Did you have any chance to test the following hack, Anshuman?
On Thu 01-02-18 14:48:29, Michal Hocko wrote:
[...]
> Just a blind shot... Does the following make any difference?
> ---
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index 021fe78998ea..04b24d00c911 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -895,7 +895,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
> the correct location in memory. */
> for(i = 0, elf_ppnt = elf_phdata;
> i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
> - int elf_prot = 0, elf_flags;
> + int elf_prot = 0, elf_flags, elf_fixed = MAP_FIXED_NOREPLACE;
> unsigned long k, vaddr;
> unsigned long total_size = 0;
>
> @@ -927,6 +927,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
> */
> }
> }
> + elf_fixed = MAP_FIXED;
> }
>
> if (elf_ppnt->p_flags & PF_R)
> @@ -944,7 +945,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
> * the ET_DYN load_addr calculations, proceed normally.
> */
> if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
> - elf_flags |= MAP_FIXED_NOREPLACE;
> + elf_flags |= elf_fixed;
> } else if (loc->elf_ex.e_type == ET_DYN) {
> /*
> * This logic is run once for the first LOAD Program
> @@ -980,7 +981,7 @@ static int load_elf_binary(struct linux_binprm *bprm)
> load_bias = ELF_ET_DYN_BASE;
> if (current->flags & PF_RANDOMIZE)
> load_bias += arch_mmap_rnd();
> - elf_flags |= MAP_FIXED_NOREPLACE;
> + elf_flags |= elf_fixed;
> } else
> load_bias = 0;
>
--
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
