On 8/2/16 7:27 AM, Christoph Hellwig wrote:
> On Tue, Aug 02, 2016 at 02:49:00PM +0800, Zorro Lang wrote:
>> After XFS_GETNEXTQUOTA feature has been merged into linux kernel and
>> xfsprogs, xfs_quota use Q_XGETNEXTQUOTA for report and dump, and
>> fall back to old XFS_GETQUOTA ioctl if XFS_GETNEXTQUOTA fails.
>>
>> But when XFS_GETNEXTQUOTA fails, xfs_quota print a warning as
>> "XFS_GETQUOTA: Invalid argument". That's due to kernel can't
>> recognize XFS_GETNEXTQUOTA ioctl and return EINVAL. At this time,
>> the warning is helpless, xfs_quota just need to fall back.
>
> We'd still want to report other errors, right?
I advised Zorro to do it this way, because -EINVAL can have a few
meanings, and we don't know for sure why we got it. Could be a
bad cmd, or a bad type, or ...
If it fails, we're going to fall back anyway, so for any other
error we'd print it twice; on the fallback, we'd print the real
unexpected error anyway, so I think the user will get the relevant
information in this case.
But I guess we could do:
+ /* EINVAL is expected for XFS_GETNEXTQUOTA on older kernels */
if (xfsquotactl(cmd, dev, type, id, (void *)&d) < 0) {
+ if (errno != ENOENT && errno != ENOSYS && errno != ESRCH &&
+ (cmd == XFS_GETNEXTQUOTA && errno != EINVAL)
and then change what we print (not perror("XFS_GETQUOTA") in all cases)?
But if we got i.e. EPERM, we'd print the EPERM error twice; once for the
first call, and once for the fallback. I suppose that'd be ok, but not
sure it's helpful.
-Eric
|