Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

Please cc linux-api for user visible API proposals (now done). Keep the
rest of the email intact for reference.

On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> From: Matthew Garrett <mjg59@google.com>
> 
> Applications that hold secrets and wish to avoid them leaking can use
> mlock() to prevent the page from being pushed out to swap and
> MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> can also use atexit() handlers to overwrite secrets on application exit.
> However, if an attacker can reboot the system into another OS, they can
> dump the contents of RAM and extract secrets. We can avoid this by setting
> CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> firmware wipe the contents of RAM before booting another OS, but this means
> rebooting takes a *long* time - the expected behaviour is for a clean
> shutdown to remove the request after scrubbing secrets from RAM in order to
> avoid this.
> 
> Unfortunately, if an application exits uncleanly, its secrets may still be
> present in RAM. This can't be easily fixed in userland (eg, if the OOM
> killer decides to kill a process holding secrets, we're not going to be able
> to avoid that), so this patch adds a new flag to madvise() to allow userland
> to request that the kernel clear the covered pages whenever the page
> reference count hits zero. Since vm_flags is already full on 32-bit, it
> will only work on 64-bit systems.
> 
> Signed-off-by: Matthew Garrett <mjg59@google.com>
> ---
> 
> Modified to wipe when the VMA is released rather than on page freeing
> 
>  include/linux/mm.h                     |  6 ++++++
>  include/uapi/asm-generic/mman-common.h |  2 ++
>  mm/madvise.c                           | 21 +++++++++++++++++++++
>  mm/memory.c                            |  3 +++
>  4 files changed, 32 insertions(+)
> 
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index 6b10c21630f5..64bdab679275 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -257,6 +257,8 @@ extern unsigned int kobjsize(const void *objp);
>  #define VM_HIGH_ARCH_2	BIT(VM_HIGH_ARCH_BIT_2)
>  #define VM_HIGH_ARCH_3	BIT(VM_HIGH_ARCH_BIT_3)
>  #define VM_HIGH_ARCH_4	BIT(VM_HIGH_ARCH_BIT_4)
> +
> +#define VM_WIPEONRELEASE BIT(37)       /* Clear pages when releasing them */
>  #endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
>  
>  #ifdef CONFIG_ARCH_HAS_PKEYS
> @@ -298,6 +300,10 @@ extern unsigned int kobjsize(const void *objp);
>  # define VM_GROWSUP	VM_NONE
>  #endif
>  
> +#ifndef VM_WIPEONRELEASE
> +# define VM_WIPEONRELEASE VM_NONE
> +#endif
> +
>  /* Bits set in the VMA until the stack is in its final location */
>  #define VM_STACK_INCOMPLETE_SETUP	(VM_RAND_READ | VM_SEQ_READ)
>  
> diff --git a/include/uapi/asm-generic/mman-common.h b/include/uapi/asm-generic/mman-common.h
> index abd238d0f7a4..82dfff4a8e3d 100644
> --- a/include/uapi/asm-generic/mman-common.h
> +++ b/include/uapi/asm-generic/mman-common.h
> @@ -64,6 +64,8 @@
>  #define MADV_WIPEONFORK 18		/* Zero memory on fork, child only */
>  #define MADV_KEEPONFORK 19		/* Undo MADV_WIPEONFORK */
>  
> +#define MADV_WIPEONRELEASE 20
> +#define MADV_DONTWIPEONRELEASE 21
>  /* compatibility flags */
>  #define MAP_FILE	0
>  
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 21a7881a2db4..989c2fde15cf 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
>  	case MADV_KEEPONFORK:
>  		new_flags &= ~VM_WIPEONFORK;
>  		break;
> +	case MADV_WIPEONRELEASE:
> +		/* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> +		if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> +		    vma->vm_flags & VM_SHARED) {
> +			error = -EINVAL;
> +			goto out;
> +		}
> +		new_flags |= VM_WIPEONRELEASE;
> +		break;
> +	case MADV_DONTWIPEONRELEASE:
> +		if (VM_WIPEONRELEASE == 0) {
> +			error = -EINVAL;
> +			goto out;
> +		}
> +		new_flags &= ~VM_WIPEONRELEASE;
> +		break;
>  	case MADV_DONTDUMP:
>  		new_flags |= VM_DONTDUMP;
>  		break;
> @@ -727,6 +743,8 @@ madvise_behavior_valid(int behavior)
>  	case MADV_DODUMP:
>  	case MADV_WIPEONFORK:
>  	case MADV_KEEPONFORK:
> +	case MADV_WIPEONRELEASE:
> +	case MADV_DONTWIPEONRELEASE:
>  #ifdef CONFIG_MEMORY_FAILURE
>  	case MADV_SOFT_OFFLINE:
>  	case MADV_HWPOISON:
> @@ -785,6 +803,9 @@ madvise_behavior_valid(int behavior)
>   *  MADV_DONTDUMP - the application wants to prevent pages in the given range
>   *		from being included in its core dump.
>   *  MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump.
> + *  MADV_WIPEONRELEASE - clear the contents of the memory after the last
> + *		reference to it has been released
> + *  MADV_DONTWIPEONRELEASE - cancel MADV_WIPEONRELEASE
>   *
>   * return values:
>   *  zero    - success
> diff --git a/mm/memory.c b/mm/memory.c
> index ab650c21bccd..ff78b527660e 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1091,6 +1091,9 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
>  			page_remove_rmap(page, false);
>  			if (unlikely(page_mapcount(page) < 0))
>  				print_bad_pte(vma, addr, ptent, page);
> +			if (unlikely(vma->vm_flags & VM_WIPEONRELEASE) &&
> +			    page_mapcount(page) == 0)
> +				clear_highpage(page);
>  			if (unlikely(__tlb_remove_page(tlb, page))) {
>  				force_flush = 1;
>  				addr += PAGE_SIZE;
> -- 
> 2.21.0.593.g511ec345e18-goog

-- 
Michal Hocko
SUSE Labs

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

On Thu 25-04-19 14:14:10, Michal Hocko wrote:
> Please cc linux-api for user visible API proposals (now done). Keep the
> rest of the email intact for reference.
> 
> On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > From: Matthew Garrett <mjg59@google.com>
> > 
> > Applications that hold secrets and wish to avoid them leaking can use
> > mlock() to prevent the page from being pushed out to swap and
> > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > can also use atexit() handlers to overwrite secrets on application exit.
> > However, if an attacker can reboot the system into another OS, they can
> > dump the contents of RAM and extract secrets. We can avoid this by setting
> > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > firmware wipe the contents of RAM before booting another OS, but this means
> > rebooting takes a *long* time - the expected behaviour is for a clean
> > shutdown to remove the request after scrubbing secrets from RAM in order to
> > avoid this.
> > 
> > Unfortunately, if an application exits uncleanly, its secrets may still be
> > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > killer decides to kill a process holding secrets, we're not going to be able
> > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > to request that the kernel clear the covered pages whenever the page
> > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > will only work on 64-bit systems.

The changelog seems stale. You are hooking into unmap path where the
reference count might be still > 0 and the page still held by somebody.
A previous email from Willy said
"
It could be the target/source of direct I/O, or userspace could have
registered it with an RDMA device, or ...

It depends on the semantics you want.  There's no legacy code to
worry about here.  I was seeing this as the equivalent of an atexit()
handler; userspace is saying "When this page is unmapped, zero it".
So it doesn't matter that somebody else might be able to reference it --
userspace could have zeroed it themselves.
"

I am not sure this is really a bullet proof argumentation but it should
definitely be part of the changelog.

Besides that you inherently assume that the user would do mlock because
you do not try to wipe the swap content. Is this intentional?

Another question would be regarding the targeted user API. There are
some attempts to make all the freed memory to be zeroed/poisoned. Are
users who would like to use this feature also be interested in using
system wide setting as well?
-- 
Michal Hocko
SUSE Labs

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Matthew Garrett]

On Thu, Apr 25, 2019 at 5:37 AM Michal Hocko <mhocko@kernel.org> wrote:
> Besides that you inherently assume that the user would do mlock because
> you do not try to wipe the swap content. Is this intentional?

Yes, given MADV_DONTDUMP doesn't imply mlock I thought it'd be more
consistent to keep those independent.

> Another question would be regarding the targeted user API. There are
> some attempts to make all the freed memory to be zeroed/poisoned. Are
> users who would like to use this feature also be interested in using
> system wide setting as well?

I think that depends on the performance overhead of a global setting,
but it's also influenced by the semantics around when the freeing
occurs, which is something I haven't nailed down yet. If the
expectation is that the page is freed whenever the process exits, even
if the page is in use somewhere else, then we'd still want this to be
separate to poisoning on final page free.

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

On Thu 25-04-19 13:39:01, Matthew Garrett wrote:
> On Thu, Apr 25, 2019 at 5:37 AM Michal Hocko <mhocko@kernel.org> wrote:
> > Besides that you inherently assume that the user would do mlock because
> > you do not try to wipe the swap content. Is this intentional?
> 
> Yes, given MADV_DONTDUMP doesn't imply mlock I thought it'd be more
> consistent to keep those independent.

Do we want to fail madvise call on VMAs that are not mlocked then? What
if the munlock happens later after the madvise is called?

-- 
Michal Hocko
SUSE Labs

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Matthew Garrett]

On Thu, Apr 25, 2019 at 10:25 PM Michal Hocko <mhocko@kernel.org> wrote:
>
> On Thu 25-04-19 13:39:01, Matthew Garrett wrote:
> > Yes, given MADV_DONTDUMP doesn't imply mlock I thought it'd be more
> > consistent to keep those independent.
>
> Do we want to fail madvise call on VMAs that are not mlocked then? What
> if the munlock happens later after the madvise is called?

I'm not sure if it's strictly necessary. We already have various
combinations of features that only make sense when used together and
which can be undermined by later actions. I can see the appeal of
designing this in a way that makes it harder to misuse, but is that
worth additional implementation complexity?

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

On Fri 26-04-19 11:08:44, Matthew Garrett wrote:
> On Thu, Apr 25, 2019 at 10:25 PM Michal Hocko <mhocko@kernel.org> wrote:
> >
> > On Thu 25-04-19 13:39:01, Matthew Garrett wrote:
> > > Yes, given MADV_DONTDUMP doesn't imply mlock I thought it'd be more
> > > consistent to keep those independent.
> >
> > Do we want to fail madvise call on VMAs that are not mlocked then? What
> > if the munlock happens later after the madvise is called?
> 
> I'm not sure if it's strictly necessary. We already have various
> combinations of features that only make sense when used together and
> which can be undermined by later actions. I can see the appeal of
> designing this in a way that makes it harder to misuse, but is that
> worth additional implementation complexity?

If the complexity is not worth the usual usecases then this should be
really documented and noted that without an mlock you are not getting
the full semantic and you can leave memory behind on the swap partition.

I cannot judge how much that matter but it certainly looks half feature
to me but if nobody is going to use the madvise without mlock then it
looks certainly much easier to implement.
-- 
Michal Hocko
SUSE Labs

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Vlastimil Babka]

On 4/25/19 2:14 PM, Michal Hocko wrote:
> Please cc linux-api for user visible API proposals (now done). Keep the
> rest of the email intact for reference.
> 
> On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
>> From: Matthew Garrett <mjg59@google.com>
>>
>> Applications that hold secrets and wish to avoid them leaking can use
>> mlock() to prevent the page from being pushed out to swap and
>> MADV_DONTDUMP to prevent it from being included in core dumps. Applications

So, do we really need a new madvise() flag and VMA flag, or can we just
infer this page clearing from mlock+MADV_DONTDUMP being both applied?

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Matthew Garrett]

On Thu, Apr 25, 2019 at 5:44 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> On 4/25/19 2:14 PM, Michal Hocko wrote:
> > Please cc linux-api for user visible API proposals (now done). Keep the
> > rest of the email intact for reference.
> >
> > On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> >> From: Matthew Garrett <mjg59@google.com>
> >>
> >> Applications that hold secrets and wish to avoid them leaking can use
> >> mlock() to prevent the page from being pushed out to swap and
> >> MADV_DONTDUMP to prevent it from being included in core dumps. Applications
>
> So, do we really need a new madvise() flag and VMA flag, or can we just
> infer this page clearing from mlock+MADV_DONTDUMP being both applied?

I think the combination would probably imply that this is the
behaviour you want, but I'm a little concerned about changing the
semantics given the corner cases described earlier in the thread. If
we can figure those out in a way that won't break any existing code, I
could buy this.

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Jann Horn]

On Thu, Apr 25, 2019 at 2:14 PM Michal Hocko <mhocko@kernel.org> wrote:
[...]
> On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > From: Matthew Garrett <mjg59@google.com>
> >
> > Applications that hold secrets and wish to avoid them leaking can use
> > mlock() to prevent the page from being pushed out to swap and
> > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > can also use atexit() handlers to overwrite secrets on application exit.
> > However, if an attacker can reboot the system into another OS, they can
> > dump the contents of RAM and extract secrets. We can avoid this by setting
> > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > firmware wipe the contents of RAM before booting another OS, but this means
> > rebooting takes a *long* time - the expected behaviour is for a clean
> > shutdown to remove the request after scrubbing secrets from RAM in order to
> > avoid this.
> >
> > Unfortunately, if an application exits uncleanly, its secrets may still be
> > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > killer decides to kill a process holding secrets, we're not going to be able
> > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > to request that the kernel clear the covered pages whenever the page
> > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > will only work on 64-bit systems.
[...]
> > diff --git a/mm/madvise.c b/mm/madvise.c
> > index 21a7881a2db4..989c2fde15cf 100644
> > --- a/mm/madvise.c
> > +++ b/mm/madvise.c
> > @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
> >       case MADV_KEEPONFORK:
> >               new_flags &= ~VM_WIPEONFORK;
> >               break;
> > +     case MADV_WIPEONRELEASE:
> > +             /* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> > +             if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> > +                 vma->vm_flags & VM_SHARED) {
> > +                     error = -EINVAL;
> > +                     goto out;
> > +             }
> > +             new_flags |= VM_WIPEONRELEASE;
> > +             break;

An interesting effect of this is that it will be possible to set this
on a CoW anon VMA in a fork() child, and then the semantics in the
parent will be subtly different - e.g. if the parent vmsplice()d a
CoWed page into a pipe, then forked an unprivileged child, the child
set MADV_WIPEONRELEASE on its VMA, the parent died somehow, and then
the child died, the page in the pipe would be zeroed out. A child
should not be able to affect its parent like this, I think. If this
was an mmap() flag instead of a madvise() command, that issue could be
avoided. Alternatively, if adding more mmap() flags doesn't work,
perhaps you could scan the VMA and ensure that it contains no pages
yet, or something like that?

> > diff --git a/mm/memory.c b/mm/memory.c
> > index ab650c21bccd..ff78b527660e 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -1091,6 +1091,9 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
> >                       page_remove_rmap(page, false);
> >                       if (unlikely(page_mapcount(page) < 0))
> >                               print_bad_pte(vma, addr, ptent, page);
> > +                     if (unlikely(vma->vm_flags & VM_WIPEONRELEASE) &&
> > +                         page_mapcount(page) == 0)
> > +                             clear_highpage(page);
> >                       if (unlikely(__tlb_remove_page(tlb, page))) {
> >                               force_flush = 1;
> >                               addr += PAGE_SIZE;

Should something like this perhaps be added in page_remove_rmap()
instead? That's where the mapcount is decremented; and looking at
other callers of page_remove_rmap(), in particular the following ones
look interesting:

 - do_huge_pmd_wp_page()/do_huge_pmd_wp_page_fallback() might be
relevant in the case where a forking process contains transparent
hugepages?
 - zap_huge_pmd() is relevant when transparent hugepages are used, I
think (otherwise transparent hugepages might not be wiped?)
 - there are various callers related to migration; I think this is
relevant on a NUMA system where memory is moved between nodes to
improve locality (moving memory to a new page and freeing the old one,
in which case you'd want to wipe the old page)

I think all the callers have a reference to the VMA, so perhaps you
could add a VMA parameter to page_remove_rmap() and then look at the
VMA in there?

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Matthew Garrett]

On Thu, Apr 25, 2019 at 5:43 AM Jann Horn <jannh@google.com> wrote:
> An interesting effect of this is that it will be possible to set this
> on a CoW anon VMA in a fork() child, and then the semantics in the
> parent will be subtly different - e.g. if the parent vmsplice()d a
> CoWed page into a pipe, then forked an unprivileged child, the child
> set MADV_WIPEONRELEASE on its VMA, the parent died somehow, and then
> the child died, the page in the pipe would be zeroed out. A child
> should not be able to affect its parent like this, I think. If this
> was an mmap() flag instead of a madvise() command, that issue could be
> avoided. Alternatively, if adding more mmap() flags doesn't work,
> perhaps you could scan the VMA and ensure that it contains no pages
> yet, or something like that?

I /think/ my argument here would be not to do that? I agree that it's
unexpected, but I guess the other alternative would be to force a copy
on any existing COW pages in the VMA at madvise() time, and maybe also
at fork() time (sort of like the behaviour of MADV_WIPEONFORK, but
copying the page rather than providing a new zero page)

> I think all the callers have a reference to the VMA, so perhaps you
> could add a VMA parameter to page_remove_rmap() and then look at the
> VMA in there?

I'll dig into that, thanks!

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

On Thu 25-04-19 14:42:52, Jann Horn wrote:
> On Thu, Apr 25, 2019 at 2:14 PM Michal Hocko <mhocko@kernel.org> wrote:
> [...]
> > On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > > From: Matthew Garrett <mjg59@google.com>
> > >
> > > Applications that hold secrets and wish to avoid them leaking can use
> > > mlock() to prevent the page from being pushed out to swap and
> > > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > > can also use atexit() handlers to overwrite secrets on application exit.
> > > However, if an attacker can reboot the system into another OS, they can
> > > dump the contents of RAM and extract secrets. We can avoid this by setting
> > > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > > firmware wipe the contents of RAM before booting another OS, but this means
> > > rebooting takes a *long* time - the expected behaviour is for a clean
> > > shutdown to remove the request after scrubbing secrets from RAM in order to
> > > avoid this.
> > >
> > > Unfortunately, if an application exits uncleanly, its secrets may still be
> > > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > > killer decides to kill a process holding secrets, we're not going to be able
> > > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > > to request that the kernel clear the covered pages whenever the page
> > > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > > will only work on 64-bit systems.
> [...]
> > > diff --git a/mm/madvise.c b/mm/madvise.c
> > > index 21a7881a2db4..989c2fde15cf 100644
> > > --- a/mm/madvise.c
> > > +++ b/mm/madvise.c
> > > @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
> > >       case MADV_KEEPONFORK:
> > >               new_flags &= ~VM_WIPEONFORK;
> > >               break;
> > > +     case MADV_WIPEONRELEASE:
> > > +             /* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> > > +             if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> > > +                 vma->vm_flags & VM_SHARED) {
> > > +                     error = -EINVAL;
> > > +                     goto out;
> > > +             }
> > > +             new_flags |= VM_WIPEONRELEASE;
> > > +             break;
> 
> An interesting effect of this is that it will be possible to set this
> on a CoW anon VMA in a fork() child, and then the semantics in the
> parent will be subtly different - e.g. if the parent vmsplice()d a
> CoWed page into a pipe, then forked an unprivileged child, the child

Maybe a stupid question. How do you fork an unprivileged child (without
exec)? Child would have to drop priviledges on its own, no?

> set MADV_WIPEONRELEASE on its VMA, the parent died somehow, and then
> the child died, the page in the pipe would be zeroed out. A child
> should not be able to affect its parent like this, I think. If this
> was an mmap() flag instead of a madvise() command, that issue could be
> avoided.

With a VMA flag underneath, I think you can do an early CoW during fork
to prevent from that.

> Alternatively, if adding more mmap() flags doesn't work,
> perhaps you could scan the VMA and ensure that it contains no pages
> yet, or something like that?
> 
> > > diff --git a/mm/memory.c b/mm/memory.c
> > > index ab650c21bccd..ff78b527660e 100644
> > > --- a/mm/memory.c
> > > +++ b/mm/memory.c
> > > @@ -1091,6 +1091,9 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
> > >                       page_remove_rmap(page, false);
> > >                       if (unlikely(page_mapcount(page) < 0))
> > >                               print_bad_pte(vma, addr, ptent, page);
> > > +                     if (unlikely(vma->vm_flags & VM_WIPEONRELEASE) &&
> > > +                         page_mapcount(page) == 0)
> > > +                             clear_highpage(page);
> > >                       if (unlikely(__tlb_remove_page(tlb, page))) {
> > >                               force_flush = 1;
> > >                               addr += PAGE_SIZE;
> 
> Should something like this perhaps be added in page_remove_rmap()
> instead? That's where the mapcount is decremented; and looking at
> other callers of page_remove_rmap(), in particular the following ones
> look interesting:

Well spotted!

-- 
Michal Hocko
SUSE Labs

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Jann Horn]

On Fri, Apr 26, 2019 at 7:31 AM Michal Hocko <mhocko@kernel.org> wrote:
> On Thu 25-04-19 14:42:52, Jann Horn wrote:
> > On Thu, Apr 25, 2019 at 2:14 PM Michal Hocko <mhocko@kernel.org> wrote:
> > [...]
> > > On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > > > From: Matthew Garrett <mjg59@google.com>
> > > >
> > > > Applications that hold secrets and wish to avoid them leaking can use
> > > > mlock() to prevent the page from being pushed out to swap and
> > > > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > > > can also use atexit() handlers to overwrite secrets on application exit.
> > > > However, if an attacker can reboot the system into another OS, they can
> > > > dump the contents of RAM and extract secrets. We can avoid this by setting
> > > > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > > > firmware wipe the contents of RAM before booting another OS, but this means
> > > > rebooting takes a *long* time - the expected behaviour is for a clean
> > > > shutdown to remove the request after scrubbing secrets from RAM in order to
> > > > avoid this.
> > > >
> > > > Unfortunately, if an application exits uncleanly, its secrets may still be
> > > > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > > > killer decides to kill a process holding secrets, we're not going to be able
> > > > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > > > to request that the kernel clear the covered pages whenever the page
> > > > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > > > will only work on 64-bit systems.
> > [...]
> > > > diff --git a/mm/madvise.c b/mm/madvise.c
> > > > index 21a7881a2db4..989c2fde15cf 100644
> > > > --- a/mm/madvise.c
> > > > +++ b/mm/madvise.c
> > > > @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
> > > >       case MADV_KEEPONFORK:
> > > >               new_flags &= ~VM_WIPEONFORK;
> > > >               break;
> > > > +     case MADV_WIPEONRELEASE:
> > > > +             /* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> > > > +             if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> > > > +                 vma->vm_flags & VM_SHARED) {
> > > > +                     error = -EINVAL;
> > > > +                     goto out;
> > > > +             }
> > > > +             new_flags |= VM_WIPEONRELEASE;
> > > > +             break;
> >
> > An interesting effect of this is that it will be possible to set this
> > on a CoW anon VMA in a fork() child, and then the semantics in the
> > parent will be subtly different - e.g. if the parent vmsplice()d a
> > CoWed page into a pipe, then forked an unprivileged child, the child
>
> Maybe a stupid question. How do you fork an unprivileged child (without
> exec)? Child would have to drop priviledges on its own, no?

Sorry, yes, that's what I meant.

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

On Fri 26-04-19 15:33:25, Jann Horn wrote:
> On Fri, Apr 26, 2019 at 7:31 AM Michal Hocko <mhocko@kernel.org> wrote:
> > On Thu 25-04-19 14:42:52, Jann Horn wrote:
> > > On Thu, Apr 25, 2019 at 2:14 PM Michal Hocko <mhocko@kernel.org> wrote:
> > > [...]
> > > > On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > > > > From: Matthew Garrett <mjg59@google.com>
> > > > >
> > > > > Applications that hold secrets and wish to avoid them leaking can use
> > > > > mlock() to prevent the page from being pushed out to swap and
> > > > > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > > > > can also use atexit() handlers to overwrite secrets on application exit.
> > > > > However, if an attacker can reboot the system into another OS, they can
> > > > > dump the contents of RAM and extract secrets. We can avoid this by setting
> > > > > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > > > > firmware wipe the contents of RAM before booting another OS, but this means
> > > > > rebooting takes a *long* time - the expected behaviour is for a clean
> > > > > shutdown to remove the request after scrubbing secrets from RAM in order to
> > > > > avoid this.
> > > > >
> > > > > Unfortunately, if an application exits uncleanly, its secrets may still be
> > > > > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > > > > killer decides to kill a process holding secrets, we're not going to be able
> > > > > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > > > > to request that the kernel clear the covered pages whenever the page
> > > > > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > > > > will only work on 64-bit systems.
> > > [...]
> > > > > diff --git a/mm/madvise.c b/mm/madvise.c
> > > > > index 21a7881a2db4..989c2fde15cf 100644
> > > > > --- a/mm/madvise.c
> > > > > +++ b/mm/madvise.c
> > > > > @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
> > > > >       case MADV_KEEPONFORK:
> > > > >               new_flags &= ~VM_WIPEONFORK;
> > > > >               break;
> > > > > +     case MADV_WIPEONRELEASE:
> > > > > +             /* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> > > > > +             if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> > > > > +                 vma->vm_flags & VM_SHARED) {
> > > > > +                     error = -EINVAL;
> > > > > +                     goto out;
> > > > > +             }
> > > > > +             new_flags |= VM_WIPEONRELEASE;
> > > > > +             break;
> > >
> > > An interesting effect of this is that it will be possible to set this
> > > on a CoW anon VMA in a fork() child, and then the semantics in the
> > > parent will be subtly different - e.g. if the parent vmsplice()d a
> > > CoWed page into a pipe, then forked an unprivileged child, the child
> >
> > Maybe a stupid question. How do you fork an unprivileged child (without
> > exec)? Child would have to drop priviledges on its own, no?
> 
> Sorry, yes, that's what I meant.

But then the VMA is gone along with the flag so why does it matter?

-- 
Michal Hocko
SUSE Labs

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Jann Horn]

On Fri, Apr 26, 2019 at 3:47 PM Michal Hocko <mhocko@kernel.org> wrote:
> On Fri 26-04-19 15:33:25, Jann Horn wrote:
> > On Fri, Apr 26, 2019 at 7:31 AM Michal Hocko <mhocko@kernel.org> wrote:
> > > On Thu 25-04-19 14:42:52, Jann Horn wrote:
> > > > On Thu, Apr 25, 2019 at 2:14 PM Michal Hocko <mhocko@kernel.org> wrote:
> > > > [...]
> > > > > On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > > > > > From: Matthew Garrett <mjg59@google.com>
> > > > > >
> > > > > > Applications that hold secrets and wish to avoid them leaking can use
> > > > > > mlock() to prevent the page from being pushed out to swap and
> > > > > > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > > > > > can also use atexit() handlers to overwrite secrets on application exit.
> > > > > > However, if an attacker can reboot the system into another OS, they can
> > > > > > dump the contents of RAM and extract secrets. We can avoid this by setting
> > > > > > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > > > > > firmware wipe the contents of RAM before booting another OS, but this means
> > > > > > rebooting takes a *long* time - the expected behaviour is for a clean
> > > > > > shutdown to remove the request after scrubbing secrets from RAM in order to
> > > > > > avoid this.
> > > > > >
> > > > > > Unfortunately, if an application exits uncleanly, its secrets may still be
> > > > > > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > > > > > killer decides to kill a process holding secrets, we're not going to be able
> > > > > > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > > > > > to request that the kernel clear the covered pages whenever the page
> > > > > > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > > > > > will only work on 64-bit systems.
> > > > [...]
> > > > > > diff --git a/mm/madvise.c b/mm/madvise.c
> > > > > > index 21a7881a2db4..989c2fde15cf 100644
> > > > > > --- a/mm/madvise.c
> > > > > > +++ b/mm/madvise.c
> > > > > > @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
> > > > > >       case MADV_KEEPONFORK:
> > > > > >               new_flags &= ~VM_WIPEONFORK;
> > > > > >               break;
> > > > > > +     case MADV_WIPEONRELEASE:
> > > > > > +             /* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> > > > > > +             if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> > > > > > +                 vma->vm_flags & VM_SHARED) {
> > > > > > +                     error = -EINVAL;
> > > > > > +                     goto out;
> > > > > > +             }
> > > > > > +             new_flags |= VM_WIPEONRELEASE;
> > > > > > +             break;
> > > >
> > > > An interesting effect of this is that it will be possible to set this
> > > > on a CoW anon VMA in a fork() child, and then the semantics in the
> > > > parent will be subtly different - e.g. if the parent vmsplice()d a
> > > > CoWed page into a pipe, then forked an unprivileged child, the child
> > >
> > > Maybe a stupid question. How do you fork an unprivileged child (without
> > > exec)? Child would have to drop priviledges on its own, no?
> >
> > Sorry, yes, that's what I meant.
>
> But then the VMA is gone along with the flag so why does it matter?

But in theory, the page might still be used somewhere, e.g. as data in
a pipe (into which the parent wrote it) or whatever. Parent
vmsplice()s a page into a pipe, parent exits, child marks the VMA as
WIPEONRELEASE and exits, page gets wiped, someone else reads the page
from the pipe.

Yes, this is very theoretical, and you'd have to write some pretty
weird software for this to matter. But it doesn't seem clean to me to
allow a child to affect the data in e.g. a pipe that it isn't supposed
to have access to like this.

Then again, this could probably already happen, since do_wp_page()
reuses pages depending on only the mapcount, without looking at the
refcount.

Re: [PATCH V2] mm: Allow userland to request that the kernel clear memory on release

[Michal Hocko]

On Fri 26-04-19 16:03:26, Jann Horn wrote:
> On Fri, Apr 26, 2019 at 3:47 PM Michal Hocko <mhocko@kernel.org> wrote:
> > On Fri 26-04-19 15:33:25, Jann Horn wrote:
> > > On Fri, Apr 26, 2019 at 7:31 AM Michal Hocko <mhocko@kernel.org> wrote:
> > > > On Thu 25-04-19 14:42:52, Jann Horn wrote:
> > > > > On Thu, Apr 25, 2019 at 2:14 PM Michal Hocko <mhocko@kernel.org> wrote:
> > > > > [...]
> > > > > > On Wed 24-04-19 14:10:39, Matthew Garrett wrote:
> > > > > > > From: Matthew Garrett <mjg59@google.com>
> > > > > > >
> > > > > > > Applications that hold secrets and wish to avoid them leaking can use
> > > > > > > mlock() to prevent the page from being pushed out to swap and
> > > > > > > MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> > > > > > > can also use atexit() handlers to overwrite secrets on application exit.
> > > > > > > However, if an attacker can reboot the system into another OS, they can
> > > > > > > dump the contents of RAM and extract secrets. We can avoid this by setting
> > > > > > > CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> > > > > > > firmware wipe the contents of RAM before booting another OS, but this means
> > > > > > > rebooting takes a *long* time - the expected behaviour is for a clean
> > > > > > > shutdown to remove the request after scrubbing secrets from RAM in order to
> > > > > > > avoid this.
> > > > > > >
> > > > > > > Unfortunately, if an application exits uncleanly, its secrets may still be
> > > > > > > present in RAM. This can't be easily fixed in userland (eg, if the OOM
> > > > > > > killer decides to kill a process holding secrets, we're not going to be able
> > > > > > > to avoid that), so this patch adds a new flag to madvise() to allow userland
> > > > > > > to request that the kernel clear the covered pages whenever the page
> > > > > > > reference count hits zero. Since vm_flags is already full on 32-bit, it
> > > > > > > will only work on 64-bit systems.
> > > > > [...]
> > > > > > > diff --git a/mm/madvise.c b/mm/madvise.c
> > > > > > > index 21a7881a2db4..989c2fde15cf 100644
> > > > > > > --- a/mm/madvise.c
> > > > > > > +++ b/mm/madvise.c
> > > > > > > @@ -92,6 +92,22 @@ static long madvise_behavior(struct vm_area_struct *vma,
> > > > > > >       case MADV_KEEPONFORK:
> > > > > > >               new_flags &= ~VM_WIPEONFORK;
> > > > > > >               break;
> > > > > > > +     case MADV_WIPEONRELEASE:
> > > > > > > +             /* MADV_WIPEONRELEASE is only supported on anonymous memory. */
> > > > > > > +             if (VM_WIPEONRELEASE == 0 || vma->vm_file ||
> > > > > > > +                 vma->vm_flags & VM_SHARED) {
> > > > > > > +                     error = -EINVAL;
> > > > > > > +                     goto out;
> > > > > > > +             }
> > > > > > > +             new_flags |= VM_WIPEONRELEASE;
> > > > > > > +             break;
> > > > >
> > > > > An interesting effect of this is that it will be possible to set this
> > > > > on a CoW anon VMA in a fork() child, and then the semantics in the
> > > > > parent will be subtly different - e.g. if the parent vmsplice()d a
> > > > > CoWed page into a pipe, then forked an unprivileged child, the child
> > > >
> > > > Maybe a stupid question. How do you fork an unprivileged child (without
> > > > exec)? Child would have to drop priviledges on its own, no?
> > >
> > > Sorry, yes, that's what I meant.
> >
> > But then the VMA is gone along with the flag so why does it matter?
> 
> But in theory, the page might still be used somewhere, e.g. as data in
> a pipe (into which the parent wrote it) or whatever. Parent
> vmsplice()s a page into a pipe, parent exits, child marks the VMA as
> WIPEONRELEASE and exits, page gets wiped, someone else reads the page
> from the pipe.
> 
> Yes, this is very theoretical, and you'd have to write some pretty
> weird software for this to matter. But it doesn't seem clean to me to
> allow a child to affect the data in e.g. a pipe that it isn't supposed
> to have access to like this.
> 
> Then again, this could probably already happen, since do_wp_page()
> reuses pages depending on only the mapcount, without looking at the
> refcount.

OK, now I see your point. I was confused about the unprivileged child.
You are right that this looks weird but we have traditionally trusted
child processes to not do a harm. I guess this falls down to the same
bucket. An early CoW on these mapping should solve the problem AFAICS.

-- 
Michal Hocko
SUSE Labs

[PATCH V3] mm: Allow userland to request that the kernel clear memory on release

[Matthew Garrett]

From: Matthew Garrett <mjg59@google.com>

Applications that hold secrets and wish to avoid them leaking can use
mlock() to prevent the page from being pushed out to swap and
MADV_DONTDUMP to prevent it from being included in core dumps. Applications
can also use atexit() handlers to overwrite secrets on application exit.
However, if an attacker can reboot the system into another OS, they can
dump the contents of RAM and extract secrets. We can avoid this by setting
CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
firmware wipe the contents of RAM before booting another OS, but this means
rebooting takes a *long* time - the expected behaviour is for a clean
shutdown to remove the request after scrubbing secrets from RAM in order to
avoid this.

Unfortunately, if an application exits uncleanly, its secrets may still be
present in RAM. This can't be easily fixed in userland (eg, if the OOM
killer decides to kill a process holding secrets, we're not going to be able
to avoid that), so this patch adds a new flag to madvise() to allow userland
to request that the kernel clear the covered pages whenever the page
map count hits zero. Since vm_flags is already full on 32-bit, it
will only work on 64-bit systems. This is currently only permitted on
private mappings that have not yet been populated in order to simplify
implementation, which should suffice for the envisaged use cases. We can
extend the behaviour later if we come up with a robust set of semantics.

Signed-off-by: Matthew Garrett <mjg59@google.com>
---

Updated based on feedback from Jann - for now let's just prevent setting
the flag on anything that has already mapped some pages, which avoids
child processes being able to interfere with the parent. In addition,
move the page clearing logic into page_remove_rmap() to ensure that we
cover the full set of cases (eg, handling page migration properly).

 include/linux/mm.h                     |  6 ++++++
 include/linux/rmap.h                   |  2 +-
 include/uapi/asm-generic/mman-common.h |  2 ++
 kernel/events/uprobes.c                |  2 +-
 mm/huge_memory.c                       | 12 ++++++------
 mm/hugetlb.c                           |  4 ++--
 mm/khugepaged.c                        |  2 +-
 mm/ksm.c                               |  2 +-
 mm/madvise.c                           | 25 +++++++++++++++++++++++++
 mm/memory.c                            |  6 +++---
 mm/migrate.c                           |  4 ++--
 mm/rmap.c                              |  9 +++++++--
 12 files changed, 57 insertions(+), 19 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 6b10c21630f5..64bdab679275 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -257,6 +257,8 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_HIGH_ARCH_2	BIT(VM_HIGH_ARCH_BIT_2)
 #define VM_HIGH_ARCH_3	BIT(VM_HIGH_ARCH_BIT_3)
 #define VM_HIGH_ARCH_4	BIT(VM_HIGH_ARCH_BIT_4)
+
+#define VM_WIPEONRELEASE BIT(37)       /* Clear pages when releasing them */
 #endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
 
 #ifdef CONFIG_ARCH_HAS_PKEYS
@@ -298,6 +300,10 @@ extern unsigned int kobjsize(const void *objp);
 # define VM_GROWSUP	VM_NONE
 #endif
 
+#ifndef VM_WIPEONRELEASE
+# define VM_WIPEONRELEASE VM_NONE
+#endif
+
 /* Bits set in the VMA until the stack is in its final location */
 #define VM_STACK_INCOMPLETE_SETUP	(VM_RAND_READ | VM_SEQ_READ)
 
diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index 988d176472df..abb47d623edd 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -177,7 +177,7 @@ void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 		unsigned long, bool);
 void page_add_file_rmap(struct page *, bool);
-void page_remove_rmap(struct page *, bool);
+void page_remove_rmap(struct page *, struct vm_area_struct *, bool);
 
 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
 			    unsigned long);
diff --git a/include/uapi/asm-generic/mman-common.h b/include/uapi/asm-generic/mman-common.h
index abd238d0f7a4..82dfff4a8e3d 100644
--- a/include/uapi/asm-generic/mman-common.h
+++ b/include/uapi/asm-generic/mman-common.h
@@ -64,6 +64,8 @@
 #define MADV_WIPEONFORK 18		/* Zero memory on fork, child only */
 #define MADV_KEEPONFORK 19		/* Undo MADV_WIPEONFORK */
 
+#define MADV_WIPEONRELEASE 20
+#define MADV_DONTWIPEONRELEASE 21
 /* compatibility flags */
 #define MAP_FILE	0
 
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index c5cde87329c7..2230a1717fe3 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -196,7 +196,7 @@ static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
 	set_pte_at_notify(mm, addr, pvmw.pte,
 			mk_pte(new_page, vma->vm_page_prot));
 
-	page_remove_rmap(old_page, false);
+	page_remove_rmap(old_page, vma, false);
 	if (!page_mapped(old_page))
 		try_to_free_swap(old_page);
 	page_vma_mapped_walk_done(&pvmw);
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 165ea46bf149..1ad6ee5857b7 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1260,7 +1260,7 @@ static vm_fault_t do_huge_pmd_wp_page_fallback(struct vm_fault *vmf,
 
 	smp_wmb(); /* make pte visible before pmd */
 	pmd_populate(vma->vm_mm, vmf->pmd, pgtable);
-	page_remove_rmap(page, true);
+	page_remove_rmap(page, vma, true);
 	spin_unlock(vmf->ptl);
 
 	/*
@@ -1410,7 +1410,7 @@ vm_fault_t do_huge_pmd_wp_page(struct vm_fault *vmf, pmd_t orig_pmd)
 			add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
 		} else {
 			VM_BUG_ON_PAGE(!PageHead(page), page);
-			page_remove_rmap(page, true);
+			page_remove_rmap(page, vma, true);
 			put_page(page);
 		}
 		ret |= VM_FAULT_WRITE;
@@ -1783,7 +1783,7 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 
 		if (pmd_present(orig_pmd)) {
 			page = pmd_page(orig_pmd);
-			page_remove_rmap(page, true);
+			page_remove_rmap(page, vma, true);
 			VM_BUG_ON_PAGE(page_mapcount(page) < 0, page);
 			VM_BUG_ON_PAGE(!PageHead(page), page);
 		} else if (thp_migration_supported()) {
@@ -2146,7 +2146,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 			set_page_dirty(page);
 		if (!PageReferenced(page) && pmd_young(_pmd))
 			SetPageReferenced(page);
-		page_remove_rmap(page, true);
+		page_remove_rmap(page, vma, true);
 		put_page(page);
 		add_mm_counter(mm, mm_counter_file(page), -HPAGE_PMD_NR);
 		return;
@@ -2266,7 +2266,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 
 	if (freeze) {
 		for (i = 0; i < HPAGE_PMD_NR; i++) {
-			page_remove_rmap(page + i, false);
+			page_remove_rmap(page + i, vma, false);
 			put_page(page + i);
 		}
 	}
@@ -2954,7 +2954,7 @@ void set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
 	if (pmd_soft_dirty(pmdval))
 		pmdswp = pmd_swp_mksoft_dirty(pmdswp);
 	set_pmd_at(mm, address, pvmw->pmd, pmdswp);
-	page_remove_rmap(page, true);
+	page_remove_rmap(page, vma, true);
 	put_page(page);
 }
 
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 6cdc7b2d9100..1df046525861 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3419,7 +3419,7 @@ void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
 			set_page_dirty(page);
 
 		hugetlb_count_sub(pages_per_huge_page(h), mm);
-		page_remove_rmap(page, true);
+		page_remove_rmap(page, vma, true);
 
 		spin_unlock(ptl);
 		tlb_remove_page_size(tlb, page, huge_page_size(h));
@@ -3643,7 +3643,7 @@ static vm_fault_t hugetlb_cow(struct mm_struct *mm, struct vm_area_struct *vma,
 		mmu_notifier_invalidate_range(mm, range.start, range.end);
 		set_huge_pte_at(mm, haddr, ptep,
 				make_huge_pte(vma, new_page, 1));
-		page_remove_rmap(old_page, true);
+		page_remove_rmap(old_page, vma, true);
 		hugepage_add_new_anon_rmap(new_page, vma, haddr);
 		set_page_huge_active(new_page);
 		/* Make the old page be freed below */
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 449044378782..20df74dfd954 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -673,7 +673,7 @@ static void __collapse_huge_page_copy(pte_t *pte, struct page *page,
 			 * superfluous.
 			 */
 			pte_clear(vma->vm_mm, address, _pte);
-			page_remove_rmap(src_page, false);
+			page_remove_rmap(src_page, vma, false);
 			spin_unlock(ptl);
 			free_page_and_swap_cache(src_page);
 		}
diff --git a/mm/ksm.c b/mm/ksm.c
index fc64874dc6f4..280705f65af7 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -1193,7 +1193,7 @@ static int replace_page(struct vm_area_struct *vma, struct page *page,
 	ptep_clear_flush(vma, addr, ptep);
 	set_pte_at_notify(mm, addr, ptep, newpte);
 
-	page_remove_rmap(page, false);
+	page_remove_rmap(page, vma, false);
 	if (!page_mapped(page))
 		try_to_free_swap(page);
 	put_page(page);
diff --git a/mm/madvise.c b/mm/madvise.c
index 21a7881a2db4..2a6e616e5c0d 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -92,6 +92,26 @@ static long madvise_behavior(struct vm_area_struct *vma,
 	case MADV_KEEPONFORK:
 		new_flags &= ~VM_WIPEONFORK;
 		break;
+	case MADV_WIPEONRELEASE:
+		/*
+		 * MADV_WIPEONRELEASE is only supported on as-yet unallocated
+		 * anonymous memory.
+		 */
+		if (VM_WIPEONRELEASE == 0 || vma->vm_file || vma->anon_vma ||
+		    vma->vm_flags & VM_SHARED) {
+			error = -EINVAL;
+			goto out;
+		}
+
+		new_flags |= VM_WIPEONRELEASE;
+		break;
+	case MADV_DONTWIPEONRELEASE:
+		if (VM_WIPEONRELEASE == 0) {
+			error = -EINVAL;
+			goto out;
+		}
+		new_flags &= ~VM_WIPEONRELEASE;
+		break;
 	case MADV_DONTDUMP:
 		new_flags |= VM_DONTDUMP;
 		break;
@@ -727,6 +747,8 @@ madvise_behavior_valid(int behavior)
 	case MADV_DODUMP:
 	case MADV_WIPEONFORK:
 	case MADV_KEEPONFORK:
+	case MADV_WIPEONRELEASE:
+	case MADV_DONTWIPEONRELEASE:
 #ifdef CONFIG_MEMORY_FAILURE
 	case MADV_SOFT_OFFLINE:
 	case MADV_HWPOISON:
@@ -785,6 +807,9 @@ madvise_behavior_valid(int behavior)
  *  MADV_DONTDUMP - the application wants to prevent pages in the given range
  *		from being included in its core dump.
  *  MADV_DODUMP - cancel MADV_DONTDUMP: no longer exclude from core dump.
+ *  MADV_WIPEONRELEASE - clear the contents of the memory after the last
+ *		reference to it has been released
+ *  MADV_DONTWIPEONRELEASE - cancel MADV_WIPEONRELEASE
  *
  * return values:
  *  zero    - success
diff --git a/mm/memory.c b/mm/memory.c
index ab650c21bccd..dd9555bb9aec 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1088,7 +1088,7 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
 					mark_page_accessed(page);
 			}
 			rss[mm_counter(page)]--;
-			page_remove_rmap(page, false);
+			page_remove_rmap(page, vma, false);
 			if (unlikely(page_mapcount(page) < 0))
 				print_bad_pte(vma, addr, ptent, page);
 			if (unlikely(__tlb_remove_page(tlb, page))) {
@@ -1116,7 +1116,7 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
 
 			pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
 			rss[mm_counter(page)]--;
-			page_remove_rmap(page, false);
+			page_remove_rmap(page, vma, false);
 			put_page(page);
 			continue;
 		}
@@ -2340,7 +2340,7 @@ static vm_fault_t wp_page_copy(struct vm_fault *vmf)
 			 * mapcount is visible. So transitively, TLBs to
 			 * old page will be flushed before it can be reused.
 			 */
-			page_remove_rmap(old_page, false);
+			page_remove_rmap(old_page, vma, false);
 		}
 
 		/* Free the old page.. */
diff --git a/mm/migrate.c b/mm/migrate.c
index 663a5449367a..5d3437a6541d 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -2083,7 +2083,7 @@ int migrate_misplaced_transhuge_page(struct mm_struct *mm,
 
 	page_ref_unfreeze(page, 2);
 	mlock_migrate_page(new_page, page);
-	page_remove_rmap(page, true);
+	page_remove_rmap(page, vma, true);
 	set_page_owner_migrate_reason(new_page, MR_NUMA_MISPLACED);
 
 	spin_unlock(ptl);
@@ -2313,7 +2313,7 @@ static int migrate_vma_collect_pmd(pmd_t *pmdp,
 			 * drop page refcount. Page won't be freed, as we took
 			 * a reference just above.
 			 */
-			page_remove_rmap(page, false);
+			page_remove_rmap(page, vma, false);
 			put_page(page);
 
 			if (pte_present(pte))
diff --git a/mm/rmap.c b/mm/rmap.c
index b30c7c71d1d9..46dc9946a516 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1292,11 +1292,13 @@ static void page_remove_anon_compound_rmap(struct page *page)
 /**
  * page_remove_rmap - take down pte mapping from a page
  * @page:	page to remove mapping from
+ * @vma:	VMA the page belongs to
  * @compound:	uncharge the page as compound or small page
  *
  * The caller needs to hold the pte lock.
  */
-void page_remove_rmap(struct page *page, bool compound)
+void page_remove_rmap(struct page *page, struct vm_area_struct *vma,
+		      bool compound)
 {
 	if (!PageAnon(page))
 		return page_remove_file_rmap(page, compound);
@@ -1321,6 +1323,9 @@ void page_remove_rmap(struct page *page, bool compound)
 	if (PageTransCompound(page))
 		deferred_split_huge_page(compound_head(page));
 
+	if (unlikely(vma->vm_flags & VM_WIPEONRELEASE))
+		clear_highpage(page);
+
 	/*
 	 * It would be tidy to reset the PageAnon mapping here,
 	 * but that might overwrite a racing page_add_anon_rmap
@@ -1652,7 +1657,7 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		 *
 		 * See Documentation/vm/mmu_notifier.rst
 		 */
-		page_remove_rmap(subpage, PageHuge(page));
+		page_remove_rmap(subpage, vma, PageHuge(page));
 		put_page(page);
 	}
 
-- 
2.21.0.593.g511ec345e18-goog

Re: [PATCH V3] mm: Allow userland to request that the kernel clear memory on release

[Vlastimil Babka]

On 4/26/19 12:58 AM, Matthew Garrett wrote:
> From: Matthew Garrett <mjg59@google.com>
> 
> Applications that hold secrets and wish to avoid them leaking can use
> mlock() to prevent the page from being pushed out to swap and
> MADV_DONTDUMP to prevent it from being included in core dumps. Applications
> can also use atexit() handlers to overwrite secrets on application exit.
> However, if an attacker can reboot the system into another OS, they can
> dump the contents of RAM and extract secrets. We can avoid this by setting
> CONFIG_RESET_ATTACK_MITIGATION on UEFI systems in order to request that the
> firmware wipe the contents of RAM before booting another OS, but this means
> rebooting takes a *long* time - the expected behaviour is for a clean
> shutdown to remove the request after scrubbing secrets from RAM in order to
> avoid this.
> 
> Unfortunately, if an application exits uncleanly, its secrets may still be
> present in RAM. This can't be easily fixed in userland (eg, if the OOM
> killer decides to kill a process holding secrets, we're not going to be able
> to avoid that), so this patch adds a new flag to madvise() to allow userland
> to request that the kernel clear the covered pages whenever the page
> map count hits zero. Since vm_flags is already full on 32-bit, it
> will only work on 64-bit systems. This is currently only permitted on
> private mappings that have not yet been populated in order to simplify
> implementation, which should suffice for the envisaged use cases. We can
> extend the behaviour later if we come up with a robust set of semantics.
> 
> Signed-off-by: Matthew Garrett <mjg59@google.com>
> ---
> 
> Updated based on feedback from Jann - for now let's just prevent setting
> the flag on anything that has already mapped some pages, which avoids
> child processes being able to interfere with the parent. In addition,

That makes the API quite tricky and different from existing madvise()
modes that don't care. One would for example have to call
madvise(MADV_WIPEONRELEASE) before mlock(), otherwise mlock() would
fault the pages in (unless MLOCK_ONFAULT). As such it really looks like
a mmap() flag, but that's less flexible.

How bout just doing the CoW on any such pre-existing pages as part of
the madvise(MADV_WIPEONRELEASE) call?

Re: [PATCH V3] mm: Allow userland to request that the kernel clear memory on release

[Matthew Garrett]

On Fri, Apr 26, 2019 at 12:45 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> On 4/26/19 12:58 AM, Matthew Garrett wrote:
> > Updated based on feedback from Jann - for now let's just prevent setting
> > the flag on anything that has already mapped some pages, which avoids
> > child processes being able to interfere with the parent. In addition,
>
> That makes the API quite tricky and different from existing madvise()
> modes that don't care. One would for example have to call
> madvise(MADV_WIPEONRELEASE) before mlock(), otherwise mlock() would
> fault the pages in (unless MLOCK_ONFAULT). As such it really looks like
> a mmap() flag, but that's less flexible.
>
> How bout just doing the CoW on any such pre-existing pages as part of
> the madvise(MADV_WIPEONRELEASE) call?

I'll look into the easiest way to do that.