lkml.org 
[lkml]   [2019]   [Sep]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] kasan: fix the missing underflow in memmove and memcpy with CONFIG_KASAN_GENERIC=y
From
Date
On Fri, 2019-09-27 at 15:07 +0200, Dmitry Vyukov wrote:
> On Fri, Sep 27, 2019 at 5:43 AM Walter Wu <walter-zh.wu@mediatek.com> wrote:
> >
> > memmove() and memcpy() have missing underflow issues.
> > When -7 <= size < 0, then KASAN will miss to catch the underflow issue.
> > It looks like shadow start address and shadow end address is the same,
> > so it does not actually check anything.
> >
> > The following test is indeed not caught by KASAN:
> >
> > char *p = kmalloc(64, GFP_KERNEL);
> > memset((char *)p, 0, 64);
> > memmove((char *)p, (char *)p + 4, -2);
> > kfree((char*)p);
> >
> > It should be checked here:
> >
> > void *memmove(void *dest, const void *src, size_t len)
> > {
> > check_memory_region((unsigned long)src, len, false, _RET_IP_);
> > check_memory_region((unsigned long)dest, len, true, _RET_IP_);
> >
> > return __memmove(dest, src, len);
> > }
> >
> > We fix the shadow end address which is calculated, then generic KASAN
> > get the right shadow end address and detect this underflow issue.
> >
> > [1] https://bugzilla.kernel.org/show_bug.cgi?id=199341
> >
> > Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
> > Reported-by: Dmitry Vyukov <dvyukov@google.com>
> > ---
> > lib/test_kasan.c | 36 ++++++++++++++++++++++++++++++++++++
> > mm/kasan/generic.c | 8 ++++++--
> > 2 files changed, 42 insertions(+), 2 deletions(-)
> >
> > diff --git a/lib/test_kasan.c b/lib/test_kasan.c
> > index b63b367a94e8..8bd014852556 100644
> > --- a/lib/test_kasan.c
> > +++ b/lib/test_kasan.c
> > @@ -280,6 +280,40 @@ static noinline void __init kmalloc_oob_in_memset(void)
> > kfree(ptr);
> > }
> >
> > +static noinline void __init kmalloc_oob_in_memmove_underflow(void)
> > +{
> > + char *ptr;
> > + size_t size = 64;
> > +
> > + pr_info("underflow out-of-bounds in memmove\n");
> > + ptr = kmalloc(size, GFP_KERNEL);
> > + if (!ptr) {
> > + pr_err("Allocation failed\n");
> > + return;
> > + }
> > +
> > + memset((char *)ptr, 0, 64);
> > + memmove((char *)ptr, (char *)ptr + 4, -2);
> > + kfree(ptr);
> > +}
> > +
> > +static noinline void __init kmalloc_oob_in_memmove_overflow(void)
> > +{
> > + char *ptr;
> > + size_t size = 64;
> > +
> > + pr_info("overflow out-of-bounds in memmove\n");
> > + ptr = kmalloc(size, GFP_KERNEL);
> > + if (!ptr) {
> > + pr_err("Allocation failed\n");
> > + return;
> > + }
> > +
> > + memset((char *)ptr, 0, 64);
> > + memmove((char *)ptr + size, (char *)ptr, 2);
> > + kfree(ptr);
> > +}
> > +
> > static noinline void __init kmalloc_uaf(void)
> > {
> > char *ptr;
> > @@ -734,6 +768,8 @@ static int __init kmalloc_tests_init(void)
> > kmalloc_oob_memset_4();
> > kmalloc_oob_memset_8();
> > kmalloc_oob_memset_16();
> > + kmalloc_oob_in_memmove_underflow();
> > + kmalloc_oob_in_memmove_overflow();
> > kmalloc_uaf();
> > kmalloc_uaf_memset();
> > kmalloc_uaf2();
> > diff --git a/mm/kasan/generic.c b/mm/kasan/generic.c
> > index 616f9dd82d12..34ca23d59e67 100644
> > --- a/mm/kasan/generic.c
> > +++ b/mm/kasan/generic.c
> > @@ -131,9 +131,13 @@ static __always_inline bool memory_is_poisoned_n(unsigned long addr,
> > size_t size)
> > {
> > unsigned long ret;
> > + void *shadow_start = kasan_mem_to_shadow((void *)addr);
> > + void *shadow_end = kasan_mem_to_shadow((void *)addr + size - 1) + 1;
> >
> > - ret = memory_is_nonzero(kasan_mem_to_shadow((void *)addr),
> > - kasan_mem_to_shadow((void *)addr + size - 1) + 1);
> > + if ((long)size < 0)
> > + shadow_end = kasan_mem_to_shadow((void *)addr + size);
>
> Hi Walter,
>
> Thanks for working on this.
>
> If size<0, does it make sense to continue at all? We will still check
> 1PB of shadow memory? What happens when we pass such huge range to
> memory_is_nonzero?
> Perhaps it's better to produce an error and bail out immediately if size<0?

I agree with what you said. when size<0, it is indeed an unreasonable
behavior, it should be blocked from continuing to do.


> Also, what's the failure mode of the tests? Didn't they badly corrupt
> memory? We tried to keep tests such that they produce the KASAN
> reports, but don't badly corrupt memory b/c/ we need to run all of
> them.

Maybe we should first produce KASAN reports and then go to execute
memmove() or do nothing? It looks like it’s doing the following.or?

void *memmove(void *dest, const void *src, size_t len)
{
+ if (long(len) <= 0)
+ kasan_report_invalid_size(src, dest, len, _RET_IP_);
+
check_memory_region((unsigned long)src, len, false, _RET_IP_);
check_memory_region((unsigned long)dest, len, true, _RET_IP_);



\
 
 \ /
  Last update: 2019-09-27 16:22    [W:0.074 / U:0.920 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site