Dear Willy & list, One more thing I seem to have missed in the previous mail:
On Tue, Nov 18, 2014 at 11:37 AM, Willy Tarreau <[email protected]> wrote: > +/* Opens a socket in the namespace described by <ns> with the parameters > <domain>, > + * <type> and <protocol> and returns the FD or -1 in case of error (check > errno). > + */ > +int socketat(const struct netns_entry *ns, int domain, int type, int > protocol) > > Just thinking about something, in Linux, many syscalls already exist with > the "at" suffix to indicate a variant working based on a file descriptor > pointing to a directory. While I'm not seeing any risk that "socketat" > would one day exist, I think we still have time to invent a less > conflicting > name, what do you think ? Maybe something like ns_socket() or any other > idea ? > I'm OK with renaming it, however, as far as I know whether or not socketat() is required as a syscall was discussed and then dropped at the time when the API for network namespaces was considered. ( http://lwn.net/Articles/407495/) We can still rename it, just to be sure, though. -- KOVACS Krisztian
