1feee8
commit 2b3d020055bea4fbbfc0ca2362d46038487c6dfd
1feee8
Author: Fabian Vogt <fvogt@suse.de>
1feee8
Date:   Wed Jul 27 11:44:07 2022 +0200
1feee8
1feee8
    nscd: Fix netlink cache invalidation if epoll is used [BZ #29415]
1feee8
    
1feee8
    Processes cache network interface information such as whether IPv4 or IPv6
1feee8
    are enabled. This is only checked again if the "netlink timestamp" provided
1feee8
    by nscd changed, which is triggered by netlink socket activity.
1feee8
    
1feee8
    However, in the epoll handler for the netlink socket, it was missed to
1feee8
    assign the new timestamp to the nscd database. The handler for plain poll
1feee8
    did that properly, copy that over.
1feee8
    
1feee8
    This bug caused that e.g. processes which started before network
1feee8
    configuration got unusuable addresses from getaddrinfo, like IPv6 only even
1feee8
    though only IPv4 is available:
1feee8
    https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/1041
1feee8
    
1feee8
    It's a bit hard to reproduce, so I verified this by checking the timestamp
1feee8
    on calls to __check_pf manually. Without this patch it's stuck at 1, now
1feee8
    it's increasing on network changes as expected.
1feee8
    
1feee8
    Signed-off-by: Fabian Vogt <fvogt@suse.de>
1feee8
    (cherry picked from commit 02ca25fef2785974011e9c5beecc99b900b69fd7)
1feee8
1feee8
diff --git a/nscd/connections.c b/nscd/connections.c
1feee8
index 3f0bda4e97edb9df..bc941715cff47c49 100644
1feee8
--- a/nscd/connections.c
1feee8
+++ b/nscd/connections.c
1feee8
@@ -2285,7 +2285,8 @@ main_loop_epoll (int efd)
1feee8
 					     sizeof (buf))) != -1)
1feee8
 	      ;
1feee8
 
1feee8
-	    __bump_nl_timestamp ();
1feee8
+	    dbs[hstdb].head->extra_data[NSCD_HST_IDX_CONF_TIMESTAMP]
1feee8
+	      = __bump_nl_timestamp ();
1feee8
 	  }
1feee8
 # endif
1feee8
 	else