On 6/22/21 9:29 AM, Philippe Mathieu-Daudé wrote:
> On 6/21/21 5:36 PM, Fam Zheng wrote:
>>> On 21 Jun 2021, at 16:13, Philippe Mathieu-Daudé <phi...@redhat.com> wrote:
>>> On 6/21/21 3:18 PM, Fam Zheng wrote:
>>>>> On 21 Jun 2021, at 10:32, Philippe Mathieu-Daudé <phi...@redhat.com>
>>>>> wrote:
>>>>>
>>>>> When the NVMe block driver was introduced (see commit bdd6a90a9e5,
>>>>> January 2018), Linux VFIO_IOMMU_MAP_DMA ioctl was only returning
>>>>> -ENOMEM in case of error. The driver was correctly handling the
>>>>> error path to recycle its volatile IOVA mappings.
>>>>>
>>>>> To fix CVE-2019-3882, Linux commit 492855939bdb ("vfio/type1: Limit
>>>>> DMA mappings per container", April 2019) added the -ENOSPC error to
>>>>> signal the user exhausted the DMA mappings available for a container.
>>>>>
>>>>> The block driver started to mis-behave:
>>>>>
>>>>> qemu-system-x86_64: VFIO_MAP_DMA failed: No space left on device
>>>>> (qemu)
>>>>> (qemu) info status
>>>>> VM status: paused (io-error)
>>>>> (qemu) c
>>>>> VFIO_MAP_DMA failed: No space left on device
>>>>> qemu-system-x86_64: block/block-backend.c:1968: blk_get_aio_context:
>>>>> Assertion `ctx == blk->ctx' failed.
>>>>
>>>> Hi Phil,
>>>>
>>>>
>>>> The diff looks good to me, but I’m not sure what exactly caused the
>>>> assertion failure. There is `if (r) { goto fail; }` that handles -ENOSPC
>>>> before, so it should be treated as a general case. What am I missing?
>>>
>>> Good catch, ENOSPC ends setting BLOCK_DEVICE_IO_STATUS_NOSPACE
>>> -> BLOCK_ERROR_ACTION_STOP, so the VM is paused with DMA mapping
>>> exhausted. I don't understand the full "VM resume" path, but this
>>> is not what we want (IO_NOSPACE is to warn the operator to add
>>> more storage and resume, which is pointless in our case, resuming
>>> won't help until we flush the mappings).
>>>
>>> IIUC what we want is return ENOMEM to set BLOCK_DEVICE_IO_STATUS_FAILED.
>>
>> I agree with that. It just makes me feel there’s another bug in the resuming
>> code path. Can you get a backtrace?
>
> It seems the resuming code path bug has been fixed elsewhere:
>
> (qemu) info status
> info status
> VM status: paused (io-error)
> (qemu) c
> c
> 2021-06-22T07:27:00.745466Z qemu-system-x86_64: VFIO_MAP_DMA failed: No
> space left on device
> (qemu) info status
> info status
> VM status: paused (io-error)
> (qemu) c
> c
> 2021-06-22T07:27:12.458137Z qemu-system-x86_64: VFIO_MAP_DMA failed: No
> space left on device
> (qemu) c
> c
> 2021-06-22T07:27:13.439167Z qemu-system-x86_64: VFIO_MAP_DMA failed: No
> space left on device
> (qemu) c
> c
> 2021-06-22T07:27:14.272071Z qemu-system-x86_64: VFIO_MAP_DMA failed: No
> space left on device
> (qemu)
>
I tested all releases up to v4.1.0 and could not trigger the
blk_get_aio_context() assertion. Building using --enable-debug.
IIRC Gentoo is more aggressive, so I'll restart using -O2.
