| Message ID | 1447179434-18596-1-git-send-email-bordjukov@gmail.com |
|---|---|
| State | Changes Requested |
| Headers | show |
On Tue, Nov 10, 2015 at 08:17:13PM +0200, Petko Bordjukov wrote: > Make it possible for several instances of hostapd to listen on the same > network interface. > diff --git a/src/ap/iapp.c b/src/ap/iapp.c > @@ -447,6 +448,15 @@ struct iapp_data * iapp_init(struct hostapd_data *hapd, const char *iface) > os_memset(&uaddr, 0, sizeof(uaddr)); > uaddr.sin_family = AF_INET; > uaddr.sin_port = htons(IAPP_UDP_PORT); > + > + if (setsockopt(iapp->udp_sock, SOL_SOCKET, SO_REUSEPORT, &reuseport, > + sizeof(reuseport)) < 0) { SO_REUSEPORT is a pretty recent addition. Wouldn't this break build with older header files? As such, this should probably be protected with #ifdef SO_REUSEPORT or something similar. > + wpa_printf(MSG_INFO, "iapp_init - setsockopt[UDP,SO_REUSEPORT]: %s", > + strerror(errno)); > + iapp_deinit(iapp); > + return NULL; > + } And this should not return failure on error since anything older than Linux 3.9 would fail here. It sounds like it would be safe to ignore the error and continue to get the existing behavior.
Hello and thank you for the feedback! On нд, 2015-11-15 at 19:16 +0200, Jouni Malinen wrote: > SO_REUSEPORT is a pretty recent addition. Wouldn't this break build > with older header files? As such, this should probably be protected > with #ifdef SO_REUSEPORT or something similar. Yes, SO_REUSEPORT is indeed a fairly new addition to Linux. I could protect it with an ifdef or I could just use SO_REUSEADDR. If I understand correctly there is little difference between the two with regards to multicast (with the exception of the UID checking for SO_REUSEPORT). Having said that, I have not tested SO_REUSEADDR yet. I will and depending on the results will either submit a new version with SO_REUSEADDR or one with an ifdef around the setsockopt invocation. > It sounds like it would be safe to ignore the error and continue to > get the existing behavior. Yes, it would be safe to continue. Should I make it just log a MSG_WARNING and carry on instead? -- Regards, P.
On Mon, Nov 16, 2015 at 04:56:42AM +0200, Petko Bordjukov wrote: > On нд, 2015-11-15 at 19:16 +0200, Jouni Malinen wrote: > > SO_REUSEPORT is a pretty recent addition. Wouldn't this break build > > with older header files? As such, this should probably be protected > > with #ifdef SO_REUSEPORT or something similar. > > Yes, SO_REUSEPORT is indeed a fairly new addition to Linux. I could > protect it with an ifdef or I could just use SO_REUSEADDR. If I > understand correctly there is little difference between the two with > regards to multicast (with the exception of the UID checking for > SO_REUSEPORT). > > Having said that, I have not tested SO_REUSEADDR yet. I will and > depending on the results will either submit a new version with > SO_REUSEADDR or one with an ifdef around the setsockopt invocation. I think I'm fine with either. Taken into account the current implementation has been there for years, this cannot really be a critical issue and using SO_REUSEPORT with ifdef should be fine. > > It sounds like it would be safe to ignore the error and continue to > > get the existing behavior. > > Yes, it would be safe to continue. Should I make it just log a > MSG_WARNING and carry on instead? Yes.
diff --git a/src/ap/iapp.c b/src/ap/iapp.c index 99aa04d..61776d8 100644 --- a/src/ap/iapp.c +++ b/src/ap/iapp.c @@ -385,6 +385,7 @@ struct iapp_data * iapp_init(struct hostapd_data *hapd, const char *iface) struct sockaddr_in *paddr, uaddr; struct iapp_data *iapp; struct ip_mreqn mreq; + int reuseport = 1; iapp = os_zalloc(sizeof(*iapp)); if (iapp == NULL) @@ -447,6 +448,15 @@ struct iapp_data * iapp_init(struct hostapd_data *hapd, const char *iface) os_memset(&uaddr, 0, sizeof(uaddr)); uaddr.sin_family = AF_INET; uaddr.sin_port = htons(IAPP_UDP_PORT); + + if (setsockopt(iapp->udp_sock, SOL_SOCKET, SO_REUSEPORT, &reuseport, + sizeof(reuseport)) < 0) { + wpa_printf(MSG_INFO, "iapp_init - setsockopt[UDP,SO_REUSEPORT]: %s", + strerror(errno)); + iapp_deinit(iapp); + return NULL; + } + if (bind(iapp->udp_sock, (struct sockaddr *) &uaddr, sizeof(uaddr)) < 0) { wpa_printf(MSG_INFO, "iapp_init - bind[UDP]: %s",
Make it possible for several instances of hostapd to listen on the same network interface. Signed-off-by: Petko Bordjukov <bordjukov@gmail.com> --- src/ap/iapp.c | 10 ++++++++++ 1 file changed, 10 insertions(+)