On Mon, Nov 13, 2017 at 7:23 PM, 张军伟(基础平台部) <zhangjunweimar...@didichuxing.com> wrote: > >> On 14 Nov 2017, at 10:09, Cong Wang <xiyou.wangc...@gmail.com> wrote: >> >> On Mon, Nov 13, 2017 at 4:36 AM, Tonghao Zhang <xiangxia.m....@gmail.com> >> wrote: >>> From: Tonghao Zhang <xiangxia.m....@gmail.com> >>> >>> This patch add a member in struct netns_core. and this is >>> a counter for socket_inuse in the _net_ namespace. The patch >>> will add/sub counter in the sock_alloc or sock_release. In >>> the sock_alloc, the private data of inode saves the special >>> _net_. When releasing it, we can access the special _net_ and >>> dec the counter of socket in that namespace. >>> >>> By the way, we dont use the 'current->nsproxy->net_ns' in the >>> sock_release. In one case，when one task exits, the 'do_exit' >>> may set the current->nsproxy NULL, and then call the sock_release. >>> Use the private data of inode, saving few bytes. >> >> Why do you need to hold netns refcnt for socket? sock already holds >> it, so you can just access it by sock_net(sock->sk) ? > I think you suggestion is > replace get_net(net) ===> with sock_net(sock->sk)
Not literally, but essentially yes. > > If thus, I think it could not work. > Because sock->sk is assigned after sock_alloc. > What we change is done in sock_alloc. You don't have to do it in sock_alloc(). > By the way, sock->sk may has been released(NULL) in sock_release. My point is since sock is always paired with a sk and sk already holds refcnt, you don't need to hold it again, it looks unnecessary.
