|
On 02.04.2016 23:16, Thomas Haller
wrote:
On Sat, 2016-04-02 at 21:49 +0200, Stjepan Groš wrote:I have one problem now. Namely, function _is_root in NMNetns doesn't work, or am I doing something wrong? When constructing NMNetnsController object the first NMNetns is created and that one should be for root network namespace. I'm using _is_root() to skip device initialization (for root that is done by NMManager now) but _is_root() returns FALSE?Hi Stjepan, The NMPNetns instance that represents the root namespace is nmp_netns_get_initial(). You don't create that one, it's always there. You change to call priv->nmp_netns = nmp_netns_new(); is wrong. This creates a new namespace and switches to it. When you switch namespace at a certain moment, you *must* switch back before you return from the function; that is, you must always call nmp_netns_pop() before your function returns. The reason for that is, that at any point it must be known which namespace is currently active. If you call a function and that function switchs namespace, that function is required to restore the callers namespace before returning. For that reason, I think it's better that nm_netns_new() / nm_netns_init() does *not* call nmp_netns_new() to create a new namespace. Instead, it seems more logical that it takes over the *current* namespace nmp_netns_get_current(). So it's the job of the caller of nm_netns_new() -- create_new_namespace() -- to prepare the namespace. Then, when creating the NMNetns instance for the root namespace, the caller just does not call nmp_netns_new() but stays on the root namespace. At that point, nmp_netns_get_currenty() equals nmp_netns_get_initial() and it just works. Does that make sense? Yes, it makes sense. I tried it and it passed the problem I had. SG ThomasThe code with my changes to your branch is on the following URL: https://github.com/sgros/NETNS_NetworkManager SG |
Attachment:
signature.asc
Description: OpenPGP digital signature