Re: [Qemu-devel] [PATCH] scsi: Don't check uninitialized local variable

[Eric Blake]

On 01/08/2018 07:20 AM, Peter Lieven wrote:
> Am 08.01.2018 um 04:09 schrieb Fam Zheng:
>> After the out label there is a check on iTask.task but it is not
>> initialized yet.
>>
>> Fixes: e38bc23454ef763deb4405ebdee6a1081aa00bc8
>> Signed-off-by: Fam Zheng <address@hidden>
>> ---
>>   block/iscsi.c | 3 +--
>>   1 file changed, 1 insertion(+), 2 deletions(-)
>>

> If lbpme is 0 we run into the same error. And this is even more likely
> than an unaligned request.

In fact, my byte-based series adds an assertion that unaligned requests
aren't possible.

> 
> I think the right patch is to move the init of iTask up again where it was:

I had to rebase v7 of my byte-based series on top of the late iTask
initialization; moving it back to early initialization makes sense.

> 
> 
> diff --git a/block/iscsi.c b/block/iscsi.c
> index 5c0a9e5..6a1c537 100644
> --- a/block/iscsi.c
> +++ b/block/iscsi.c
> @@ -658,6 +658,8 @@ static int64_t coroutine_fn
> iscsi_co_get_block_status(BlockDriverState *bs,
>      uint64_t lba;
>      int64_t ret;
> 
> +    iscsi_co_init_iscsitask(iscsilun, &iTask);
> +
>      if (!is_sector_request_lun_aligned(sector_num, nb_sectors,
> iscsilun)) {
>          ret = -EINVAL;
>          goto out;
> @@ -675,7 +677,6 @@ static int64_t coroutine_fn
> iscsi_co_get_block_status(BlockDriverState *bs,
> 
>      lba = sector_qemu2lun(sector_num, iscsilun);
> 
> -    iscsi_co_init_iscsitask(iscsilun, &iTask);
>      qemu_mutex_lock(&iscsilun->mutex);
>  retry:
>      if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,

If you resubmit this as a formal patch, you can add:
Reviewed-by: Eric Blake <address@hidden>

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

signature.asc
Description: OpenPGP digital signature