On 22.03.2017 17:28, Stefan Hajnoczi wrote:
> On Mon, Mar 20, 2017 at 04:07:16PM +0100, Max Reitz wrote:
>> On 20.03.2017 11:18, Stefan Hajnoczi wrote:
>>> On Mon, Mar 13, 2017 at 10:39:46PM +0100, Max Reitz wrote:
>>>> diff --git a/block/iscsi.c b/block/iscsi.c
>>>> index ab559a6f71..5d6265c4a6 100644
>>>> --- a/block/iscsi.c
>>>> +++ b/block/iscsi.c
>>>> @@ -2060,11 +2060,16 @@ static void iscsi_reopen_commit(BDRVReopenState
>>>> *reopen_state)
>>>> }
>>>> }
>>>>
>>>> -static int iscsi_truncate(BlockDriverState *bs, int64_t offset, Error
>>>> **errp)
>>>> +static int iscsi_truncate(BlockDriverState *bs, int64_t offset,
>>>> + PreallocMode prealloc, Error **errp)
>>>> {
>>>> IscsiLun *iscsilun = bs->opaque;
>>>> Error *local_err = NULL;
>>>>
>>>> + if (prealloc != PREALLOC_MODE_OFF) {
>>>> + return -ENOTSUP;
>>>> + }
>>>> +
>>>> if (iscsilun->type != TYPE_DISK) {
>>>> return -ENOTSUP;
>>>> }
>>>
>>> Nevermind what I said about adding a BiteSizedTasks entry:
>>>
>>> The missing errp usage is not in qemu.git/master yet. Please fix up
>>> your bdrv_truncate() errp patch to use errp in all cases, e.g.
>>> error_setg("Unable to truncate non-disk LUN").
>>
>> The thing is that I wasn't comfortable doing that for all block drivers.
>> I mean, I can take another look but I'd rather have vague error messages
>> ("truncation failed: #{strerror}") than outright wrong ones because I
>> didn't know what error message to use.
>>
>> Of course you could argue that this may probably come out during review
>> but that implies that every submaintainer for every block driver would
>> actually come out for review...
>
> I'm worried about errp being set in only a subset of error cases.
>
> This is likely to cause bugs if callers use if (local_err). Grepping
> through the codebase I can see instances of:
Yes, but the generic bdrv_truncate() will always set errp if the driver
hasn't done so.
Max
> ret = foo(..., &local_err);
> if (local_err) { /* no ret check! */
> ...
> }
>
> The code would work fine with qcow2 but not iscsi, for example.
>
> IMO we should always set errp, even if the error message is vague
> ("truncation failed: #{strerror}").
>
> Stefan
>
signature.asc
Description: OpenPGP digital signature
