kvm/arm64: use-after-free in kvm_unmap_hva_handler/unmap_stage2_pmds

[Paolo Bonzini]

On 14/03/2017 12:07, Suzuki K Poulose wrote:
> On 10/03/17 13:34, Andrey Konovalov wrote:
>> Hi,
>>
>> I've got the following error report while fuzzing the kernel with
>> syzkaller.
>>
>> On linux-next commit 56b8bad5e066c23e8fa273ef5fba50bd3da2ace8 (Mar 8).
>>
>> Unfortunately I can't reproduce it.
>>
>> ==================================================================
>> BUG: KASAN: use-after-free in put_page include/linux/compiler.h:243
>> [inline]
>> BUG: KASAN: use-after-free in unmap_stage2_pmds
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:240 [inline]
>> BUG: KASAN: use-after-free in unmap_stage2_puds
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:269 [inline]
>> BUG: KASAN: use-after-free in unmap_stage2_range+0x884/0x938
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:299
>> Read of size 8 at addr ffff80004585c000 by task syz-executor/5176
> 
> 
>> [<ffff2000080d9664>] put_page include/linux/compiler.h:243 [inline]
>> [<ffff2000080d9664>] unmap_stage2_pmds
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:240 [inline]
>> [<ffff2000080d9664>] unmap_stage2_puds
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:269 [inline]
>> [<ffff2000080d9664>] unmap_stage2_range+0x884/0x938
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:299
>> [<ffff2000080d9740>] kvm_unmap_hva_handler+0x28/0x38
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:1556
>> [<ffff2000080d7578>] handle_hva_to_gpa+0x140/0x250
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:1547
>> [<ffff2000080db858>] kvm_unmap_hva_range+0x60/0x80
>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:1579
>> [<ffff2000080c12dc>]
>> kvm_mmu_notifier_invalidate_range_start+0x194/0x278
>> arch/arm64/kvm/../../../virt/kvm/kvm_main.c:357
>> [<ffff2000083fea58>] __mmu_notifier_invalidate_range_start+0x1d0/0x2a0
>> mm/mmu_notifier.c:199
>> [<ffff200008395f44>] mmu_notifier_invalidate_range_start
>> include/linux/mmu_notifier.h:282 [inline]
>> [<ffff200008395f44>] unmap_vmas+0x12c/0x198 mm/memory.c:1372
>> [<ffff2000083a8e00>] unmap_region+0x128/0x230 mm/mmap.c:2460
>> [<ffff2000083ae6e0>] update_hiwater_vm include/linux/mm.h:1483 [inline]
>> [<ffff2000083ae6e0>] remove_vma_list mm/mmap.c:2432 [inline]
>> [<ffff2000083ae6e0>] do_munmap+0x598/0x9b0 mm/mmap.c:2662
>> [<ffff2000083b14a0>] find_vma_links mm/mmap.c:495 [inline]
>> [<ffff2000083b14a0>] mmap_region+0x138/0xc78 mm/mmap.c:1610
>> [<ffff2000083b23ac>] is_file_hugepages include/linux/hugetlb.h:269
>> [inline]
>> [<ffff2000083b23ac>] do_mmap+0x3cc/0x848 mm/mmap.c:1446
>> [<ffff200008369b0c>] do_mmap_pgoff include/linux/mm.h:2039 [inline]
>> [<ffff200008369b0c>] vm_mmap_pgoff+0xec/0x120 mm/util.c:305
>> [<ffff2000083ac558>] SYSC_mmap_pgoff mm/mmap.c:1475 [inline]
>> [<ffff2000083ac558>] SyS_mmap_pgoff+0x220/0x420 mm/mmap.c:1458
>> [<ffff20000808ecf0>] sys_mmap+0x58/0x80 arch/arm64/kernel/sys.c:37
>> [<ffff200008083f70>] el0_svc_naked+0x24/0x28
>>
> 
> We hold kvm->mmu_lock, while manipulating the stage2 ranges. However, I
> find that
> we don't take the lock, when we do it f rom kvm_free_stage2_pgd(), which
> could
> potentially have caused a problem with an munmap of a memslot ?
> 
> Lightly tested...
> 
> 
> commit fa75684dbf0fe845cf8403517d6e0c2c3344a544
> Author: Suzuki K Poulose <suzuki.poulose at arm.com>
> Date:   Tue Mar 14 10:26:54 2017 +0000
> 
>     kvm: arm: Fix locking for kvm_free_stage2_pgd
>         In kvm_free_stage2_pgd() we don't hold the kvm->mmu_lock while
> calling
>     unmap_stage2_range() on the entire memory range for the guest. This
> could
>     cause problems with other callers (e.g, munmap on a memslot) trying to
>     unmap a range.
>         Cc: Marc Zyngier <marc.zyngier at arm.com>
>     Cc: Christoffer Dall <christoffer.dall at linaro.org>
>     Signed-off-by: Suzuki K Poulose <suzuki.poulose at arm.com>
> 
> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> index a07ce3e..7f97063 100644
> --- a/arch/arm/kvm/mmu.c
> +++ b/arch/arm/kvm/mmu.c
> @@ -831,7 +831,10 @@ void kvm_free_stage2_pgd(struct kvm *kvm)
>         if (kvm->arch.pgd == NULL)
>                 return;
>  
> +       spin_lock(&kvm->mmu_lock);
>         unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
> +       spin_unlock(&kvm->mmu_lock);
> +
>         /* Free the HW pgd, one page at a time */
>         free_pages_exact(kvm->arch.pgd, S2_PGD_SIZE);
>         kvm->arch.pgd = NULL;

Reviewed-by: Paolo Bonzini <pbonzini at redhat.com>
> 
> 
> 
>> The buggy address belongs to the page:
>> page:ffff7e0001161700 count:0 mapcount:0 mapping:          (null)
>> index:0x0
>> flags: 0xfffc00000000000()
>> raw: 0fffc00000000000 0000000000000000 0000000000000000 00000000ffffffff
>> raw: ffff7e00018c9120 ffff7e0000ea8b60 0000000000000000 0000000000000000
>> page dumped because: kasan: bad access detected
>>
>> Memory state around the buggy address:
>>  ffff80004585bf00: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>>  ffff80004585bf80: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>>> ffff80004585c000: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>>                    ^
>>  ffff80004585c080: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>>  ffff80004585c100: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>> ==================================================================
>>
>