Merge pull request #875 from sarthurdev/dhcp_address_wait

ipsec: T1501: T3617: Add handling for missing addresses on boot when using dhcp-interface

5 files changed, 142 insertions, 50 deletions

diff --git a/data/templates/ipsec/ipsec.conf.tmpl b/data/templates/ipsec/ipsec.conf.tmpl
index 342887883..53cba44b9 100644
--- a/data/templates/ipsec/ipsec.conf.tmpl
+++ b/data/templates/ipsec/ipsec.conf.tmpl
@@ -7,7 +7,7 @@ config setup
     uniqueids = {{ "no" if disable_uniqreqids is defined else "yes" }}
 
 {% if site_to_site is defined and site_to_site.peer is defined %}
-{%   for peer, peer_conf in site_to_site.peer.items() %}
+{%   for peer, peer_conf in site_to_site.peer.items() if peer not in dhcp_no_address %}
 {%     set peer_index = loop.index %}
 {%     set peer_ike = ike_group[peer_conf.ike_group] %}
 {%     set peer_esp = esp_group[peer_conf.default_esp_group] if peer_conf.default_esp_group is defined else None %}
diff --git a/smoketest/scripts/cli/test_vpn_ipsec.py b/smoketest/scripts/cli/test_vpn_ipsec.py
index 4a3340ffb..820762fc2 100755
--- a/smoketest/scripts/cli/test_vpn_ipsec.py
+++ b/smoketest/scripts/cli/test_vpn_ipsec.py
@@ -20,6 +20,7 @@ from base_vyostest_shim import VyOSUnitTestSHIM
 
 from vyos.util import call, process_named_running, read_file
 
+ethernet_path = ['interfaces', 'ethernet']
 tunnel_path = ['interfaces', 'tunnel']
 nhrp_path = ['protocols', 'nhrp']
 base_path = ['vpn', 'ipsec']
@@ -29,8 +30,39 @@ class TestVPNIPsec(VyOSUnitTestSHIM.TestCase):
         self.cli_delete(base_path)
         self.cli_delete(nhrp_path)
         self.cli_delete(tunnel_path)
+        self.cli_delete(ethernet_path)
         self.cli_commit()
 
+    def test_dhcp_fail_handling(self):
+        self.cli_delete(ethernet_path)
+        self.cli_delete(base_path)
+
+        # Interface for dhcp-interface
+        self.cli_set(ethernet_path + ['eth0', 'address', 'dhcp'])
+
+        # Set IKE/ESP Groups
+        self.cli_set(base_path + ["esp-group", "MyESPGroup", "proposal", "1", "encryption", "aes128"])
+        self.cli_set(base_path + ["esp-group", "MyESPGroup", "proposal", "1", "hash", "sha1"])
+        self.cli_set(base_path + ["ike-group", "MyIKEGroup", "proposal", "1", "dh-group", "2"])
+        self.cli_set(base_path + ["ike-group", "MyIKEGroup", "proposal", "1", "encryption", "aes128"])
+        self.cli_set(base_path + ["ike-group", "MyIKEGroup", "proposal", "1", "hash", "sha1"])
+
+        # Site to site
+        self.cli_set(base_path + ["ipsec-interfaces", "interface", "eth0"])
+        self.cli_set(base_path + ["site-to-site", "peer", "203.0.113.45", "authentication", "mode", "pre-shared-secret"])
+        self.cli_set(base_path + ["site-to-site", "peer", "203.0.113.45", "authentication", "pre-shared-secret", "MYSECRETKEY"])
+        self.cli_set(base_path + ["site-to-site", "peer", "203.0.113.45", "ike-group", "MyIKEGroup"])
+        self.cli_set(base_path + ["site-to-site", "peer", "203.0.113.45", "default-esp-group", "MyESPGroup"])
+        self.cli_set(base_path + ["site-to-site", "peer", "203.0.113.45", "dhcp-interface", "eth0"])
+        self.cli_set(base_path + ["site-to-site", "peer", "203.0.113.45", "tunnel", "1", "protocol", "gre"])
+
+        self.cli_commit()
+
+        ipsec_dhcp_waiting = read_file('/tmp/ipsec_dhcp_waiting')
+
+        self.assertIn('eth0', ipsec_dhcp_waiting) # Ensure dhcp-failed interface was added for dhclient hook
+        self.assertTrue(process_named_running('charon')) # Commit should've still succeeded and launched charon
+
     def test_site_to_site(self):
         self.cli_delete(base_path)
 
diff --git a/src/conf_mode/vpn_ipsec.py b/src/conf_mode/vpn_ipsec.py
index 4efedd995..3eaa78a4b 100755
--- a/src/conf_mode/vpn_ipsec.py
+++ b/src/conf_mode/vpn_ipsec.py
@@ -23,6 +23,7 @@ from vyos.config import Config
 from vyos.configdict import leaf_node_changed
 from vyos.configverify import verify_interface_exists
 from vyos.ifconfig import Interface
+from vyos.template import ip_from_cidr
 from vyos.template import render
 from vyos.util import call
 from vyos.util import dict_search
@@ -73,12 +74,16 @@ any_log_modes = [
 ike_ciphers = {}
 esp_ciphers = {}
 
+dhcp_wait_attempts = 2
+dhcp_wait_sleep = 1
+
 mark_base = 0x900000
 
 CA_PATH = "/etc/ipsec.d/cacerts/"
 CRL_PATH = "/etc/ipsec.d/crls/"
 
 DHCP_BASE = "/var/lib/dhcp/dhclient"
+DHCP_HOOK_IFLIST="/tmp/ipsec_dhcp_waiting"
 
 LOCAL_KEY_PATHS = ['/config/auth/', '/config/ipsec.d/rsa-keys/']
 X509_PATH = '/config/auth/'
@@ -96,6 +101,7 @@ def get_config(config=None):
     ipsec = conf.get_config_dict(base, key_mangling=('-', '_'),
                                  get_first_key=True, no_tag_node_value_mangle=True)
 
+    ipsec['dhcp_no_address'] = {}
     ipsec['interface_change'] = leaf_node_changed(conf, base + ['ipsec-interfaces', 'interface'])
     ipsec['l2tp_exists'] = conf.exists('vpn l2tp remote-access ipsec-settings ')
     ipsec['nhrp_exists'] = conf.exists('protocols nhrp tunnel')
@@ -162,6 +168,15 @@ def verify_rsa_local_key(ipsec):
 def verify_rsa_key(ipsec, key_name):
     return dict_search(f'rsa_key_name.{key_name}.rsa_key', ipsec['rsa_keys'])
 
+def get_dhcp_address(iface):
+    addresses = Interface(iface).get_addr()
+    if not addresses:
+        return None
+    for address in addresses:
+        if not address.startswith("fe80:"): # Skip link-local ipv6
+            return ip_from_cidr(address)
+    return None
+
 def verify(ipsec):
     if not ipsec:
         return None
@@ -252,9 +267,17 @@ def verify(ipsec):
                 if not os.path.exists(f'{DHCP_BASE}_{dhcp_interface}.conf'):
                     raise ConfigError(f"Invalid dhcp-interface on site-to-site peer {peer}")
 
-                address = Interface(dhcp_interface).get_addr()
+                address = get_dhcp_address(dhcp_interface)
+                count = 0
+                while not address and count < dhcp_wait_attempts:
+                    address = get_dhcp_address(dhcp_interface)
+                    count += 1
+                    sleep(dhcp_wait_sleep)
+
                 if not address:
-                    raise ConfigError(f"Failed to get address from dhcp-interface on site-to-site peer {peer}")
+                    ipsec['dhcp_no_address'][peer] = dhcp_interface
+                    print(f"Failed to get address from dhcp-interface on site-to-site peer {peer} -- skipped")
+                    continue
 
             if 'vti' in peer_conf:
                 if 'local_address' in peer_conf and 'dhcp_interface' in peer_conf:
@@ -291,6 +314,10 @@ def generate(ipsec):
     data = {}
 
     if ipsec:
+        if ipsec['dhcp_no_address']:
+            with open(DHCP_HOOK_IFLIST, 'w') as f:
+                f.write(" ".join(ipsec['dhcp_no_address'].values()))
+
         data = ipsec
         data['authby'] = authby_translate
         data['ciphers'] = {'ike': ike_ciphers, 'esp': esp_ciphers}
@@ -300,6 +327,9 @@ def generate(ipsec):
 
         if 'site_to_site' in data and 'peer' in data['site_to_site']:
             for peer, peer_conf in ipsec['site_to_site']['peer'].items():
+                if peer in ipsec['dhcp_no_address']:
+                    continue
+
                 if peer_conf['authentication']['mode'] == 'x509':
                     ca_cert_file = os.path.join(X509_PATH, peer_conf['authentication']['x509']['ca_cert_file'])
                     call(f'cp -f {ca_cert_file} {CA_PATH}')
@@ -312,7 +342,7 @@ def generate(ipsec):
                 if 'local_address' in peer_conf:
                     local_ip = peer_conf['local_address']
                 elif 'dhcp_interface' in peer_conf:
-                    local_ip = Interface(peer_conf['dhcp_interface']).get_addr()
+                    local_ip = get_dhcp_address(peer_conf['dhcp_interface'])
 
                 data['site_to_site']['peer'][peer]['local_address'] = local_ip
 
diff --git a/src/etc/dhcp/dhclient-exit-hooks.d/ipsec-dhclient-hook b/src/etc/dhcp/dhclient-exit-hooks.d/ipsec-dhclient-hook
new file mode 100755
index 000000000..e00e5fe6e
--- /dev/null
+++ b/src/etc/dhcp/dhclient-exit-hooks.d/ipsec-dhclient-hook
@@ -0,0 +1,76 @@
+#!/bin/bash
+
+if [ "$reason" == "REBOOT" ] || [ "$reason" == "EXPIRE" ]; then
+    exit 0
+fi
+
+DHCP_HOOK_IFLIST="/tmp/ipsec_dhcp_waiting"
+
+if [ -f $DHCP_HOOK_IFLIST ] && [ "$reason" == "BOUND" ]; then
+    if grep -qw $interface $DHCP_HOOK_IFLIST; then
+        sudo rm $DHCP_HOOK_IFLIST
+        sudo python3 /usr/libexec/vyos/conf_mode/vpn_ipsec.py
+        exit 0
+    fi
+fi
+
+if [ "$old_ip_address" == "$new_ip_address" ] && [ "$reason" == "BOUND" ]; then
+    exit 0
+fi
+
+python3 - <<PYEND
+import os
+import re
+from vyos.util import call
+from vyos.util import cmd
+
+IPSEC_CONF="/etc/ipsec.conf"
+IPSEC_SECRETS="/etc/ipsec.secrets"
+
+def getlines(file):
+    with open(file, 'r') as f:
+        return f.readlines()
+
+def writelines(file, lines):
+    with open(file, 'w') as f:
+        f.writelines(lines)
+
+def ipsec_down(ip_address):
+    # This prevents the need to restart ipsec and kill all active connections, only the stale connection is closed
+    status = cmd('sudo ipsec statusall')
+    connection_name = None
+    for line in status.split("\n"):
+        if line.find(ip_address) > 0:
+            regex_match = re.search(r'(peer-[^:\[]+)', line)
+            if regex_match:
+                connection_name = regex_match[1]
+                break
+    if connection_name:
+        call(f'sudo ipsec down {connection_name}')
+
+if __name__ == '__main__':
+    interface = os.getenv('interface')
+    new_ip = os.getenv('new_ip_address')
+    old_ip = os.getenv('old_ip_address')
+
+    conf_lines = getlines(IPSEC_CONF)
+    secrets_lines = getlines(IPSEC_SECRETS)
+    found = False
+    to_match = f'# dhcp:{interface}'
+
+    for i, line in enumerate(conf_lines):
+        if line.find(to_match) > 0:
+            conf_lines[i] = line.replace(old_ip, new_ip)
+            found = True
+
+    for i, line in enumerate(secrets_lines):
+        if line.find(to_match) > 0:
+            secrets_lines[i] = line.replace(old_ip, new_ip)
+
+    if found:
+        writelines(IPSEC_CONF, conf_lines)
+        writelines(IPSEC_SECRETS, secrets_lines)
+        ipsec_down(old_ip)
+        call('sudo /usr/sbin/ipsec rereadall')
+        call('sudo /usr/sbin/ipsec reload')
+PYEND
\ No newline at end of file
diff --git a/src/etc/dhcp/dhclient-exit-hooks.d/ipsec-dhclient-hook.py b/src/etc/dhcp/dhclient-exit-hooks.d/ipsec-dhclient-hook.py
deleted file mode 100644
index 36edf04f3..000000000
--- a/src/etc/dhcp/dhclient-exit-hooks.d/ipsec-dhclient-hook.py
+++ /dev/null
@@ -1,46 +0,0 @@
-#!/usr/bin/env python3
-
-import os
-import sys
-
-from vyos.util import call
-
-IPSEC_CONF="/etc/ipsec.conf"
-IPSEC_SECRETS="/etc/ipsec.secrets"
-
-def getlines(file):
-    with open(file, 'r') as f:
-        return f.readlines()
-
-def writelines(file, lines):
-    with open(file, 'w') as f:
-        f.writelines(lines)
-
-if __name__ == '__main__':
-    interface = os.getenv('interface')
-    new_ip = os.getenv('new_ip_address')
-    old_ip = os.getenv('old_ip_address')
-    reason = os.getenv('reason')
-
-    if (old_ip == new_ip and reason != 'BOUND') or reason in ['REBOOT', 'EXPIRE']:
-        sys.exit(0)
-
-    conf_lines = getlines(IPSEC_CONF)
-    secrets_lines = getlines(IPSEC_SECRETS)
-    found = False
-    to_match = f'# dhcp:{interface}'
-
-    for i, line in enumerate(conf_lines):
-        if line.find(to_match) > 0:
-            conf_lines[i] = line.replace(old_ip, new_ip)
-            found = True
-
-    for i, line in enumerate(secrets_lines):
-        if line.find(to_match) > 0:
-            secrets_lines[i] = line.replace(old_ip, new_ip)
-
-    if found:
-        writelines(IPSEC_CONF, conf_lines)
-        writelines(IPSEC_SECRETS, secrets_lines)
-        call('sudo /usr/sbin/ipsec rereadall')
-        call('sudo /usr/sbin/ipsec reload')