On 11/27/18 at 04:39pm, Kees Cook wrote:
> >> >> diff --git a/arch/x86/kernel/machine_kexec_64.c b/arch/x86/kernel/machine_kexec_64.c
> >> >> index 4c8acdfdc5a7..6161d77c5bfb 100644
> >> >> --- a/arch/x86/kernel/machine_kexec_64.c
> >> >> +++ b/arch/x86/kernel/machine_kexec_64.c
> >> >> @@ -356,6 +356,9 @@ void arch_crash_save_vmcoreinfo(void)
> >> >> VMCOREINFO_SYMBOL(init_top_pgt);
> >> >> vmcoreinfo_append_str("NUMBER(pgtable_l5_enabled)=%d\n",
> >> >> pgtable_l5_enabled());
> >> >> +#ifdef CONFIG_RANDOMIZE_BASE
>
> Okay, gotcha. In that case, shouldn't this be CONFIG_RANDOMIZE_MEMORY?
Currently, Kirill added level5 support to x86_64, and kernel with level5
enabled can be boot switched into level4 or level5 with kernel option
"no5lvl". So page_offset_base will be changed accordingly. You can see
code pasted at bottom, DYNAMIC_MEMORY_LAYOUT is added for this change,
not only KASLR, but 5LEVEL.
If only put it inside ifdef CONFIG_RANDOMIZE_MEMORY, change between l4
and l5 will force us to decide page_offset again if
CONFIG_RANDOMIZE_MEMORY or CONFIG_RANDOMIZE_BASE is not set. Besides,
below commit change the starting address of the direct mapping again, if
only judge CONFIG_RANDOMIZE_MEMORY, in case KASLR is disabled, code in
userspace may need many if-else checking as below. So if we pass, better
pass it for all.
get_page_offset()
{
if(get_page_offset_from_vmcoreinfo()) {
xxx //in KASLR case
return;
} else if (check_5level_paging()) {
if (version < 4.21) {
page_offset = 0xff10000000000000;
} else //version > = 4.21
{
page_offset = 0xff11000000000000;
}
} else { //4level
if (version < 2.6.27) {
page_offset = 0xffff810000000000;
} else if (version < 4.21) {
page_offset = 0xffff880000000000;
} else //version > = 4.21
{
page_offset = 0xffff888000000000,;
}
}
}
Sign, seeing above code, I still think that deducing it from
kcore/vmcore elf header is better. Can't we be better to ourselves?
commit d52888aa2753e3063a9d3a0c9f72f94aa9809c15
Author: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
Date: Fri Oct 26 15:28:54 2018 +0300
x86/mm: Move LDT remap out of KASLR region on 5-level paging
[bhe@ linux]$ git describe --contains d52888aa2753e3063a9d3a0c9f72f94aa9809c15
v4.20-rc2~5^2~4
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#ifdef CONFIG_DYNAMIC_MEMORY_LAYOUT
unsigned long page_offset_base __ro_after_init = __PAGE_OFFSET_BASE_L4;
EXPORT_SYMBOL(page_offset_base);
unsigned long vmalloc_base __ro_after_init = __VMALLOC_BASE_L4;
EXPORT_SYMBOL(vmalloc_base);
unsigned long vmemmap_base __ro_after_init = __VMEMMAP_BASE_L4;
EXPORT_SYMBOL(vmemmap_base);
#endif
config DYNAMIC_MEMORY_LAYOUT
bool
---help---
This option makes base addresses of vmalloc and vmemmap as well as
__PAGE_OFFSET movable during boot.
config RANDOMIZE_MEMORY
bool "Randomize the kernel memory sections"
depends on X86_64
depends on RANDOMIZE_BASE
select DYNAMIC_MEMORY_LAYOUT
default RANDOMIZE_BASE
config X86_5LEVEL
bool "Enable 5-level page tables support"
select DYNAMIC_MEMORY_LAYOUT
select SPARSEMEM_VMEMMAP
depends on X86_64
_______________________________________________
kexec mailing list
kexec@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/kexec
