@@ -169,19 +169,14 @@ static void
handler_nl_event(struct uloop_fd *u, unsigned int events)
{
struct event_socket *ev = container_of(u, struct event_socket, uloop);
- int err;
- socklen_t errlen = sizeof(err);
+ int ret;
- if (!u->error) {
- nl_recvmsgs_default(ev->sock);
+ ret = nl_recvmsgs_default(ev->sock);
+ if (ret >= 0)
return;
- }
-
- if (getsockopt(u->fd, SOL_SOCKET, SO_ERROR, (void *)&err, &errlen))
- goto abort;
- switch(err) {
- case ENOBUFS:
+ switch (-ret) {
+ case NLE_NOMEM:
/* Increase rx buffer size on netlink socket */
ev->bufsize *= 2;
if (nl_socket_set_buffer_size(ev->sock, ev->bufsize, 0))
@@ -195,7 +190,6 @@ handler_nl_event(struct uloop_fd *u, unsigned int events)
default:
goto abort;
}
- u->error = false;
return;
abort:
@@ -797,24 +791,19 @@ handle_hotplug_event(struct uloop_fd *u, unsigned int events)
struct sockaddr_nl nla;
unsigned char *buf = NULL;
int size;
- int err;
- socklen_t errlen = sizeof(err);
- if (!u->error) {
- while ((size = nl_recv(ev->sock, &nla, &buf, NULL)) > 0) {
- if (nla.nl_pid == 0)
- handle_hotplug_msg((char *) buf, size);
+ while ((size = nl_recv(ev->sock, &nla, &buf, NULL)) > 0) {
+ if (nla.nl_pid == 0)
+ handle_hotplug_msg((char *) buf, size);
- free(buf);
- }
- return;
+ free(buf);
}
- if (getsockopt(u->fd, SOL_SOCKET, SO_ERROR, (void *)&err, &errlen))
- goto abort;
+ switch (-size) {
+ case 0:
+ return;
- switch(err) {
- case ENOBUFS:
+ case NLE_NOMEM:
/* Increase rx buffer size on netlink socket */
ev->bufsize *= 2;
if (nl_socket_set_buffer_size(ev->sock, ev->bufsize, 0))
@@ -824,7 +813,6 @@ handle_hotplug_event(struct uloop_fd *u, unsigned int events)
default:
goto abort;
}
- u->error = false;
return;
abort:
The error handling needed for the buffer growth logic relies on uloop_fd's error flag, which is set based on epoll events. Doing so without handling recvmsg's error codes is racy, as an error state may be set between receiving epoll events and the next recvmsg, but calling recvmsg clears the error state. To fix this, add handling for errors returned by nl_recvmsgs_default() and nl_recv(); checking for u->error and retrieving the error status using getsockopt() becomes redundant. We have observed this issue on Gluon (recent OpenWrt 23.05); on some devices with DSA switches, the bridge interface's carrier-on event would consistenly get lost during boot due to insufficient buffer space (see [1]). We have bisected the issue to netifd commit 516ab774cc16 ("system-linux: fix race condition on bringing up wireless devices"), but that commit only uncovered the preexisting bug by switching from getting the carrier state from sysfs to using the netlink messages in cb_rtnl_event(). I suspect that other recent issues about netifd missing a carrier state change like [2] may have the same underlying cause. [1] https://github.com/freifunk-gluon/gluon/issues/3130 [2] https://github.com/openwrt/openwrt/issues/13863 Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net> --- system-linux.c | 38 +++++++++++++------------------------- 1 file changed, 13 insertions(+), 25 deletions(-)