On 29.10.2014 19:28, Paolo Bonzini wrote:
> On 10/29/2014 05:18 PM, Peter Lieven wrote:
>> Am 29.10.2014 um 14:31 schrieb Paolo Bonzini:
>>> On 10/29/2014 02:13 PM, Fam Zheng wrote:
>>>> Before, when a write protected iSCSI target is attached as scsi-disk
>>>> with BDRV_O_RDWR, we report it as writable, while in fact all writes
>>>> will fail.
>>>>
>>>> One way to improve this is to report write protect flag as true to
>>>> guest, but a even better way is to refuse using a write protected LUN to
>>>> guest.
>>>>
>>>> Target write protect flag is checked with a mode sense query.
>>>>
>>>> Signed-off-by: Fam Zheng <famz@redhat.com>
>>>> ---
>>>>   block/iscsi.c | 30 ++++++++++++++++++++++++++++++
>>>>   1 file changed, 30 insertions(+)
>>>>
>>>> diff --git a/block/iscsi.c b/block/iscsi.c
>>>> index 233f462..c154928 100644
>>>> --- a/block/iscsi.c
>>>> +++ b/block/iscsi.c
>>>> @@ -1339,6 +1339,36 @@ static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
>>>>       scsi_free_scsi_task(task);
>>>>       task = NULL;
>>>>   
>>>> +    /* Check the write protect flag of the LUN if we want to write */
>>>> +    if (flags & BDRV_O_RDWR) {
>>>> +        struct scsi_mode_sense *ms;
>>>> +
>>>> +        task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
>>>> +                                      1, SCSI_MODESENSE_PC_CURRENT,
>>>> +                                      0x3F,
>>>> +                                      0, 255);
>>>> +
>>>> +        if (task == NULL) {
>>>> +            error_setg(errp, "Failed to send MODE_SENSE10 command: %s\n",
>>> This is MODE SENSE(6).  Fixed and applied.
>>>
>>> Paolo
>>>
>>>> +                       iscsi_get_error(iscsilun->iscsi));
>>>> +            ret = -EINVAL;
>>>> +            goto out;
>>>> +        }
>>>> +
>>>> +        if (task->status != SCSI_STATUS_GOOD) {
>>>> +            error_setg(errp, "MODE_SENSE10 failed: %s\n",
>>>> +                       iscsi_get_error(iscsi));
>>>> +            ret = -EINVAL;
>>>> +            goto out;
>>>> +        }
>>>> +        ms = scsi_datain_unmarshall(task);
>> scsi_datain_unmarshall may fail. You need to check for NULL here.
> Thanks for the remark, I fixed this too.
>
> Paolo

I am not 100% happy with this patch as it may break support for some targets.

It seems that MODESENSE is somewhat tricky with some targets. See function  sd_read_write_protect_flag
in <http://lxr.free-electrons.com/ident?i=sd_read_write_protect_flag>drivers/scsi/sd.c of the Linux kernel.

I would not fail if this modesense fails, but just assume write enabled. Maybe drop a warning.

In the command itself:
  0x3F => SCSI_MODEPAGE_RETURN_ALL_PAGES

Can you please post the resulting patch somewhere for review.

Thanks,
Peter