On Mon, Feb 06, 2017 at 03:58:52PM +0300, Konstantin Khorenko wrote: > So, what is the result of discussion here? > > -- > Best regards, > > Konstantin Khorenko, > Virtuozzo Linux Kernel Team > > On 12/23/2016 10:19 AM, Cyrill Gorcunov wrote: > > On Fri, Dec 23, 2016 at 09:59:28AM +0300, Cyrill Gorcunov wrote: > > > On Thu, Dec 22, 2016 at 04:45:10PM -0800, Andrey Vagin wrote: > > > > > > > > Actually, this patch breaks the old behaviour even when MSG_PEEK isn't > > > > set. > > > > > > > > I was thinking a bit more and I don't understand why it is a problem. If > > > > we can't fill a buffer, an error will be returned and a user will be > > > > able > > > > to set peek_offset to get these data again. > > > > > > A user doesn't have to set peek again, without the patch the internal > > > state > > > of sk is context-dependant, which is broken design. Take a look on unix > > > sockets code, they DON'T modify offset if something earlier failed for > > > exactly same reason. > > > > Another option -- set offset only iff data copying passed.
I think the patch is correct. Up to Andrew, since it's his code in first place. _______________________________________________ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel