> From: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
> Subject: Re: [PATCH v2 01/20] vmcore: refer to e_phoff member explicitly
> Date: Sun, 10 Mar 2013 14:46:31 +0800
> 
> > 于 2013年03月05日 15:35, Zhang Yanfei 写道:
> >> 于 2013年03月02日 16:35, HATAYAMA Daisuke 写道:
> <cut>
> > 
> > One minor suggestion.
> > 
> > Previously, when the code assumes program headers are following immediately
> > the ELF header, it uses
> > 
> >  elfcorebuf_sz = sizeof(Elf64_Ehdr) + ehdr.e_phnum * sizeof(Elf64_Phdr);
> > 
> > to calculate the size of ELF header and ELF program headers
> > 
> > This patch avoids the assumption, and uses ehdr.e_phoff to get the program
> > headers' address. But it will read unrelated contents into elfcorebuf if
> > program headers are not following immediately the ELF header. So could the
> > code be:
> > 
> > elfcorebuf_sz = sizeof(Elf64_Ehdr) + ehdr.e_phnum * sizeof(Elf64_Phdr);
> > addr = elfcorehdr_addr + ehdr.e_phoff;
> > memcpy(elfcorebuf, &ehdr, sizeof(Elf64_Ehdr));
> > read_from_oldmem(elfcorebuf + sizeof(Elf64_Ehdr), elfcorebuf_sz -
> >                  sizeof(Elf64_Ehdr), &addr, 0);
> > (Elf64_Ehdr *)elfcorebuf->e_phoff = sizeof(Elf64_Ehdr);
> 
> Thanks. This is not minor suggestion. This is critical. My code is
> completely broken. On ELF, segments and headers other than ELF header
> can occur in any positions. This means program header table can occur
> after segments. So, on terabyte systems, e_phoff can be more than
> terabytes.

Agreed. It is safer to not copy al the bits till e_phoff.

Thanks
Vivek