[PATCH] ipvsadm: catch the original errno from netlink answer

[PATCH] ipvsadm: catch the original errno from netlink answer

[Julian Anastasov]

nl_recvmsgs_default() returns NLE_* error codes and not
errno values. As result, attempt to delete virtual service
returns NLE_OBJ_NOTFOUND (12) which matches the ENOMEM value.

Problem as reported by Emanuele Rocca:

ipvsadm -D -t example.org:80
Memory allocation problem

Fix it by providing generic error handler to catch the errno
value as returned in netlink answer. By this way all netlink
commands will get proper error string. The problem is present
only when ipvsadm is compiled with libnl.

ipvsadm -D -t example.org:80
No such service

Reported-by: Emanuele Rocca <ema@wikimedia.org>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
 libipvs/libipvs.c | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/libipvs/libipvs.c b/libipvs/libipvs.c
index 180ea42..1cc46b9 100644
--- a/libipvs/libipvs.c
+++ b/libipvs/libipvs.c
@@ -74,9 +74,23 @@ static int ipvs_nl_noop_cb(struct nl_msg *msg, void *arg)
 	return NL_OK;
 }
 
+struct cb_err_data {
+	int	err;
+};
+
+static int ipvs_nl_err_cb(struct sockaddr_nl *nla, struct nlmsgerr *nlerr,
+			  void *arg)
+{
+	struct cb_err_data *data = arg;
+
+	data->err = nlerr->error;
+	return -nl_syserr2nlerr(nlerr->error);
+}
+
 int ipvs_nl_send_message(struct nl_msg *msg, nl_recvmsg_msg_cb_t func, void *arg)
 {
 	int err = EINVAL;
+	struct cb_err_data err_data = { .err = 0 };
 
 	sock = nl_socket_alloc();
 	if (!sock) {
@@ -100,12 +114,18 @@ int ipvs_nl_send_message(struct nl_msg *msg, nl_recvmsg_msg_cb_t func, void *arg
 
 	if (nl_socket_modify_cb(sock, NL_CB_VALID, NL_CB_CUSTOM, func, arg) != 0)
 		goto fail_genl;
+	if (nl_socket_modify_err_cb(sock, NL_CB_CUSTOM, ipvs_nl_err_cb,
+				    &err_data) != 0)
+		goto fail_genl;
 
 	if (nl_send_auto_complete(sock, msg) < 0)
 		goto fail_genl;
 
-	if ((err = -nl_recvmsgs_default(sock)) > 0)
+	if (nl_recvmsgs_default(sock) < 0) {
+		if (err_data.err)
+			err = -err_data.err;
 		goto fail_genl;
+	}
 
 	nlmsg_free(msg);
 
-- 
2.9.4

Re: [PATCH] ipvsadm: catch the original errno from netlink answer

[Jesper Dangaard Brouer]

On Sat,  5 Aug 2017 14:38:28 +0300
Julian Anastasov <ja@ssi.bg> wrote:

> nl_recvmsgs_default() returns NLE_* error codes and not
> errno values. As result, attempt to delete virtual service
> returns NLE_OBJ_NOTFOUND (12) which matches the ENOMEM value.
> 
> Problem as reported by Emanuele Rocca:
> 
> ipvsadm -D -t example.org:80
> Memory allocation problem
> 
> Fix it by providing generic error handler to catch the errno
> value as returned in netlink answer. By this way all netlink
> commands will get proper error string. The problem is present
> only when ipvsadm is compiled with libnl.
> 
> ipvsadm -D -t example.org:80
> No such service
> 
> Reported-by: Emanuele Rocca <ema@wikimedia.org>
> Signed-off-by: Julian Anastasov <ja@ssi.bg>

I've gone through the full call path from the kernel via netlink, and I
was going to claim that we also needed to handle errno "-EEXIST" in the
ipvsadm translation function ipvs_strerror().  Note, this fix uses the
errno "-ESRCH".

As kernel function ip_vs_del_service() return -EEXIST (if svc==NULL)
 http://elixir.free-electrons.com/linux/v4.12.5/source/net/netfilter/ipvs/ip_vs_ctl.c#L1480

BUT a closer look shows that -EEXIST will never get returned by
ip_vs_del_service() as all callers of this function does a svc==NULL
check and return "-ESRCH" instead.

Thus, this patch is correct, but the kernel code is confusing ;-)

Applied:
 https://git.kernel.org/pub/scm/utils/kernel/ipvsadm/ipvsadm.git/commit/?id=f8cff0808a24b

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer