oem-exoscale: make DHCP address retrieval more resilient

When retrievent the DHCP IP address (to get user data), two steps were
used: the first one waited for one lease file to become present and the
second one extracted and validated the server IP address. If for some
reason the second step failed, an empty IP was provided (and notably,
coreos-setup-environment.service never finishes correctly).

To avoid that, we combine the two steps into a loop: for each available
lease file, extract the IP and validate it. If it succeeds, return
it. If not continue. If nothing was found, restart the same process
after sleeping a bit.

sdk_container/src/third_party/coreos-overlay/coreos-base/oem-exoscale/files/exoscale-dhcp

@@ -2,22 +2,20 @@
 
 get_dhcp_ip() {
     local leases_dir="/run/systemd/netif/leases"
+    local found=0
     while true; do
-        if [[ "$(find "${leases_dir}" -type f -size +1c)" ]]; then
+        for leasefile in $(find "${leases_dir}" -type f -size +1c); do
-            break
-        fi
-        sleep .5
-    done
-
-    for leasefile in "${leases_dir}/"*; do
             dhcp_server_ip=$(cat $leasefile | awk -F= '/SERVER_ADDRESS/ { print $2 }')
             if [[ -n "${dhcp_server_ip}" ]]; then
                 metadata_url="http://${dhcp_server_ip}/latest/meta-data/"
-            curl --fail -s "${metadata_url}" >/dev/null
+                if curl --fail -s "${metadata_url}" >/dev/null; then
-            if [[ $? -eq 0 ]]; then
                     echo $dhcp_server_ip
+                    found=1
                     break
                 fi
             fi
         done
+        [[ $found -eq 0 ]] || break
+        sleep .5
+    done
 }