This patch expands functionality of CONFIG_DEBUG_RODATA to set main
(static) kernel data area as NX.
The following steps are taken to achieve this:
1. Linker script is adjusted so .text always starts and ends on a page boundary
2. Linker script is adjusted so .rodata and .data always start and
end on a page boundary
3. void mark_nxdata_nx(void) added to arch/x86/mm/init.c with actual
functionality: NX is set for all pages from _etext through _end.
4. mark_nxdata_nx() called from free_initmem() (after init has been released)
5. free_init_pages() sets released memory NX in arch/x86/mm/init.c
The results of patch application may be observed in the diff of kernel page
table dumps:
--- data_nx_pt_before.txt 2009-10-13 07:48:59.000000000 -0400
+++ data_nx_pt_after.txt 2009-10-13 07:26:46.000000000 -0400
@@ -2,8 +2,9 @@
0x00000000-0xc0000000 3G pmd
---[ Kernel Mapping ]---
0xc0000000-0xc0100000 1M RW GLB x pte
-0xc0100000-0xc048d000 3636K ro GLB x pte
-0xc048d000-0xc0600000 1484K RW GLB x pte
+0xc0100000-0xc0381000 2564K ro GLB x pte
+0xc0381000-0xc048d000 1072K ro GLB NX pte
+0xc048d000-0xc0600000 1484K RW GLB NX pte
0xc0600000-0xf7800000 882M RW PSE GLB NX pmd
0xf7800000-0xf79fe000 2040K RW GLB NX pte
0xf79fe000-0xf7a00000 8K pte
The patch have been developed for Linux 2.6.31-rc7 x86 by Siarhei Liakh
<[email protected]> and Xuxian Jiang <[email protected]>.
V1: initial patch for 2.6.30
V2: patch for 2.6.31-rc7
V3: moved all code into arch/x86, adjusted credits
V4: fixed ifdef, removed credits from CREDITS
V5: fixed an address calculation bug in mark_nxdata_nx()
V6: added acked-by and PT dump diff to commit log
---
Signed-off-by: Siarhei Liakh <[email protected]>
Signed-off-by: Xuxian Jiang <[email protected]>
Acked-by: Arjan van de Ven <[email protected]>
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
index 78d185d..83ae734 100644
--- a/arch/x86/kernel/vmlinux.lds.S
+++ b/arch/x86/kernel/vmlinux.lds.S
@@ -43,14 +43,14 @@ jiffies_64 = jiffies;
PHDRS {
text PT_LOAD FLAGS(5); /* R_E */
- data PT_LOAD FLAGS(7); /* RWE */
+ data PT_LOAD FLAGS(6); /* RW_ */
#ifdef CONFIG_X86_64
- user PT_LOAD FLAGS(7); /* RWE */
- data.init PT_LOAD FLAGS(7); /* RWE */
+ user PT_LOAD FLAGS(6); /* RW_ */
+ data.init PT_LOAD FLAGS(6); /* RW_ */
#ifdef CONFIG_SMP
- percpu PT_LOAD FLAGS(7); /* RWE */
+ percpu PT_LOAD FLAGS(6); /* RW_ */
#endif
- data.init2 PT_LOAD FLAGS(7); /* RWE */
+ data.init2 PT_LOAD FLAGS(6); /* RW_ */
#endif
note PT_NOTE FLAGS(0); /* ___ */
}
@@ -89,6 +89,8 @@ SECTIONS
IRQENTRY_TEXT
*(.fixup)
*(.gnu.warning)
+ /* .text should occupy whole number of pages */
+ . = ALIGN(PAGE_SIZE);
/* End of text section */
_etext = .;
} :text = 0x9090
@@ -151,6 +153,8 @@ SECTIONS
.data.read_mostly : AT(ADDR(.data.read_mostly) - LOAD_OFFSET) {
*(.data.read_mostly)
+ /* .data should occupy whole number of pages */
+ . = ALIGN(PAGE_SIZE);
/* End of data section */
_edata = .;
}
diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index 0607119..7bfd411 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -423,9 +423,10 @@ void free_init_pages(char *what, unsigned long
begin, unsigned long end)
/*
* We just marked the kernel text read only above, now that
* we are going to free part of that, we need to make that
- * writeable first.
+ * writeable and non-executable first.
*/
set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
+ set_memory_nx(begin, (end - begin) >> PAGE_SHIFT);
printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
@@ -440,11 +441,29 @@ void free_init_pages(char *what, unsigned long
begin, unsigned long end)
#endif
}
+void mark_nxdata_nx(void)
+{
+#ifdef CONFIG_DEBUG_RODATA
+ /*
+ * When this called, init has already been executed and released,
+ * so everything past _etext sould be NX.
+ */
+ unsigned long start = PAGE_ALIGN((unsigned long)(&_etext));
+ unsigned long size = PAGE_ALIGN((unsigned long)(&_end)) - start;
+
+ printk(KERN_INFO "NX-protecting the kernel data: %lx, %lu pages\n",
+ start, size >> PAGE_SHIFT);
+ set_memory_nx(start, size >> PAGE_SHIFT);
+#endif
+}
+
void free_initmem(void)
{
free_init_pages("unused kernel memory",
(unsigned long)(&__init_begin),
(unsigned long)(&__init_end));
+ /* Set kernel's data as NX */
+ mark_nxdata_nx();
}
#ifdef CONFIG_BLK_DEV_INITRD
* Siarhei Liakh <[email protected]> wrote:
> This patch expands functionality of CONFIG_DEBUG_RODATA to set main
> (static) kernel data area as NX.
thanks - this one looks good now.
Would be nice if you could send this with the patch that also deals with
the first 1 MB.
Ingo
> thanks - this one looks good now.
>
> Would be nice if you could send this with the patch that also deals with
> the first 1 MB.
I do plan to have that first megabyte fixed. But for now, let's keep
these patches separate, as there are many things that rely on that 1MB
area and I want to make sure they all still work properly with the
patch.
* Siarhei Liakh <[email protected]> wrote:
> > thanks - this one looks good now.
> >
> > Would be nice if you could send this with the patch that also deals
> > with the first 1 MB.
>
> I do plan to have that first megabyte fixed. But for now, let's keep
> these patches separate, as there are many things that rely on that 1MB
> area and I want to make sure they all still work properly with the
> patch.
Well, if we are going to touch this area i'd like to see them addressed
together.
The 1MB thing would obviously be a default-off Kconfig option so in that
sense it should not break anything by default.
Ingo
On Fri, Oct 16, 2009 at 8:27 AM, Ingo Molnar <[email protected]> wrote:
>
[...]
> Well, if we are going to touch this area i'd like to see them addressed
> together.
>
> The 1MB thing would obviously be a default-off Kconfig option so in that
> sense it should not break anything by default.
Ingo,
I've got this 1MB thing fixed. I have also adjusted my previous
patches to be compatible with 2.6.33-rc5.
I will be re-posting this patch set to LKML shortly.
Let me know if there is anything else you would like me to adjust.
Thanks.