ping: try next addrinfo on connect failure

On hosts that have routing rules matching on the outgoing interface [1],
getaddrinfo() may return results sorted in a suboptimal order because it is
not aware of the network interface passed to ping via the "-I" option.  In
particular, address reachability detection may fail and getaddrinfo() will
return ipv6 results first, even though the only routes available are ipv4.

Improve user experience by trying next addrinfo entry if we encounter a
failure at connect() time because of missing or unreachable routes.

[1] For example, on switches running Cumulus Linux, the default VRF is used
for front ports and a "mgmt" VRF is used for the management interface, which
also handles all DNS traffic.  (VRFs apply different routing rules based on
the iif/oif, ie.  influenced by SO_BINDTODEVICE.) In the default vrf, it's
possible to ping an ipv4 address via the mgmt vrf by specifying "-I mgmt".
However, that will fail if the target host is specified by name, has a AAAA
record and there is no ipv6 route to it.

Since libc commit 5ddb5bf5fb, getaddrinfo() does a udp connect to result
addresses to check if there is a route to them.  This is to implement
RFC3484 §6 Rule 1 ("Avoid unusable destinations") which is part of the
algorithm to order results.  getaddrinfo() is unaware of ping's "-I" option
and tries to connect its socket via the default vrf, which has no ipv6 route
to the target host (and, in fact, no ipv4 route either).  Following this
failure, getaddrinfo() returns results ordered according to
/etc/gai.conf (Rule 6) - by default, ipv6 first.

ping tries only the first entry returned by getaddrinfo() and fails to
connect to it because there is no ipv6 route to the host, even in the mgmt
vrf.  However, if getaddrinfo() had ordered the ipv4 result first or ping
had tried the next addrinfo entry (the ipv4 one), ping could connect a udp
socket to it and later successfully exchange icmp messages with it.

Example:

    cumulus@act-5812-10:~$ ip vrf list
    Name              Table
    -----------------------
    mgmt              1001
    cumulus@act-5812-10:~$ ip vrf identify
    cumulus@act-5812-10:~$ # --> default vrf
    cumulus@act-5812-10:~$
    cumulus@act-5812-10:~$ ip rule
    99:     from all to 10.230.0.53 ipproto udp dport 53 lookup mgmt
    99:     from all to 10.20.249.1 ipproto udp dport 53 lookup mgmt
    1000:   from all lookup [l3mdev-table]
    32765:  from all lookup local
    32766:  from all lookup main
    32767:  from all lookup default

    cumulus@act-5812-10:~$ ip route

    cumulus@act-5812-10:~$ ip -6 route
    ::1 dev lo proto kernel metric 256 pref medium

    cumulus@act-5812-10:~$ ip route show vrf mgmt
    default via 10.230.130.1 dev eth0
    unreachable default metric 4278198272
    10.230.130.0/24 dev eth0 proto kernel scope link src 10.230.130.211
    127.0.0.0/8 dev mgmt proto kernel scope link src 127.0.0.1

    cumulus@act-5812-10:~$ ip -6 route show vrf mgmt
    ::1 dev mgmt proto kernel metric 256 pref medium
    anycast fe80:: dev eth0 proto kernel metric 0 pref medium
    fe80::/64 dev eth0 proto kernel metric 256 pref medium
    ff00::/8 dev eth0 metric 256 pref medium
    unreachable default dev lo metric 4278198272 pref medium

    cumulus@act-5812-10:~$ host google.com
    google.com has address 172.217.0.46
    google.com has IPv6 address 2607:f8b0:4005:802::200e
    google.com mail is handled by 30 alt2.aspmx.l.google.com.
    google.com mail is handled by 40 alt3.aspmx.l.google.com.
    google.com mail is handled by 20 alt1.aspmx.l.google.com.
    google.com mail is handled by 10 aspmx.l.google.com.
    google.com mail is handled by 50 alt4.aspmx.l.google.com.

Success with numeric address

    cumulus@act-5812-10:~$ ping -n -c1 -I mgmt 172.217.0.46
    ping: Warning: source address might be selected on device other than mgmt.
    PING 172.217.0.46 (172.217.0.46) from 10.230.130.211 mgmt: 56(84) bytes of data.
    64 bytes from 172.217.0.46: icmp_seq=1 ttl=51 time=4.68 ms

    --- 172.217.0.46 ping statistics ---
    1 packets transmitted, 1 received, 0% packet loss, time 0ms
    rtt min/avg/max/mdev = 4.675/4.675/4.675/0.000 ms

Failure with host by name

    cumulus@act-5812-10:~$ ping -n -c1 -I mgmt google.com
    connect: No route to host

Success when running in the mgmt vrf because getaddrinfo()'s address
reachability test is effective and ipv4 result(s) are ordered first.

    cumulus@act-5812-10:~$ ip vrf exec mgmt ping -n -c1 google.com
    PING google.com (172.217.0.46) 56(84) bytes of data.
    64 bytes from 172.217.0.46: icmp_seq=1 ttl=51 time=4.65 ms

    --- google.com ping statistics ---
    1 packets transmitted, 1 received, 0% packet loss, time 0ms
    rtt min/avg/max/mdev = 4.650/4.650/4.650/0.000 ms

For demonstration purposes, the following configuration allows to reproduce
a similar problem.  Starting from a host with a vanilla configuration,
default ipv4 route using eth0, no ipv6 global routes:

    root@vsid:~# ip route
    default via 192.168.15.1 dev eth0
    192.168.15.0/24 dev eth0 proto kernel scope link src 192.168.15.100

    root@vsid:~# ip -6 route
    ::1 dev lo proto kernel metric 256 pref medium
    fe80::/64 dev eth0 proto kernel metric 256 pref medium

    root@vsid:~# ip rou flush table main

    root@vsid:~# ip rou add table 1 192.168.15.0/24 dev eth0

    root@vsid:~# ip rou add table 1 default via 192.168.15.1

    root@vsid:~# ip rule
    0:    from all lookup local
    32766:  from all lookup main
    32767:  from all lookup default
    root@vsid:~# ip rule add pref 1 to 192.168.15.1 ipproto udp dport 53 lookup 1
    root@vsid:~# ip rule add pref 2 oif eth0 lookup 1
    root@vsid:~# ping -c1 -I eth0 google.com

    ping: connect: Network is unreachable

With the current patch

    root@vsid:~# /src/iputils/builddir/ping/ping -c1 -I eth0 google.com
    PING  (172.217.174.110) from 192.168.15.100 eth0: 56(84) bytes of data.
    64 bytes from nrt12s28-in-f14.1e100.net (172.217.174.110): icmp_seq=1 ttl=53 time=11.3 ms

    ---  ping statistics ---
    1 packets transmitted, 1 received, 0% packet loss, time 0ms
    rtt min/avg/max/mdev = 11.313/11.313/11.313/0.000 ms

Signed-off-by: Benjamin Poirier <[email protected]>

ping/ping.c

@@ -697,6 +697,9 @@ int ping4_run(struct ping_rts *rts, int argc, char **argv, struct addrinfo *ai,
 					error(2, errno, _("cannot set broadcasting"));
 				if (connect(probe_fd, (struct sockaddr *)&dst, sizeof(dst)) == -1)
 					error(2, errno, "connect");
+			} else if ((errno == EHOSTUNREACH || errno == ENETUNREACH) && ai->ai_next) {
+				close(probe_fd);
+				return -1;
 			} else
 				error(2, errno, "connect");
 		}

ping/ping6_common.c

@@ -174,8 +174,13 @@ int ping6_run(struct ping_rts *rts, int argc, char **argv, struct addrinfo *ai,
 			rts->firsthop.sin6_family = AF_INET6;
 
 		rts->firsthop.sin6_port = htons(1025);
-		if (connect(probe_fd, (struct sockaddr *)&rts->firsthop, sizeof(rts->firsthop)) == -1)
+		if (connect(probe_fd, (struct sockaddr *)&rts->firsthop, sizeof(rts->firsthop)) == -1) {
+			if ((errno == EHOSTUNREACH || errno == ENETUNREACH) && ai->ai_next) {
+				close(probe_fd);
+				return -1;
+			}
 			error(2, errno, "connect");
+		}
 		alen = sizeof rts->source6;
 		if (getsockname(probe_fd, (struct sockaddr *)&rts->source6, &alen) == -1)
 			error(2, errno, "getsockname");