Messages in this thread | | | Date | Wed, 25 Mar 2015 19:29:26 +0100 | From | Hector Marco <> | Subject | Re: [PATCH] mm/x86: AMD Bulldozer ASLR fix |
| |
El 24/03/15 a las 20:15, Borislav Petkov escribió: > On Tue, Mar 24, 2015 at 07:00:48PM +0100, Hector Marco-Gisbert wrote: >> diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c >> index 15c5df9..a693d54 100644 >> --- a/arch/x86/kernel/cpu/amd.c >> +++ b/arch/x86/kernel/cpu/amd.c >> @@ -5,6 +5,7 @@ >> >> #include <linux/io.h> >> #include <linux/sched.h> >> +#include <linux/random.h> >> #include <asm/processor.h> >> #include <asm/apic.h> >> #include <asm/cpu.h> >> @@ -18,6 +19,8 @@ >> >> #include "cpu.h" >> >> +unsigned long rnd_bulldozer_bits = 0; >> + >> static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p) >> { >> u32 gprs[8] = { 0 }; >> @@ -488,6 +491,8 @@ static void bsp_init_amd(struct cpuinfo_x86 *c) >> >> va_align.mask = (upperbit - 1) & PAGE_MASK; >> va_align.flags = ALIGN_VA_32 | ALIGN_VA_64; >> + /* A random value per boot for bits 12,13 and 14 */ >> + rnd_bulldozer_bits = get_random_int() & va_align.mask; > > Hmm, this should be done differently: > > va_align should have a ->bits member which gets ORed in into the hole > made my va_align.mask... >
Yes, It looks better using va_align.
>> } >> } >> >> diff --git a/arch/x86/kernel/sys_x86_64.c b/arch/x86/kernel/sys_x86_64.c >> index 30277e2..5b8ad01 100644 >> --- a/arch/x86/kernel/sys_x86_64.c >> +++ b/arch/x86/kernel/sys_x86_64.c >> @@ -18,6 +18,7 @@ >> >> #include <asm/ia32.h> >> #include <asm/syscalls.h> >> +#include <asm/amd_15h.h> >> >> /* >> * Align a virtual address to avoid aliasing in the I$ on AMD F15h. >> @@ -34,10 +35,16 @@ static unsigned long get_align_mask(void) >> return va_align.mask; >> } >> >> +static unsigned long get_bulldozer_bits(void){ >> + >> + return rnd_bulldozer_bits & get_align_mask(); >> +} >> + >> unsigned long align_vdso_addr(unsigned long addr) >> { >> unsigned long align_mask = get_align_mask(); >> - return (addr + align_mask) & ~align_mask; >> + addr = (addr + align_mask) & ~align_mask; >> + return addr | get_bulldozer_bits(); >> } >> >> static int __init control_va_addr_alignment(char *str) >> @@ -137,7 +144,10 @@ arch_get_unmapped_area(struct file *filp, unsigned long addr, >> info.high_limit = end; >> info.align_mask = filp ? get_align_mask() : 0; > > info.align_bits = get_align_bits() : 0; >
I see your point. The drawback of adding a new field (align_bits) to the info struct is that all architectures need to initialize the info.align_bits. In addition, the generic functions unmapped_area()/unmapped_area_topdown() in file mm/mmap.c, need to be modified to set the bits [12..14] using the new field info.align_bits.
A possible alternative which does not add a new field, is to use the "align_offset" variable. By adding the "get_align_bits()" value to the "align_offset" the bits [12..14] are randomized per boot.
Patch coming. Hector.
>> info.align_offset = pgoff << PAGE_SHIFT; >> - return vm_unmapped_area(&info); >> + addr = vm_unmapped_area(&info); >> + if (!(addr & ~PAGE_MASK)) >> + return filp ? (addr|get_bulldozer_bits()) : addr; >> + return addr; >> } >> >> unsigned long >> @@ -178,7 +188,7 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, >> info.align_offset = pgoff << PAGE_SHIFT; >> addr = vm_unmapped_area(&info); >> if (!(addr & ~PAGE_MASK)) >> - return addr; >> + return filp ? (addr|get_bulldozer_bits()) : addr; > > Ditto. >
| |