Re: [Qemu-devel] [PATCH] ide: ahci: reset ncq object to unused on error

[John Snow <jsnow@redhat.com>]

On 01/08/2016 07:08 AM, P J P wrote:
> From: Prasad J Pandit <pjp@fedoraproject.org>
> 
> When processing NCQ commands, ACHI device emulation prepares a
> NCQ transfer object; To which an aio control block(aiocb) object
> is assigned in 'execute_ncq_command'. In case, when the NCQ
> command is invalid, the 'aiocb' object is not assigned, and NCQ
> transfer object is left as 'used'. This leads to a use after
> free error in 'bdrv_aio_cancel_async' via 'ahci_reset_port'.
> Reset NCQ transfer object to 'unused' to avoid it.
> 

The commit message is a little misleading:

In both of these error pathways, AIOCB is actually never assigned to
begin with. However, neglecting to set used = 0 will indeed cause
ahci_reset_port to try to read these values.

So it's not necessarily a use-after-free.

Regardless, good find!

> Reported-by: Qinghao Tang <luodalongde@gmail.com>
> Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
> ---
>  hw/ide/ahci.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/hw/ide/ahci.c b/hw/ide/ahci.c
> index dd1912e..e359127 100644
> --- a/hw/ide/ahci.c
> +++ b/hw/ide/ahci.c
> @@ -1014,6 +1014,7 @@ static void execute_ncq_command(NCQTransferState *ncq_tfs)
>          DPRINTF(port, "error: unsupported NCQ command (0x%02x) received\n",
>                  ncq_tfs->cmd);
>          qemu_sglist_destroy(&ncq_tfs->sglist);
> +        ncq_tfs->used = 0;
>          ncq_err(ncq_tfs);
>      }
>  }
> @@ -1081,6 +1082,7 @@ static void process_ncq_command(AHCIState *s, int port, uint8_t *cmd_fis,
>                       "is smaller than the requested size (0x%zx)",
>                       ncq_tfs->sglist.size, size);
>          qemu_sglist_destroy(&ncq_tfs->sglist);
> +        ncq_tfs->used = 0;
>          ncq_err(ncq_tfs);
>          ahci_trigger_irq(ad->hba, ad, PORT_IRQ_OVERFLOW);
>          return;
> 

I think it should be safe to put ncq_tfs->used = 0 directly inside of
ncq_err, and that way we won't have any other error pathways omitting
this in the future.

--js