Re: kvm/arm64: use-after-free in kvm_vm_ioctl/vmacache_update

From: Marc Zyngier
Date: Tue Mar 14 2017 - 08:26:39 EST

On 14/03/17 11:03, Suzuki K Poulose wrote:
> On 13/03/17 09:58, Marc Zyngier wrote:
>> On 10/03/17 18:37, Suzuki K Poulose wrote:
>>> On 10/03/17 15:50, Andrey Konovalov wrote:
>>>> On Fri, Mar 10, 2017 at 2:38 PM, Andrey Konovalov <andreyknvl@xxxxxxxxxx> 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 vmacache_update+0x114/0x118 mm/vmacache.c:63
>>>>> Read of size 8 at addr ffff80003b9a2040 by task syz-executor/26615
>>>>>
>>>>> CPU: 1 PID: 26615 Comm: syz-executor Not tainted
>>>>> 4.11.0-rc1-next-20170308-xc2-dirty #3
>>>>> Hardware name: Hardkernel ODROID-C2 (DT)
>>>>> Call trace:
>>>>> [<ffff20000808fbb0>] dump_backtrace+0x0/0x440 arch/arm64/kernel/traps.c:505
>>>>> [<ffff200008090010>] show_stack+0x20/0x30 arch/arm64/kernel/traps.c:228
>>>>> [<ffff2000088e9578>] __dump_stack lib/dump_stack.c:16 [inline]
>>>>> [<ffff2000088e9578>] dump_stack+0x110/0x168 lib/dump_stack.c:52
>>>>> [<ffff200008414018>] print_address_description+0x60/0x248 mm/kasan/report.c:250
>>>>> [<ffff2000084142e8>] kasan_report_error+0xe8/0x250 mm/kasan/report.c:349
>>>>> [<ffff200008414564>] kasan_report mm/kasan/report.c:372 [inline]
>>>>> [<ffff200008414564>] __asan_report_load8_noabort+0x3c/0x48 mm/kasan/report.c:393
>>>>> [<ffff200008383f64>] vmacache_update+0x114/0x118 mm/vmacache.c:63
>>>>> [<ffff2000083a9000>] find_vma+0xf8/0x150 mm/mmap.c:2124
>>>>> [<ffff2000080dc19c>] kvm_arch_prepare_memory_region+0x2ac/0x488
>>>>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:1817
>>>>> [<ffff2000080c2920>] __kvm_set_memory_region+0x3d8/0x12b8
>>>>> arch/arm64/kvm/../../../virt/kvm/kvm_main.c:1026
>>>>> [<ffff2000080c3838>] kvm_set_memory_region+0x38/0x58
>>>>> arch/arm64/kvm/../../../virt/kvm/kvm_main.c:1075
>>>>> [<ffff2000080c747c>] kvm_vm_ioctl_set_memory_region
>>>>> arch/arm64/kvm/../../../virt/kvm/kvm_main.c:1087 [inline]
>>>>> [<ffff2000080c747c>] kvm_vm_ioctl+0xb94/0x1308
>>>>> arch/arm64/kvm/../../../virt/kvm/kvm_main.c:2960
>>>>> [<ffff20000848f928>] vfs_ioctl fs/ioctl.c:45 [inline]
>>>>> [<ffff20000848f928>] do_vfs_ioctl+0x128/0xfc0 fs/ioctl.c:685
>>>>> [<ffff200008490868>] SYSC_ioctl fs/ioctl.c:700 [inline]
>>>>> [<ffff200008490868>] SyS_ioctl+0xa8/0xb8 fs/ioctl.c:691
>>>>> [<ffff200008083f70>] el0_svc_naked+0x24/0x28
>>>>>
>>>>> Allocated by task 26657:
>>>>> save_stack_trace_tsk+0x0/0x330 arch/arm64/kernel/stacktrace.c:133
>>>>> save_stack_trace+0x20/0x30 arch/arm64/kernel/stacktrace.c:216
>>>>> save_stack mm/kasan/kasan.c:515 [inline]
>>>>> set_track mm/kasan/kasan.c:527 [inline]
>>>>> kasan_kmalloc+0xd4/0x180 mm/kasan/kasan.c:619
>>>>> kasan_slab_alloc+0x14/0x20 mm/kasan/kasan.c:557
>>>>> slab_post_alloc_hook mm/slab.h:456 [inline]
>>>>> slab_alloc_node mm/slub.c:2718 [inline]
>>>>> slab_alloc mm/slub.c:2726 [inline]
>>>>> kmem_cache_alloc+0x144/0x230 mm/slub.c:2731
>>>>> __split_vma+0x118/0x608 mm/mmap.c:2515
>>>>> do_munmap+0x194/0x9b0 mm/mmap.c:2636
>>>>> Freed by task 26657:
>>>>> save_stack_trace_tsk+0x0/0x330 arch/arm64/kernel/stacktrace.c:133
>>>>> save_stack_trace+0x20/0x30 arch/arm64/kernel/stacktrace.c:216
>>>>> save_stack mm/kasan/kasan.c:515 [inline]
>>>>> set_track mm/kasan/kasan.c:527 [inline]
>>>>> kasan_slab_free+0x84/0x198 mm/kasan/kasan.c:592
>>>>> slab_free_hook mm/slub.c:1357 [inline]
>>>>> slab_free_freelist_hook mm/slub.c:1379 [inline]
>>>>> slab_free mm/slub.c:2961 [inline]
>>>>> kmem_cache_free+0x80/0x258 mm/slub.c:2983
>>>>> __vma_adjust+0x6b0/0xf mm/mmap.c:890] el0_svc_naked+0x24/0x28
>>>>>
>>>>> The buggy address belongs to the object at ffff80003b9a2000
>>>>> which belongs to the cache vm_area_struct(647:session-6.scope) of size 184
>>>>> The buggy address is located 64 bytes inside of
>>>>> 184-byte region [ffff80003b9a2000, ffff80003b9a20b8)
>>>>> The buggy address belongs to the page:
>>>>> page:ffff7e0000ee6880 count:1 mapcount:0 mapping: (null) index:0x0
>>>>> flags: 0xfffc00000000100(slab)
>>>>> raw: 0fffc00000000100 0000000000000000 0000000000000000 0000000180100010
>>>>> raw: 0000000000000000 0000000c00000001 ffff80005a5cc600 ffff80005ac99980
>>>>> page dumped because: kasan: bad access detected
>>>>> page->mem_cgroup:ffff80005ac99980
>>>>>
>>>>> Memory state around the buggy address:
>>>>> ffff80003b9a1f00: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>>>>> ffff80003b9a1f80: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
>>>>>> ffff80003b9a2000: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>>>>> ^
>>>>> ffff80003b9a2080: fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc fb
>>>>> ffff80003b9a2100: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>>>>> ==================================================================
>>>>
>>>> Another one that looks related and doesn't have parts of stack traces missing:
>>>>
>>>> ==================================================================
>>>> BUG: KASAN: use-after-free in find_vma+0x140/0x150 mm/mmap.c:2114
>>>> Read of size 8 at addr ffff800031a03e90 by task syz-executor/4360
>>>>
>>>> CPU: 2 PID: 4360 Comm: syz-executor Not tainted
>>>> 4.11.0-rc1-next-20170308-xc2-dirty #3
>>>> Hardware name: Hardkernel ODROID-C2 (DT)
>>>> Call trace:
>>>> [<ffff20000808fbb0>] dump_backtrace+0x0/0x440 arch/arm64/kernel/traps.c:505
>>>> [<ffff200008090010>] show_stack+0x20/0x30 arch/arm64/kernel/traps.c:228
>>>> [<ffff2000088e9578>] __dump_stack lib/dump_stack.c:16 [inline]
>>>> [<ffff2000088e9578>] dump_stack+0x110/0x168 lib/dump_stack.c:52
>>>> [<ffff200008414018>] print_address_description+0x60/0x248 mm/kasan/report.c:250
>>>> [<ffff2000084142e8>] kasan_report_error+0xe8/0x250 mm/kasan/report.c:349
>>>> [<ffff200008414564>] kasan_report mm/kasan/report.c:372 [inline]
>>>> [<ffff200008414564>] __asan_report_load8_noabort+0x3c/0x48 mm/kasan/report.c:393
>>>> [<ffff2000083a9048>] find_vma+0x140/0x150 mm/mmap.c:2114
>>>> [<ffff2000080dc19c>] kvm_arch_prepare_memory_region+0x2ac/0x488
>>>> arch/arm64/kvm/../../../arch/arm/kvm/mmu.c:1817
>>>
>>> It looks like we don't take the mmap_sem before calling find_vma() in
>>> stage2_unmap_memslot() and in kvm_arch_prepare_memory_region(), which is causing
>>> the race, with probably the test trying to unmap ranges in between.
>>
>> That indeed seems like a possible failure mode. The annoying thing is
>> that we're not exactly in a position to take mmap_sem in
>> stage2_unmap_memslot, since we hold the kvm->mmu_lock spinlock. We may
>> have to hold mmap_sem while iterating over all the memslots.
>>
>> How about the following (very lightly tested):
>>
>> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
>> index 962616fd4ddd..2006a79d5912 100644
>> --- a/arch/arm/kvm/mmu.c
>> +++ b/arch/arm/kvm/mmu.c
>> @@ -803,6 +803,7 @@ void stage2_unmap_vm(struct kvm *kvm)
>> int idx;
>>
>> idx = srcu_read_lock(&kvm->srcu);
>> + down_read(&current->mm->mmap_sem);
>> spin_lock(&kvm->mmu_lock);
>>
>> slots = kvm_memslots(kvm);
>> @@ -810,6 +811,7 @@ void stage2_unmap_vm(struct kvm *kvm)
>> stage2_unmap_memslot(kvm, memslot);
>>
>> spin_unlock(&kvm->mmu_lock);
>> + up_read(&current->mm->mmap_sem);
>> srcu_read_unlock(&kvm->srcu, idx);
>> }
>>
>> @@ -1813,6 +1815,7 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
>> * | memory region |
>> * +--------------------------------------------+
>> */
>> + down_read(&current->mm->mmap_sem);
>> do {
>> struct vm_area_struct *vma = find_vma(current->mm, hva);
>> hva_t vm_start, vm_end;
>
> I have added the following hunk :
>
> @@ -1844,8 +1847,10 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
> pa += vm_start - vma->vm_start;
>
> /* IO region dirty page logging not allowed */
> - if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES)
> - return -EINVAL;
> + if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES) {
> + ret = -EINVAL;
> + goto out;
> + }
>
> ret = kvm_phys_addr_ioremap(kvm, gpa, pa,
> vm_end - vm_start,
>

Ah, of course... Thanks for pointing that out, I'll fix it as I post a
proper patch.

Thanks,

M.
--
Jazz is not dead. It just smells funny...