> It looks like this is down to autofs not handling names using only > IPv6 addresses. Not sure why this happens given that I would have > expected it to just pass this directly to mount(8) but it's > presumably doing more than that. Not sure exactly what the cause > is, though /usr/lib/x86_64-linux-gnu/autofs/mount_nfs.so is using > getaddrinfo and I can't see any obvious defect with a quick glance > over the sources.
Autofs does some initial NFS probing of its own, as part of handling server replication. From the debug output for a successful mount from a dual-stack server, we can also see that get_nfs_info() is only using the IPv4 address(es) of the server. IPv6 support seems to need libtirpc. If autofs is rebuilt using --with-libtirpc, both IPv6 and IPv4 addresses are used in get_nfs_info(), and the IPv6 only case now works. In wheezy at least, the following patch (or something similar): --- a/lib/rpc_subs.c +++ b/lib/rpc_subs.c @@ -34,6 +34,7 @@ #include <pthread.h> #include <poll.h> +/* #ifdef WITH_LIBTIRPC #undef auth_destroy #define auth_destroy(auth) \ @@ -43,6 +44,7 @@ ((*((auth)->ah_ops->ah_destroy))(auth));\ } while (0) #endif +*/ #include "mount.h" #include "rpc_subs.h" which removes a redefinition of auth_destroy() is also needed. The redefinition seems to have been an attempt to avoid a symbol clash on log_debug(), but in the wheezy macro there is no symbol clash, and the redefinition instead creates an undefined symbol auth_put(). Arne -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org