ARM64's pfn_valid() shifts away the upper PAGE_SHIFT bits of the input
before seeing if the PFN is valid. This leads to false positives when
some of the upper bits are set, but the lower bits match a valid PFN.
For example, the following userspace code looks up a bogus entry in
/proc/kpageflags:
int pagemap = open("/proc/self/pagemap", O_RDONLY);
int pageflags = open("/proc/kpageflags", O_RDONLY);
uint64_t pfn, val;
lseek64(pagemap, [...], SEEK_SET);
read(pagemap, &pfn, sizeof(pfn));
if (pfn & (1UL << 63)) { /* valid PFN */
pfn &= ((1UL << 55) - 1); /* clear flag bits */
pfn |= (1UL << 55);
lseek64(pageflags, pfn * sizeof(uint64_t), SEEK_SET);
read(pageflags, &val, sizeof(val));
}
On ARM64 this causes the userspace process to crash with SIGSEGV rather
than reading (1 << KPF_NOPAGE). kpageflags_read() treats the offset as
valid, and stable_page_flags() will try to access an address between the
user and kernel address ranges.
Signed-off-by: Greg Hackmann <[email protected]>
---
arch/arm64/mm/init.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
index 9abf8a1e7b25..787e27964ab9 100644
--- a/arch/arm64/mm/init.c
+++ b/arch/arm64/mm/init.c
@@ -287,7 +287,11 @@ static void __init zone_sizes_init(unsigned long min, unsigned long max)
#ifdef CONFIG_HAVE_ARCH_PFN_VALID
int pfn_valid(unsigned long pfn)
{
- return memblock_is_map_memory(pfn << PAGE_SHIFT);
+ phys_addr_t addr = pfn << PAGE_SHIFT;
+
+ if ((addr >> PAGE_SHIFT) != pfn)
+ return 0;
+ return memblock_is_map_memory(addr);
}
EXPORT_SYMBOL(pfn_valid);
#endif
--
2.18.0.597.ga71716f1ad-goog
Hi Greg,
On Mon, Aug 13, 2018 at 12:30:11PM -0700, Greg Hackmann wrote:
> ARM64's pfn_valid() shifts away the upper PAGE_SHIFT bits of the input
> before seeing if the PFN is valid. This leads to false positives when
> some of the upper bits are set, but the lower bits match a valid PFN.
>
> For example, the following userspace code looks up a bogus entry in
> /proc/kpageflags:
>
> int pagemap = open("/proc/self/pagemap", O_RDONLY);
> int pageflags = open("/proc/kpageflags", O_RDONLY);
> uint64_t pfn, val;
>
> lseek64(pagemap, [...], SEEK_SET);
> read(pagemap, &pfn, sizeof(pfn));
> if (pfn & (1UL << 63)) { /* valid PFN */
> pfn &= ((1UL << 55) - 1); /* clear flag bits */
> pfn |= (1UL << 55);
> lseek64(pageflags, pfn * sizeof(uint64_t), SEEK_SET);
> read(pageflags, &val, sizeof(val));
> }
>
> On ARM64 this causes the userspace process to crash with SIGSEGV rather
> than reading (1 << KPF_NOPAGE). kpageflags_read() treats the offset as
> valid, and stable_page_flags() will try to access an address between the
> user and kernel address ranges.
>
> Signed-off-by: Greg Hackmann <[email protected]>
> ---
> arch/arm64/mm/init.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
Thanks, this looks like a sensible fix to me. Do you think it warrants a
CC stable?
Will
> diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
> index 9abf8a1e7b25..787e27964ab9 100644
> --- a/arch/arm64/mm/init.c
> +++ b/arch/arm64/mm/init.c
> @@ -287,7 +287,11 @@ static void __init zone_sizes_init(unsigned long min, unsigned long max)
> #ifdef CONFIG_HAVE_ARCH_PFN_VALID
> int pfn_valid(unsigned long pfn)
> {
> - return memblock_is_map_memory(pfn << PAGE_SHIFT);
> + phys_addr_t addr = pfn << PAGE_SHIFT;
> +
> + if ((addr >> PAGE_SHIFT) != pfn)
> + return 0;
> + return memblock_is_map_memory(addr);
> }
> EXPORT_SYMBOL(pfn_valid);
> #endif
> --
> 2.18.0.597.ga71716f1ad-goog
>
On 08/14/2018 03:40 AM, Will Deacon wrote:
> Hi Greg,
>
> On Mon, Aug 13, 2018 at 12:30:11PM -0700, Greg Hackmann wrote:
>> ARM64's pfn_valid() shifts away the upper PAGE_SHIFT bits of the input
>> before seeing if the PFN is valid. This leads to false positives when
>> some of the upper bits are set, but the lower bits match a valid PFN.
>>
>> For example, the following userspace code looks up a bogus entry in
>> /proc/kpageflags:
>>
>> int pagemap = open("/proc/self/pagemap", O_RDONLY);
>> int pageflags = open("/proc/kpageflags", O_RDONLY);
>> uint64_t pfn, val;
>>
>> lseek64(pagemap, [...], SEEK_SET);
>> read(pagemap, &pfn, sizeof(pfn));
>> if (pfn & (1UL << 63)) { /* valid PFN */
>> pfn &= ((1UL << 55) - 1); /* clear flag bits */
>> pfn |= (1UL << 55);
>> lseek64(pageflags, pfn * sizeof(uint64_t), SEEK_SET);
>> read(pageflags, &val, sizeof(val));
>> }
>>
>> On ARM64 this causes the userspace process to crash with SIGSEGV rather
>> than reading (1 << KPF_NOPAGE). kpageflags_read() treats the offset as
>> valid, and stable_page_flags() will try to access an address between the
>> user and kernel address ranges.
>>
>> Signed-off-by: Greg Hackmann <[email protected]>
>> ---
>> arch/arm64/mm/init.c | 6 +++++-
>> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> Thanks, this looks like a sensible fix to me. Do you think it warrants a
> CC stable?
>
> Will
Yes, I think so. Should I resend with a "Fixes" field?
On Tue, Aug 14, 2018 at 08:17:48AM -0700, Greg Hackmann wrote:
> On 08/14/2018 03:40 AM, Will Deacon wrote:
> > Hi Greg,
> >
> > On Mon, Aug 13, 2018 at 12:30:11PM -0700, Greg Hackmann wrote:
> >> ARM64's pfn_valid() shifts away the upper PAGE_SHIFT bits of the input
> >> before seeing if the PFN is valid. This leads to false positives when
> >> some of the upper bits are set, but the lower bits match a valid PFN.
> >>
> >> For example, the following userspace code looks up a bogus entry in
> >> /proc/kpageflags:
> >>
> >> int pagemap = open("/proc/self/pagemap", O_RDONLY);
> >> int pageflags = open("/proc/kpageflags", O_RDONLY);
> >> uint64_t pfn, val;
> >>
> >> lseek64(pagemap, [...], SEEK_SET);
> >> read(pagemap, &pfn, sizeof(pfn));
> >> if (pfn & (1UL << 63)) { /* valid PFN */
> >> pfn &= ((1UL << 55) - 1); /* clear flag bits */
> >> pfn |= (1UL << 55);
> >> lseek64(pageflags, pfn * sizeof(uint64_t), SEEK_SET);
> >> read(pageflags, &val, sizeof(val));
> >> }
> >>
> >> On ARM64 this causes the userspace process to crash with SIGSEGV rather
> >> than reading (1 << KPF_NOPAGE). kpageflags_read() treats the offset as
> >> valid, and stable_page_flags() will try to access an address between the
> >> user and kernel address ranges.
> >>
> >> Signed-off-by: Greg Hackmann <[email protected]>
> >> ---
> >> arch/arm64/mm/init.c | 6 +++++-
> >> 1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > Thanks, this looks like a sensible fix to me. Do you think it warrants a
> > CC stable?
> >
> > Will
>
> Yes, I think so. Should I resend with a "Fixes" field?
Could do, but I think this goes all the way back to day 1! Doesn't arch/arm/
also suffer from the same issue?
Will
On 08/14/2018 08:29 AM, Will Deacon wrote:
> On Tue, Aug 14, 2018 at 08:17:48AM -0700, Greg Hackmann wrote:
>> On 08/14/2018 03:40 AM, Will Deacon wrote:
>>> Hi Greg,
>>>
>>> On Mon, Aug 13, 2018 at 12:30:11PM -0700, Greg Hackmann wrote:
>>>> ARM64's pfn_valid() shifts away the upper PAGE_SHIFT bits of the input
>>>> before seeing if the PFN is valid. This leads to false positives when
>>>> some of the upper bits are set, but the lower bits match a valid PFN.
>>>>
>>>> For example, the following userspace code looks up a bogus entry in
>>>> /proc/kpageflags:
>>>>
>>>> int pagemap = open("/proc/self/pagemap", O_RDONLY);
>>>> int pageflags = open("/proc/kpageflags", O_RDONLY);
>>>> uint64_t pfn, val;
>>>>
>>>> lseek64(pagemap, [...], SEEK_SET);
>>>> read(pagemap, &pfn, sizeof(pfn));
>>>> if (pfn & (1UL << 63)) { /* valid PFN */
>>>> pfn &= ((1UL << 55) - 1); /* clear flag bits */
>>>> pfn |= (1UL << 55);
>>>> lseek64(pageflags, pfn * sizeof(uint64_t), SEEK_SET);
>>>> read(pageflags, &val, sizeof(val));
>>>> }
>>>>
>>>> On ARM64 this causes the userspace process to crash with SIGSEGV rather
>>>> than reading (1 << KPF_NOPAGE). kpageflags_read() treats the offset as
>>>> valid, and stable_page_flags() will try to access an address between the
>>>> user and kernel address ranges.
>>>>
>>>> Signed-off-by: Greg Hackmann <[email protected]>
>>>> ---
>>>> arch/arm64/mm/init.c | 6 +++++-
>>>> 1 file changed, 5 insertions(+), 1 deletion(-)
>>>
>>> Thanks, this looks like a sensible fix to me. Do you think it warrants a
>>> CC stable?
>>>
>>> Will
>>
>> Yes, I think so. Should I resend with a "Fixes" field?
>
> Could do, but I think this goes all the way back to day 1! Doesn't arch/arm/
> also suffer from the same issue?
>
> Will
>
Yeah, it looks like this happens on non-LPAE 32-bit kernels too. LPAE
kernels aren't affected since __pfn_to_phys() promotes to a 64-bit type.
I can submit a fix for that too while I'm at it.