blob: 17d37c1617232abb5b93555ff1d0bb5479950a07 [file] [log] [blame] [raw]
/* SPDX-License-Identifier: LGPL-2.1+ */
#include <netinet/in.h>
#include <linux/if.h>
#include <linux/if_arp.h>
#include <unistd.h>
#include "alloc-util.h"
#include "bus-util.h"
#include "dhcp-identifier.h"
#include "dhcp-lease-internal.h"
#include "env-file.h"
#include "fd-util.h"
#include "fileio.h"
#include "missing_network.h"
#include "netdev/bond.h"
#include "netdev/bridge.h"
#include "netdev/ipvlan.h"
#include "netdev/vrf.h"
#include "netlink-util.h"
#include "network-internal.h"
#include "networkd-can.h"
#include "networkd-dhcp-server.h"
#include "networkd-dhcp4.h"
#include "networkd-dhcp6.h"
#include "networkd-ipv4ll.h"
#include "networkd-ipv6-proxy-ndp.h"
#include "networkd-link-bus.h"
#include "networkd-link.h"
#include "networkd-lldp-tx.h"
#include "networkd-manager.h"
#include "networkd-ndisc.h"
#include "networkd-neighbor.h"
#include "networkd-radv.h"
#include "networkd-routing-policy-rule.h"
#include "set.h"
#include "socket-util.h"
#include "stdio-util.h"
#include "string-table.h"
#include "strv.h"
#include "sysctl-util.h"
#include "tmpfile-util.h"
#include "udev-util.h"
#include "util.h"
#include "virt.h"
uint32_t link_get_vrf_table(Link *link) {
return link->network->vrf ? VRF(link->network->vrf)->table : RT_TABLE_MAIN;
}
uint32_t link_get_dhcp_route_table(Link *link) {
/* When the interface is part of an VRF use the VRFs routing table, unless
* another table is explicitly specified. */
if (link->network->dhcp_route_table_set)
return link->network->dhcp_route_table;
return link_get_vrf_table(link);
}
uint32_t link_get_ipv6_accept_ra_route_table(Link *link) {
if (link->network->ipv6_accept_ra_route_table_set)
return link->network->ipv6_accept_ra_route_table;
return link_get_vrf_table(link);
}
DUID* link_get_duid(Link *link) {
if (link->network->duid.type != _DUID_TYPE_INVALID)
return &link->network->duid;
else
return &link->manager->duid;
}
int link_sysctl_ipv6_enabled(Link *link) {
_cleanup_free_ char *value = NULL;
int r;
assert(link);
assert(link->ifname);
if (link->sysctl_ipv6_enabled >= 0)
return link->sysctl_ipv6_enabled;
const char *ifname = link->ifname; /* work around bogus gcc warning */
r = sysctl_read_ip_property(AF_INET6, ifname, "disable_ipv6", &value);
if (r < 0)
return log_link_warning_errno(link, r,
"Failed to read net.ipv6.conf.%s.disable_ipv6 sysctl property: %m",
ifname);
link->sysctl_ipv6_enabled = value[0] == '0';
return link->sysctl_ipv6_enabled;
}
static bool link_dhcp6_enabled(Link *link) {
assert(link);
if (!socket_ipv6_is_supported())
return false;
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->network->bond)
return false;
if (link->iftype == ARPHRD_CAN)
return false;
if (link_sysctl_ipv6_enabled(link) == 0)
return false;
return link->network->dhcp & ADDRESS_FAMILY_IPV6;
}
static bool link_dhcp4_enabled(Link *link) {
assert(link);
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->network->bond)
return false;
if (link->iftype == ARPHRD_CAN)
return false;
return link->network->dhcp & ADDRESS_FAMILY_IPV4;
}
static bool link_dhcp4_server_enabled(Link *link) {
assert(link);
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->network->bond)
return false;
if (link->iftype == ARPHRD_CAN)
return false;
return link->network->dhcp_server;
}
bool link_ipv4ll_enabled(Link *link, AddressFamilyBoolean mask) {
assert(link);
assert((mask & ~(ADDRESS_FAMILY_IPV4 | ADDRESS_FAMILY_FALLBACK_IPV4)) == 0);
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->iftype == ARPHRD_CAN)
return false;
if (STRPTR_IN_SET(link->kind,
"vrf", "wireguard", "ipip", "gre", "ip6gre","ip6tnl", "sit", "vti",
"vti6", "nlmon", "xfrm"))
return false;
/* L3 or L3S mode do not support ARP. */
if (IN_SET(link_get_ipvlan_mode(link), NETDEV_IPVLAN_MODE_L3, NETDEV_IPVLAN_MODE_L3S))
return false;
if (link->network->bond)
return false;
return link->network->link_local & mask;
}
static bool link_ipv6ll_enabled(Link *link) {
assert(link);
if (!socket_ipv6_is_supported())
return false;
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->iftype == ARPHRD_CAN)
return false;
if (STRPTR_IN_SET(link->kind, "vrf", "wireguard", "ipip", "gre", "sit", "vti", "nlmon"))
return false;
if (link->network->bond)
return false;
if (link_sysctl_ipv6_enabled(link) == 0)
return false;
return link->network->link_local & ADDRESS_FAMILY_IPV6;
}
static bool link_ipv6_enabled(Link *link) {
assert(link);
if (!socket_ipv6_is_supported())
return false;
if (link->network->bond)
return false;
if (link_sysctl_ipv6_enabled(link) == 0)
return false;
if (link->iftype == ARPHRD_CAN)
return false;
/* DHCPv6 client will not be started if no IPv6 link-local address is configured. */
return link_ipv6ll_enabled(link) || network_has_static_ipv6_addresses(link->network);
}
static bool link_radv_enabled(Link *link) {
assert(link);
if (!link_ipv6ll_enabled(link))
return false;
return link->network->router_prefix_delegation != RADV_PREFIX_DELEGATION_NONE;
}
static bool link_ipv4_forward_enabled(Link *link) {
assert(link);
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->network->ip_forward == _ADDRESS_FAMILY_BOOLEAN_INVALID)
return false;
return link->network->ip_forward & ADDRESS_FAMILY_IPV4;
}
static bool link_ipv6_forward_enabled(Link *link) {
assert(link);
if (!socket_ipv6_is_supported())
return false;
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->network->ip_forward == _ADDRESS_FAMILY_BOOLEAN_INVALID)
return false;
if (link_sysctl_ipv6_enabled(link) == 0)
return false;
return link->network->ip_forward & ADDRESS_FAMILY_IPV6;
}
static bool link_proxy_arp_enabled(Link *link) {
assert(link);
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (link->network->proxy_arp < 0)
return false;
return true;
}
static bool link_ipv6_accept_ra_enabled(Link *link) {
assert(link);
if (!socket_ipv6_is_supported())
return false;
if (link->flags & IFF_LOOPBACK)
return false;
if (!link->network)
return false;
if (!link_ipv6ll_enabled(link))
return false;
/* If unset use system default (enabled if local forwarding is disabled.
* disabled if local forwarding is enabled).
* If set, ignore or enforce RA independent of local forwarding state.
*/
if (link->network->ipv6_accept_ra < 0)
/* default to accept RA if ip_forward is disabled and ignore RA if ip_forward is enabled */
return !link_ipv6_forward_enabled(link);
else if (link->network->ipv6_accept_ra > 0)
/* accept RA even if ip_forward is enabled */
return true;
else
/* ignore RA */
return false;
}
static IPv6PrivacyExtensions link_ipv6_privacy_extensions(Link *link) {
assert(link);
if (!socket_ipv6_is_supported())
return _IPV6_PRIVACY_EXTENSIONS_INVALID;
if (link->flags & IFF_LOOPBACK)
return _IPV6_PRIVACY_EXTENSIONS_INVALID;
if (!link->network)
return _IPV6_PRIVACY_EXTENSIONS_INVALID;
return link->network->ipv6_privacy_extensions;
}
static int link_enable_ipv6(Link *link) {
bool disabled;
int r;
if (link->flags & IFF_LOOPBACK)
return 0;
disabled = !link_ipv6_enabled(link);
r = sysctl_write_ip_property_boolean(AF_INET6, link->ifname, "disable_ipv6", disabled);
if (r < 0)
log_link_warning_errno(link, r, "Cannot %s IPv6: %m", enable_disable(!disabled));
else
log_link_info(link, "IPv6 successfully %sd", enable_disable(!disabled));
return 0;
}
static bool link_is_enslaved(Link *link) {
if (link->flags & IFF_SLAVE)
/* Even if the link is not managed by networkd, honor IFF_SLAVE flag. */
return true;
if (!link->network)
return false;
if (link->master_ifindex > 0 && link->network->bridge)
return true;
/* TODO: add conditions for other netdevs. */
return false;
}
static void link_update_master_operstate(Link *link, NetDev *netdev) {
Link *master;
if (!netdev)
return;
if (link_get(link->manager, netdev->ifindex, &master) < 0)
return;
link_update_operstate(master, true);
}
void link_update_operstate(Link *link, bool also_update_master) {
LinkOperationalState operstate;
LinkCarrierState carrier_state;
LinkAddressState address_state;
_cleanup_strv_free_ char **p = NULL;
uint8_t scope = RT_SCOPE_NOWHERE;
bool changed = false;
Address *address;
Iterator i;
assert(link);
if (link->kernel_operstate == IF_OPER_DORMANT)
carrier_state = LINK_CARRIER_STATE_DORMANT;
else if (link_has_carrier(link)) {
if (link_is_enslaved(link))
carrier_state = LINK_CARRIER_STATE_ENSLAVED;
else
carrier_state = LINK_CARRIER_STATE_CARRIER;
} else if (link->flags & IFF_UP)
carrier_state = LINK_CARRIER_STATE_NO_CARRIER;
else
carrier_state = LINK_CARRIER_STATE_OFF;
if (carrier_state >= LINK_CARRIER_STATE_CARRIER) {
Link *slave;
SET_FOREACH(slave, link->slaves, i) {
link_update_operstate(slave, false);
if (slave->carrier_state < LINK_CARRIER_STATE_CARRIER)
carrier_state = LINK_CARRIER_STATE_DEGRADED_CARRIER;
}
}
SET_FOREACH(address, link->addresses, i) {
if (!address_is_ready(address))
continue;
if (address->scope < scope)
scope = address->scope;
}
/* for operstate we also take foreign addresses into account */
SET_FOREACH(address, link->addresses_foreign, i) {
if (!address_is_ready(address))
continue;
if (address->scope < scope)
scope = address->scope;
}
if (scope < RT_SCOPE_SITE)
/* universally accessible addresses found */
address_state = LINK_ADDRESS_STATE_ROUTABLE;
else if (scope < RT_SCOPE_HOST)
/* only link or site local addresses found */
address_state = LINK_ADDRESS_STATE_DEGRADED;
else
/* no useful addresses found */
address_state = LINK_ADDRESS_STATE_OFF;
/* Mapping of address and carrier state vs operational state
* carrier state
* | off | no-carrier | dormant | degraded-carrier | carrier | enslaved
* ------------------------------------------------------------------------------
* off | off | no-carrier | dormant | degraded-carrier | carrier | enslaved
* address_state degraded | off | no-carrier | dormant | degraded-carrier | degraded | enslaved
* routable | off | no-carrier | dormant | degraded-carrier | routable | routable
*/
if (carrier_state < LINK_CARRIER_STATE_CARRIER || address_state == LINK_ADDRESS_STATE_OFF)
operstate = (LinkOperationalState) carrier_state;
else if (address_state == LINK_ADDRESS_STATE_ROUTABLE)
operstate = LINK_OPERSTATE_ROUTABLE;
else if (carrier_state == LINK_CARRIER_STATE_CARRIER)
operstate = LINK_OPERSTATE_DEGRADED;
else
operstate = LINK_OPERSTATE_ENSLAVED;
if (link->carrier_state != carrier_state) {
link->carrier_state = carrier_state;
changed = true;
if (strv_extend(&p, "CarrierState") < 0)
log_oom();
}
if (link->address_state != address_state) {
link->address_state = address_state;
changed = true;
if (strv_extend(&p, "AddressState") < 0)
log_oom();
}
if (link->operstate != operstate) {
link->operstate = operstate;
changed = true;
if (strv_extend(&p, "OperationalState") < 0)
log_oom();
}
if (p)
link_send_changed_strv(link, p);
if (changed)
link_dirty(link);
if (also_update_master && link->network) {
link_update_master_operstate(link, link->network->bond);
link_update_master_operstate(link, link->network->bridge);
}
}
#define FLAG_STRING(string, flag, old, new) \
(((old ^ new) & flag) \
? ((old & flag) ? (" -" string) : (" +" string)) \
: "")
static int link_update_flags(Link *link, sd_netlink_message *m, bool force_update_operstate) {
unsigned flags, unknown_flags_added, unknown_flags_removed, unknown_flags;
uint8_t operstate;
int r;
assert(link);
r = sd_rtnl_message_link_get_flags(m, &flags);
if (r < 0)
return log_link_warning_errno(link, r, "Could not get link flags: %m");
r = sd_netlink_message_read_u8(m, IFLA_OPERSTATE, &operstate);
if (r < 0)
/* if we got a message without operstate, take it to mean
the state was unchanged */
operstate = link->kernel_operstate;
if (!force_update_operstate && (link->flags == flags) && (link->kernel_operstate == operstate))
return 0;
if (link->flags != flags) {
log_link_debug(link, "Flags change:%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
FLAG_STRING("LOOPBACK", IFF_LOOPBACK, link->flags, flags),
FLAG_STRING("MASTER", IFF_MASTER, link->flags, flags),
FLAG_STRING("SLAVE", IFF_SLAVE, link->flags, flags),
FLAG_STRING("UP", IFF_UP, link->flags, flags),
FLAG_STRING("DORMANT", IFF_DORMANT, link->flags, flags),
FLAG_STRING("LOWER_UP", IFF_LOWER_UP, link->flags, flags),
FLAG_STRING("RUNNING", IFF_RUNNING, link->flags, flags),
FLAG_STRING("MULTICAST", IFF_MULTICAST, link->flags, flags),
FLAG_STRING("BROADCAST", IFF_BROADCAST, link->flags, flags),
FLAG_STRING("POINTOPOINT", IFF_POINTOPOINT, link->flags, flags),
FLAG_STRING("PROMISC", IFF_PROMISC, link->flags, flags),
FLAG_STRING("ALLMULTI", IFF_ALLMULTI, link->flags, flags),
FLAG_STRING("PORTSEL", IFF_PORTSEL, link->flags, flags),
FLAG_STRING("AUTOMEDIA", IFF_AUTOMEDIA, link->flags, flags),
FLAG_STRING("DYNAMIC", IFF_DYNAMIC, link->flags, flags),
FLAG_STRING("NOARP", IFF_NOARP, link->flags, flags),
FLAG_STRING("NOTRAILERS", IFF_NOTRAILERS, link->flags, flags),
FLAG_STRING("DEBUG", IFF_DEBUG, link->flags, flags),
FLAG_STRING("ECHO", IFF_ECHO, link->flags, flags));
unknown_flags = ~(IFF_LOOPBACK | IFF_MASTER | IFF_SLAVE | IFF_UP |
IFF_DORMANT | IFF_LOWER_UP | IFF_RUNNING |
IFF_MULTICAST | IFF_BROADCAST | IFF_POINTOPOINT |
IFF_PROMISC | IFF_ALLMULTI | IFF_PORTSEL |
IFF_AUTOMEDIA | IFF_DYNAMIC | IFF_NOARP |
IFF_NOTRAILERS | IFF_DEBUG | IFF_ECHO);
unknown_flags_added = ((link->flags ^ flags) & flags & unknown_flags);
unknown_flags_removed = ((link->flags ^ flags) & link->flags & unknown_flags);
/* link flags are currently at most 18 bits, let's align to
* printing 20 */
if (unknown_flags_added)
log_link_debug(link,
"Unknown link flags gained: %#.5x (ignoring)",
unknown_flags_added);
if (unknown_flags_removed)
log_link_debug(link,
"Unknown link flags lost: %#.5x (ignoring)",
unknown_flags_removed);
}
link->flags = flags;
link->kernel_operstate = operstate;
link_update_operstate(link, true);
return 0;
}
static int link_new(Manager *manager, sd_netlink_message *message, Link **ret) {
_cleanup_(link_unrefp) Link *link = NULL;
uint16_t type;
const char *ifname, *kind = NULL;
int r, ifindex;
unsigned short iftype;
assert(manager);
assert(message);
assert(ret);
/* check for link kind */
r = sd_netlink_message_enter_container(message, IFLA_LINKINFO);
if (r == 0) {
(void) sd_netlink_message_read_string(message, IFLA_INFO_KIND, &kind);
r = sd_netlink_message_exit_container(message);
if (r < 0)
return r;
}
r = sd_netlink_message_get_type(message, &type);
if (r < 0)
return r;
else if (type != RTM_NEWLINK)
return -EINVAL;
r = sd_rtnl_message_link_get_ifindex(message, &ifindex);
if (r < 0)
return r;
else if (ifindex <= 0)
return -EINVAL;
r = sd_rtnl_message_link_get_type(message, &iftype);
if (r < 0)
return r;
r = sd_netlink_message_read_string(message, IFLA_IFNAME, &ifname);
if (r < 0)
return r;
link = new(Link, 1);
if (!link)
return -ENOMEM;
*link = (Link) {
.n_ref = 1,
.manager = manager,
.state = LINK_STATE_PENDING,
.ifindex = ifindex,
.iftype = iftype,
.sysctl_ipv6_enabled = -1,
.n_dns = (unsigned) -1,
.dns_default_route = -1,
.llmnr = _RESOLVE_SUPPORT_INVALID,
.mdns = _RESOLVE_SUPPORT_INVALID,
.dnssec_mode = _DNSSEC_MODE_INVALID,
.dns_over_tls_mode = _DNS_OVER_TLS_MODE_INVALID,
};
link->ifname = strdup(ifname);
if (!link->ifname)
return -ENOMEM;
if (kind) {
link->kind = strdup(kind);
if (!link->kind)
return -ENOMEM;
}
r = sd_netlink_message_read_u32(message, IFLA_MASTER, (uint32_t *)&link->master_ifindex);
if (r < 0)
log_link_debug_errno(link, r, "New device has no master, continuing without");
r = sd_netlink_message_read_ether_addr(message, IFLA_ADDRESS, &link->mac);
if (r < 0)
log_link_debug_errno(link, r, "MAC address not found for new device, continuing without");
if (asprintf(&link->state_file, "/run/systemd/netif/links/%d", link->ifindex) < 0)
return -ENOMEM;
if (asprintf(&link->lease_file, "/run/systemd/netif/leases/%d", link->ifindex) < 0)
return -ENOMEM;
if (asprintf(&link->lldp_file, "/run/systemd/netif/lldp/%d", link->ifindex) < 0)
return -ENOMEM;
r = hashmap_ensure_allocated(&manager->links, NULL);
if (r < 0)
return r;
r = hashmap_put(manager->links, INT_TO_PTR(link->ifindex), link);
if (r < 0)
return r;
r = link_update_flags(link, message, false);
if (r < 0)
return r;
*ret = TAKE_PTR(link);
return 0;
}
void link_ntp_settings_clear(Link *link) {
link->ntp = strv_free(link->ntp);
}
void link_dns_settings_clear(Link *link) {
link->dns = mfree(link->dns);
link->n_dns = (unsigned) -1;
link->search_domains = ordered_set_free_free(link->search_domains);
link->route_domains = ordered_set_free_free(link->route_domains);
link->dns_default_route = -1;
link->llmnr = _RESOLVE_SUPPORT_INVALID;
link->mdns = _RESOLVE_SUPPORT_INVALID;
link->dnssec_mode = _DNSSEC_MODE_INVALID;
link->dns_over_tls_mode = _DNS_OVER_TLS_MODE_INVALID;
link->dnssec_negative_trust_anchors = set_free_free(link->dnssec_negative_trust_anchors);
}
static Link *link_free(Link *link) {
Address *address;
assert(link);
link_ntp_settings_clear(link);
link_dns_settings_clear(link);
link->routes = set_free_with_destructor(link->routes, route_free);
link->routes_foreign = set_free_with_destructor(link->routes_foreign, route_free);
link->addresses = set_free_with_destructor(link->addresses, address_free);
link->addresses_foreign = set_free_with_destructor(link->addresses_foreign, address_free);
while ((address = link->pool_addresses)) {
LIST_REMOVE(addresses, link->pool_addresses, address);
address_free(address);
}
sd_dhcp_server_unref(link->dhcp_server);
sd_dhcp_client_unref(link->dhcp_client);
sd_dhcp_lease_unref(link->dhcp_lease);
set_free(link->dhcp_routes);
link_lldp_emit_stop(link);
free(link->lease_file);
sd_lldp_unref(link->lldp);
free(link->lldp_file);
ndisc_flush(link);
sd_ipv4ll_unref(link->ipv4ll);
sd_dhcp6_client_unref(link->dhcp6_client);
sd_ndisc_unref(link->ndisc);
sd_radv_unref(link->radv);
free(link->ifname);
free(link->kind);
(void) unlink(link->state_file);
free(link->state_file);
sd_device_unref(link->sd_device);
hashmap_free(link->bound_to_links);
hashmap_free(link->bound_by_links);
set_free_with_destructor(link->slaves, link_unref);
network_unref(link->network);
return mfree(link);
}
DEFINE_TRIVIAL_REF_UNREF_FUNC(Link, link, link_free);
int link_get(Manager *m, int ifindex, Link **ret) {
Link *link;
assert(m);
assert(ifindex);
assert(ret);
link = hashmap_get(m->links, INT_TO_PTR(ifindex));
if (!link)
return -ENODEV;
*ret = link;
return 0;
}
void link_set_state(Link *link, LinkState state) {
assert(link);
if (link->state == state)
return;
log_link_debug(link, "State changed: %s -> %s",
link_state_to_string(link->state),
link_state_to_string(state));
link->state = state;
link_send_changed(link, "AdministrativeState", NULL);
}
static void link_enter_unmanaged(Link *link) {
assert(link);
link_set_state(link, LINK_STATE_UNMANAGED);
link_dirty(link);
}
int link_stop_clients(Link *link, bool may_keep_dhcp) {
int r = 0, k;
assert(link);
assert(link->manager);
assert(link->manager->event);
dhcp4_release_old_lease(link);
if (link->dhcp_client && (!may_keep_dhcp || !link->network ||
!FLAGS_SET(link->network->keep_configuration, KEEP_CONFIGURATION_DHCP_ON_STOP))) {
k = sd_dhcp_client_stop(link->dhcp_client);
if (k < 0)
r = log_link_warning_errno(link, k, "Could not stop DHCPv4 client: %m");
}
if (link->ipv4ll) {
k = sd_ipv4ll_stop(link->ipv4ll);
if (k < 0)
r = log_link_warning_errno(link, k, "Could not stop IPv4 link-local: %m");
}
if (link->dhcp6_client) {
k = sd_dhcp6_client_stop(link->dhcp6_client);
if (k < 0)
r = log_link_warning_errno(link, k, "Could not stop DHCPv6 client: %m");
}
if (link->ndisc) {
k = sd_ndisc_stop(link->ndisc);
if (k < 0)
r = log_link_warning_errno(link, k, "Could not stop IPv6 Router Discovery: %m");
}
if (link->radv) {
k = sd_radv_stop(link->radv);
if (k < 0)
r = log_link_warning_errno(link, k, "Could not stop IPv6 Router Advertisement: %m");
}
link_lldp_emit_stop(link);
return r;
}
void link_enter_failed(Link *link) {
assert(link);
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return;
log_link_warning(link, "Failed");
link_set_state(link, LINK_STATE_FAILED);
link_stop_clients(link, false);
link_dirty(link);
}
static int link_join_netdevs_after_configured(Link *link) {
NetDev *netdev;
Iterator i;
int r;
HASHMAP_FOREACH(netdev, link->network->stacked_netdevs, i) {
if (netdev->ifindex > 0)
/* Assume already enslaved. */
continue;
if (netdev_get_create_type(netdev) != NETDEV_CREATE_AFTER_CONFIGURED)
continue;
log_struct(LOG_DEBUG,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(netdev),
LOG_LINK_MESSAGE(link, "Enslaving by '%s'", netdev->ifname));
r = netdev_join(netdev, link, NULL);
if (r < 0)
return log_struct_errno(LOG_WARNING, r,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(netdev),
LOG_LINK_MESSAGE(link, "Could not join netdev '%s': %m", netdev->ifname));
}
return 0;
}
static void link_enter_configured(Link *link) {
assert(link);
assert(link->network);
if (link->state != LINK_STATE_CONFIGURING)
return;
log_link_info(link, "Configured");
link_set_state(link, LINK_STATE_CONFIGURED);
(void) link_join_netdevs_after_configured(link);
link_dirty(link);
}
static int link_request_set_routing_policy_rule(Link *link) {
RoutingPolicyRule *rule, *rrule = NULL;
int r;
assert(link);
assert(link->network);
link->routing_policy_rules_configured = false;
LIST_FOREACH(rules, rule, link->network->rules) {
r = routing_policy_rule_get(link->manager, rule, &rrule);
if (r >= 0) {
if (r == 0)
(void) routing_policy_rule_make_local(link->manager, rrule);
continue;
}
r = routing_policy_rule_configure(rule, link, NULL);
if (r < 0)
return log_link_warning_errno(link, r, "Could not set routing policy rules: %m");
if (r > 0)
link->routing_policy_rule_messages++;
}
routing_policy_rule_purge(link->manager, link);
if (link->routing_policy_rule_messages == 0) {
link->routing_policy_rules_configured = true;
link_check_ready(link);
} else {
log_link_debug(link, "Setting routing policy rules");
link_set_state(link, LINK_STATE_CONFIGURING);
}
return 0;
}
static int route_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(link);
assert(link->route_messages > 0);
assert(IN_SET(link->state, LINK_STATE_CONFIGURING,
LINK_STATE_FAILED, LINK_STATE_LINGER));
link->route_messages--;
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0 && r != -EEXIST) {
log_link_warning_errno(link, r, "Could not set route: %m");
link_enter_failed(link);
return 1;
}
if (link->route_messages == 0) {
log_link_debug(link, "Routes set");
link->static_routes_configured = true;
link_check_ready(link);
}
return 1;
}
int link_request_set_routes(Link *link) {
enum {
PHASE_NON_GATEWAY, /* First phase: Routes without a gateway */
PHASE_GATEWAY, /* Second phase: Routes with a gateway */
_PHASE_MAX
} phase;
Route *rt;
int r;
assert(link);
assert(link->network);
assert(link->addresses_configured);
assert(link->address_messages == 0);
assert(link->state != _LINK_STATE_INVALID);
link->static_routes_configured = false;
if (!link_has_carrier(link) && !link->network->configure_without_carrier)
/* During configuring addresses, the link lost its carrier. As networkd is dropping
* the addresses now, let's not configure the routes either. */
return 0;
r = link_request_set_routing_policy_rule(link);
if (r < 0)
return r;
/* First add the routes that enable us to talk to gateways, then add in the others that need a gateway. */
for (phase = 0; phase < _PHASE_MAX; phase++)
LIST_FOREACH(routes, rt, link->network->static_routes) {
if (in_addr_is_null(rt->family, &rt->gw) != (phase == PHASE_NON_GATEWAY))
continue;
r = route_configure(rt, link, route_handler);
if (r < 0)
return log_link_warning_errno(link, r, "Could not set routes: %m");
if (r > 0)
link->route_messages++;
}
if (link->route_messages == 0) {
link->static_routes_configured = true;
link_check_ready(link);
} else {
log_link_debug(link, "Setting routes");
link_set_state(link, LINK_STATE_CONFIGURING);
}
return 0;
}
void link_check_ready(Link *link) {
Address *a;
Iterator i;
int r;
assert(link);
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return;
if (!link->network)
return;
if (!link->addresses_configured)
return;
if (!link->neighbors_configured)
return;
SET_FOREACH(a, link->addresses, i)
if (!address_is_ready(a))
return;
if (!link->addresses_ready) {
link->addresses_ready = true;
r = link_request_set_routes(link);
if (r < 0)
link_enter_failed(link);
return;
}
if (!link->static_routes_configured)
return;
if (!link->routing_policy_rules_configured)
return;
if (link_has_carrier(link) || !link->network->configure_without_carrier) {
if (link_ipv4ll_enabled(link, ADDRESS_FAMILY_IPV4) && !(link->ipv4ll_address && link->ipv4ll_route))
return;
if (link_ipv6ll_enabled(link) &&
in_addr_is_null(AF_INET6, (const union in_addr_union*) &link->ipv6ll_address))
return;
if ((link_dhcp4_enabled(link) || link_dhcp6_enabled(link)) &&
!link->dhcp4_configured &&
!link->dhcp6_configured &&
!(link_ipv4ll_enabled(link, ADDRESS_FAMILY_FALLBACK_IPV4) && link->ipv4ll_address && link->ipv4ll_route))
/* When DHCP is enabled, at least one protocol must provide an address, or
* an IPv4ll fallback address must be configured. */
return;
if (link_ipv6_accept_ra_enabled(link) && !link->ndisc_configured)
return;
}
if (link->state != LINK_STATE_CONFIGURED)
link_enter_configured(link);
return;
}
static int link_request_set_neighbors(Link *link) {
Neighbor *neighbor;
int r;
assert(link);
assert(link->network);
assert(link->state != _LINK_STATE_INVALID);
link->neighbors_configured = false;
LIST_FOREACH(neighbors, neighbor, link->network->neighbors) {
r = neighbor_configure(neighbor, link, NULL);
if (r < 0)
return log_link_warning_errno(link, r, "Could not set neighbor: %m");
}
if (link->neighbor_messages == 0) {
link->neighbors_configured = true;
link_check_ready(link);
} else {
log_link_debug(link, "Setting neighbors");
link_set_state(link, LINK_STATE_CONFIGURING);
}
return 0;
}
static int address_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(rtnl);
assert(m);
assert(link);
assert(link->ifname);
assert(link->address_messages > 0);
assert(IN_SET(link->state, LINK_STATE_CONFIGURING,
LINK_STATE_FAILED, LINK_STATE_LINGER));
link->address_messages--;
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0 && r != -EEXIST) {
log_link_warning_errno(link, r, "could not set address: %m");
link_enter_failed(link);
return 1;
} else if (r >= 0)
(void) manager_rtnl_process_address(rtnl, m, link->manager);
if (link->address_messages == 0) {
log_link_debug(link, "Addresses set");
link->addresses_configured = true;
link_check_ready(link);
}
return 1;
}
static int link_set_bridge_fdb(Link *link) {
FdbEntry *fdb_entry;
int r;
LIST_FOREACH(static_fdb_entries, fdb_entry, link->network->static_fdb_entries) {
r = fdb_entry_configure(link, fdb_entry);
if (r < 0)
return log_link_error_errno(link, r, "Failed to add MAC entry to static MAC table: %m");
}
return 0;
}
static int link_request_set_addresses(Link *link) {
AddressLabel *label;
Address *ad;
int r;
assert(link);
assert(link->network);
assert(link->state != _LINK_STATE_INVALID);
/* Reset all *_configured flags we are configuring. */
link->addresses_configured = false;
link->addresses_ready = false;
link->neighbors_configured = false;
link->static_routes_configured = false;
link->routing_policy_rules_configured = false;
r = link_set_bridge_fdb(link);
if (r < 0)
return r;
r = link_request_set_neighbors(link);
if (r < 0)
return r;
LIST_FOREACH(addresses, ad, link->network->static_addresses) {
bool update;
update = address_get(link, ad->family, &ad->in_addr, ad->prefixlen, NULL) > 0;
r = address_configure(ad, link, address_handler, update);
if (r < 0)
return log_link_warning_errno(link, r, "Could not set addresses: %m");
if (r > 0)
link->address_messages++;
}
LIST_FOREACH(labels, label, link->network->address_labels) {
r = address_label_configure(label, link, NULL, false);
if (r < 0)
return log_link_warning_errno(link, r, "Could not set address label: %m");
link->address_label_messages++;
}
/* now that we can figure out a default address for the dhcp server,
start it */
if (link_dhcp4_server_enabled(link) && (link->flags & IFF_UP)) {
r = dhcp4_server_configure(link);
if (r < 0)
return r;
log_link_debug(link, "Offering DHCPv4 leases");
}
if (link->address_messages == 0) {
link->addresses_configured = true;
link_check_ready(link);
} else {
log_link_debug(link, "Setting addresses");
link_set_state(link, LINK_STATE_CONFIGURING);
}
return 0;
}
static int link_set_bridge_vlan(Link *link) {
int r;
r = br_vlan_configure(link, link->network->pvid, link->network->br_vid_bitmap, link->network->br_untagged_bitmap);
if (r < 0)
log_link_error_errno(link, r, "Failed to assign VLANs to bridge port: %m");
return r;
}
static int link_set_proxy_arp(Link *link) {
int r;
if (!link_proxy_arp_enabled(link))
return 0;
r = sysctl_write_ip_property_boolean(AF_INET, link->ifname, "proxy_arp", link->network->proxy_arp > 0);
if (r < 0)
log_link_warning_errno(link, r, "Cannot configure proxy ARP for interface: %m");
return 0;
}
static int link_configure_after_setting_mtu(Link *link);
static int set_mtu_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(m);
assert(link);
assert(link->ifname);
link->setting_mtu = false;
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0)
log_link_warning_errno(link, r, "Could not set MTU, ignoring: %m");
else
log_link_debug(link, "Setting MTU done.");
if (link->state == LINK_STATE_INITIALIZED) {
r = link_configure_after_setting_mtu(link);
if (r < 0)
link_enter_failed(link);
}
return 1;
}
int link_set_mtu(Link *link, uint32_t mtu) {
_cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
int r;
assert(link);
assert(link->manager);
assert(link->manager->rtnl);
if (mtu == 0 || link->setting_mtu)
return 0;
if (link->mtu == mtu)
return 0;
log_link_debug(link, "Setting MTU: %" PRIu32, mtu);
r = sd_rtnl_message_new_link(link->manager->rtnl, &req, RTM_SETLINK, link->ifindex);
if (r < 0)
return log_link_error_errno(link, r, "Could not allocate RTM_SETLINK message: %m");
/* If IPv6 not configured (no static IPv6 address and IPv6LL autoconfiguration is disabled)
* for this interface, then disable IPv6 else enable it. */
(void) link_enable_ipv6(link);
/* IPv6 protocol requires a minimum MTU of IPV6_MTU_MIN(1280) bytes
* on the interface. Bump up MTU bytes to IPV6_MTU_MIN. */
if (link_ipv6_enabled(link) && mtu < IPV6_MIN_MTU) {
log_link_warning(link, "Bumping MTU to " STRINGIFY(IPV6_MIN_MTU) ", as "
"IPv6 is requested and requires a minimum MTU of " STRINGIFY(IPV6_MIN_MTU) " bytes: %m");
mtu = IPV6_MIN_MTU;
}
r = sd_netlink_message_append_u32(req, IFLA_MTU, mtu);
if (r < 0)
return log_link_error_errno(link, r, "Could not append MTU: %m");
r = netlink_call_async(link->manager->rtnl, NULL, req, set_mtu_handler,
link_netlink_destroy_callback, link);
if (r < 0)
return log_link_error_errno(link, r, "Could not send rtnetlink message: %m");
link_ref(link);
link->setting_mtu = true;
return 0;
}
static bool link_reduces_vlan_mtu(Link *link) {
/* See netif_reduces_vlan_mtu() in kernel. */
return streq_ptr(link->kind, "macsec");
}
static uint32_t link_get_requested_mtu_by_stacked_netdevs(Link *link) {
uint32_t mtu = 0;
NetDev *dev;
Iterator i;
HASHMAP_FOREACH(dev, link->network->stacked_netdevs, i)
if (dev->kind == NETDEV_KIND_VLAN && dev->mtu > 0)
/* See vlan_dev_change_mtu() in kernel. */
mtu = MAX(mtu, link_reduces_vlan_mtu(link) ? dev->mtu + 4 : dev->mtu);
else if (dev->kind == NETDEV_KIND_MACVLAN && dev->mtu > mtu)
/* See macvlan_change_mtu() in kernel. */
mtu = dev->mtu;
return mtu;
}
static int link_configure_mtu(Link *link) {
uint32_t mtu;
assert(link);
assert(link->network);
if (link->network->mtu > 0)
return link_set_mtu(link, link->network->mtu);
mtu = link_get_requested_mtu_by_stacked_netdevs(link);
if (link->mtu >= mtu)
return 0;
log_link_notice(link, "Bumping MTU bytes from %"PRIu32" to %"PRIu32" because of stacked device. "
"If it is not desired, then please explicitly specify MTUBytes= setting.",
link->mtu, mtu);
return link_set_mtu(link, mtu);
}
static int set_flags_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(m);
assert(link);
assert(link->ifname);
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0)
log_link_warning_errno(link, r, "Could not set link flags, ignoring: %m");
return 1;
}
static int link_set_flags(Link *link) {
_cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
unsigned ifi_change = 0;
unsigned ifi_flags = 0;
int r;
assert(link);
assert(link->manager);
assert(link->manager->rtnl);
if (link->flags & IFF_LOOPBACK)
return 0;
if (!link->network)
return 0;
if (link->network->arp < 0 && link->network->multicast < 0 && link->network->allmulticast < 0)
return 0;
r = sd_rtnl_message_new_link(link->manager->rtnl, &req, RTM_SETLINK, link->ifindex);
if (r < 0)
return log_link_error_errno(link, r, "Could not allocate RTM_SETLINK message: %m");
if (link->network->arp >= 0) {
ifi_change |= IFF_NOARP;
SET_FLAG(ifi_flags, IFF_NOARP, link->network->arp == 0);
}
if (link->network->multicast >= 0) {
ifi_change |= IFF_MULTICAST;
SET_FLAG(ifi_flags, IFF_MULTICAST, link->network->multicast);
}
if (link->network->allmulticast >= 0) {
ifi_change |= IFF_ALLMULTI;
SET_FLAG(ifi_flags, IFF_ALLMULTI, link->network->allmulticast);
}
r = sd_rtnl_message_link_set_flags(req, ifi_flags, ifi_change);
if (r < 0)
return log_link_error_errno(link, r, "Could not set link flags: %m");
r = netlink_call_async(link->manager->rtnl, NULL, req, set_flags_handler,
link_netlink_destroy_callback, link);
if (r < 0)
return log_link_error_errno(link, r, "Could not send rtnetlink message: %m");
link_ref(link);
return 0;
}
static int link_acquire_ipv6_conf(Link *link) {
int r;
assert(link);
if (link_ipv6_accept_ra_enabled(link)) {
assert(link->ndisc);
log_link_debug(link, "Discovering IPv6 routers");
r = sd_ndisc_start(link->ndisc);
if (r < 0 && r != -EBUSY)
return log_link_warning_errno(link, r, "Could not start IPv6 Router Discovery: %m");
}
if (link_radv_enabled(link)) {
assert(link->radv);
assert(in_addr_is_link_local(AF_INET6, (const union in_addr_union*)&link->ipv6ll_address) > 0);
log_link_debug(link, "Starting IPv6 Router Advertisements");
r = sd_radv_start(link->radv);
if (r < 0 && r != -EBUSY)
return log_link_warning_errno(link, r, "Could not start IPv6 Router Advertisement: %m");
}
(void) dhcp6_request_prefix_delegation(link);
return 0;
}
static int link_acquire_ipv4_conf(Link *link) {
int r;
assert(link);
assert(link->manager);
assert(link->manager->event);
if (link_ipv4ll_enabled(link, ADDRESS_FAMILY_IPV4)) {
assert(link->ipv4ll);
log_link_debug(link, "Acquiring IPv4 link-local address");
r = sd_ipv4ll_start(link->ipv4ll);
if (r < 0)
return log_link_warning_errno(link, r, "Could not acquire IPv4 link-local address: %m");
}
if (link_dhcp4_enabled(link)) {
assert(link->dhcp_client);
log_link_debug(link, "Acquiring DHCPv4 lease");
r = sd_dhcp_client_start(link->dhcp_client);
if (r < 0)
return log_link_warning_errno(link, r, "Could not acquire DHCPv4 lease: %m");
}
return 0;
}
static int link_acquire_conf(Link *link) {
int r;
assert(link);
r = link_acquire_ipv4_conf(link);
if (r < 0)
return r;
if (!in_addr_is_null(AF_INET6, (const union in_addr_union*) &link->ipv6ll_address)) {
r = link_acquire_ipv6_conf(link);
if (r < 0)
return r;
}
if (link_lldp_emit_enabled(link)) {
r = link_lldp_emit_start(link);
if (r < 0)
return log_link_warning_errno(link, r, "Failed to start LLDP transmission: %m");
}
return 0;
}
bool link_has_carrier(Link *link) {
/* see Documentation/networking/operstates.txt in the kernel sources */
if (link->kernel_operstate == IF_OPER_UP)
return true;
if (link->kernel_operstate == IF_OPER_UNKNOWN)
/* operstate may not be implemented, so fall back to flags */
if (FLAGS_SET(link->flags, IFF_LOWER_UP | IFF_RUNNING) &&
!FLAGS_SET(link->flags, IFF_DORMANT))
return true;
return false;
}
static int link_address_genmode_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(link);
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0)
log_link_warning_errno(link, r, "Could not set address genmode for interface, ignoring: %m");
return 1;
}
static int link_configure_addrgen_mode(Link *link) {
_cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
uint8_t ipv6ll_mode;
int r;
assert(link);
assert(link->network);
assert(link->manager);
assert(link->manager->rtnl);
if (!socket_ipv6_is_supported())
return 0;
log_link_debug(link, "Setting address genmode for link");
r = sd_rtnl_message_new_link(link->manager->rtnl, &req, RTM_SETLINK, link->ifindex);
if (r < 0)
return log_link_error_errno(link, r, "Could not allocate RTM_SETLINK message: %m");
r = sd_netlink_message_open_container(req, IFLA_AF_SPEC);
if (r < 0)
return log_link_error_errno(link, r, "Could not open IFLA_AF_SPEC container: %m");
r = sd_netlink_message_open_container(req, AF_INET6);
if (r < 0)
return log_link_error_errno(link, r, "Could not open AF_INET6 container: %m");
if (!link_ipv6ll_enabled(link))
ipv6ll_mode = IN6_ADDR_GEN_MODE_NONE;
else if (sysctl_read_ip_property(AF_INET6, link->ifname, "stable_secret", NULL) < 0)
/* The file may not exist. And event if it exists, when stable_secret is unset,
* reading the file fails with EIO. */
ipv6ll_mode = IN6_ADDR_GEN_MODE_EUI64;
else
ipv6ll_mode = IN6_ADDR_GEN_MODE_STABLE_PRIVACY;
r = sd_netlink_message_append_u8(req, IFLA_INET6_ADDR_GEN_MODE, ipv6ll_mode);
if (r < 0)
return log_link_error_errno(link, r, "Could not append IFLA_INET6_ADDR_GEN_MODE: %m");
r = sd_netlink_message_close_container(req);
if (r < 0)
return log_link_error_errno(link, r, "Could not close AF_INET6 container: %m");
r = sd_netlink_message_close_container(req);
if (r < 0)
return log_link_error_errno(link, r, "Could not close IFLA_AF_SPEC container: %m");
r = netlink_call_async(link->manager->rtnl, NULL, req, link_address_genmode_handler,
link_netlink_destroy_callback, link);
if (r < 0)
return log_link_error_errno(link, r, "Could not send rtnetlink message: %m");
link_ref(link);
return 0;
}
static int link_up_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(link);
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0)
/* we warn but don't fail the link, as it may be brought up later */
log_link_warning_errno(link, r, "Could not bring up interface: %m");
return 1;
}
static int link_up(Link *link) {
_cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
int r;
assert(link);
assert(link->network);
assert(link->manager);
assert(link->manager->rtnl);
log_link_debug(link, "Bringing link up");
r = sd_rtnl_message_new_link(link->manager->rtnl, &req, RTM_SETLINK, link->ifindex);
if (r < 0)
return log_link_error_errno(link, r, "Could not allocate RTM_SETLINK message: %m");
/* set it free if not enslaved with networkd */
if (!link->network->bridge && !link->network->bond && !link->network->vrf) {
r = sd_netlink_message_append_u32(req, IFLA_MASTER, 0);
if (r < 0)
return log_link_error_errno(link, r, "Could not append IFLA_MASTER attribute: %m");
}
r = sd_rtnl_message_link_set_flags(req, IFF_UP, IFF_UP);
if (r < 0)
return log_link_error_errno(link, r, "Could not set link flags: %m");
if (link->network->mac) {
r = sd_netlink_message_append_ether_addr(req, IFLA_ADDRESS, link->network->mac);
if (r < 0)
return log_link_error_errno(link, r, "Could not set MAC address: %m");
}
r = netlink_call_async(link->manager->rtnl, NULL, req, link_up_handler,
link_netlink_destroy_callback, link);
if (r < 0)
return log_link_error_errno(link, r, "Could not send rtnetlink message: %m");
link_ref(link);
return 0;
}
static int link_down_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(link);
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0)
log_link_warning_errno(link, r, "Could not bring down interface: %m");
return 1;
}
int link_down(Link *link, link_netlink_message_handler_t callback) {
_cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
int r;
assert(link);
assert(link->manager);
assert(link->manager->rtnl);
log_link_debug(link, "Bringing link down");
r = sd_rtnl_message_new_link(link->manager->rtnl, &req,
RTM_SETLINK, link->ifindex);
if (r < 0)
return log_link_error_errno(link, r, "Could not allocate RTM_SETLINK message: %m");
r = sd_rtnl_message_link_set_flags(req, 0, IFF_UP);
if (r < 0)
return log_link_error_errno(link, r, "Could not set link flags: %m");
r = netlink_call_async(link->manager->rtnl, NULL, req,
callback ?: link_down_handler,
link_netlink_destroy_callback, link);
if (r < 0)
return log_link_error_errno(link, r, "Could not send rtnetlink message: %m");
link_ref(link);
return 0;
}
static int link_handle_bound_to_list(Link *link) {
Link *l;
Iterator i;
int r;
bool required_up = false;
bool link_is_up = false;
assert(link);
if (hashmap_isempty(link->bound_to_links))
return 0;
if (link->flags & IFF_UP)
link_is_up = true;
HASHMAP_FOREACH (l, link->bound_to_links, i)
if (link_has_carrier(l)) {
required_up = true;
break;
}
if (!required_up && link_is_up) {
r = link_down(link, NULL);
if (r < 0)
return r;
} else if (required_up && !link_is_up) {
r = link_up(link);
if (r < 0)
return r;
}
return 0;
}
static int link_handle_bound_by_list(Link *link) {
Iterator i;
Link *l;
int r;
assert(link);
if (hashmap_isempty(link->bound_by_links))
return 0;
HASHMAP_FOREACH (l, link->bound_by_links, i) {
r = link_handle_bound_to_list(l);
if (r < 0)
return r;
}
return 0;
}
static int link_put_carrier(Link *link, Link *carrier, Hashmap **h) {
int r;
assert(link);
assert(carrier);
if (link == carrier)
return 0;
if (hashmap_get(*h, INT_TO_PTR(carrier->ifindex)))
return 0;
r = hashmap_ensure_allocated(h, NULL);
if (r < 0)
return r;
r = hashmap_put(*h, INT_TO_PTR(carrier->ifindex), carrier);
if (r < 0)
return r;
return 0;
}
static int link_new_bound_by_list(Link *link) {
Manager *m;
Link *carrier;
Iterator i;
int r;
bool list_updated = false;
assert(link);
assert(link->manager);
m = link->manager;
HASHMAP_FOREACH(carrier, m->links, i) {
if (!carrier->network)
continue;
if (strv_isempty(carrier->network->bind_carrier))
continue;
if (strv_fnmatch(carrier->network->bind_carrier, link->ifname, 0)) {
r = link_put_carrier(link, carrier, &link->bound_by_links);
if (r < 0)
return r;
list_updated = true;
}
}
if (list_updated)
link_dirty(link);
HASHMAP_FOREACH(carrier, link->bound_by_links, i) {
r = link_put_carrier(carrier, link, &carrier->bound_to_links);
if (r < 0)
return r;
link_dirty(carrier);
}
return 0;
}
static int link_new_bound_to_list(Link *link) {
Manager *m;
Link *carrier;
Iterator i;
int r;
bool list_updated = false;
assert(link);
assert(link->manager);
if (!link->network)
return 0;
if (strv_isempty(link->network->bind_carrier))
return 0;
m = link->manager;
HASHMAP_FOREACH (carrier, m->links, i) {
if (strv_fnmatch(link->network->bind_carrier, carrier->ifname, 0)) {
r = link_put_carrier(link, carrier, &link->bound_to_links);
if (r < 0)
return r;
list_updated = true;
}
}
if (list_updated)
link_dirty(link);
HASHMAP_FOREACH (carrier, link->bound_to_links, i) {
r = link_put_carrier(carrier, link, &carrier->bound_by_links);
if (r < 0)
return r;
link_dirty(carrier);
}
return 0;
}
static int link_new_carrier_maps(Link *link) {
int r;
r = link_new_bound_by_list(link);
if (r < 0)
return r;
r = link_handle_bound_by_list(link);
if (r < 0)
return r;
r = link_new_bound_to_list(link);
if (r < 0)
return r;
r = link_handle_bound_to_list(link);
if (r < 0)
return r;
return 0;
}
static void link_free_bound_to_list(Link *link) {
Link *bound_to;
Iterator i;
HASHMAP_FOREACH (bound_to, link->bound_to_links, i) {
hashmap_remove(link->bound_to_links, INT_TO_PTR(bound_to->ifindex));
if (hashmap_remove(bound_to->bound_by_links, INT_TO_PTR(link->ifindex)))
link_dirty(bound_to);
}
return;
}
static void link_free_bound_by_list(Link *link) {
Link *bound_by;
Iterator i;
HASHMAP_FOREACH (bound_by, link->bound_by_links, i) {
hashmap_remove(link->bound_by_links, INT_TO_PTR(bound_by->ifindex));
if (hashmap_remove(bound_by->bound_to_links, INT_TO_PTR(link->ifindex))) {
link_dirty(bound_by);
link_handle_bound_to_list(bound_by);
}
}
return;
}
static void link_free_carrier_maps(Link *link) {
bool list_updated = false;
assert(link);
if (!hashmap_isempty(link->bound_to_links)) {
link_free_bound_to_list(link);
list_updated = true;
}
if (!hashmap_isempty(link->bound_by_links)) {
link_free_bound_by_list(link);
list_updated = true;
}
if (list_updated)
link_dirty(link);
return;
}
static int link_append_to_master(Link *link, NetDev *netdev) {
Link *master;
int r;
assert(link);
assert(netdev);
r = link_get(link->manager, netdev->ifindex, &master);
if (r < 0)
return r;
r = set_ensure_allocated(&master->slaves, NULL);
if (r < 0)
return r;
r = set_put(master->slaves, link);
if (r < 0)
return r;
link_ref(link);
return 0;
}
static void link_drop_from_master(Link *link, NetDev *netdev) {
Link *master;
assert(link);
if (!link->manager || !netdev)
return;
if (link_get(link->manager, netdev->ifindex, &master) < 0)
return;
link_unref(set_remove(master->slaves, link));
}
static void link_detach_from_manager(Link *link) {
if (!link || !link->manager)
return;
link_unref(set_remove(link->manager->links_requesting_uuid, link));
link_clean(link);
/* The following must be called at last. */
assert_se(hashmap_remove(link->manager->links, INT_TO_PTR(link->ifindex)) == link);
link_unref(link);
}
void link_drop(Link *link) {
if (!link || link->state == LINK_STATE_LINGER)
return;
link_set_state(link, LINK_STATE_LINGER);
link_free_carrier_maps(link);
if (link->network) {
link_drop_from_master(link, link->network->bridge);
link_drop_from_master(link, link->network->bond);
}
log_link_debug(link, "Link removed");
(void) unlink(link->state_file);
link_detach_from_manager(link);
}
static int link_joined(Link *link) {
int r;
assert(link);
assert(link->network);
if (!hashmap_isempty(link->bound_to_links)) {
r = link_handle_bound_to_list(link);
if (r < 0)
return r;
} else if (!(link->flags & IFF_UP)) {
r = link_up(link);
if (r < 0) {
link_enter_failed(link);
return r;
}
}
if (link->network->bridge) {
r = link_set_bridge(link);
if (r < 0)
log_link_error_errno(link, r, "Could not set bridge message: %m");
r = link_append_to_master(link, link->network->bridge);
if (r < 0)
log_link_error_errno(link, r, "Failed to add to bridge master's slave list: %m");
}
if (link->network->bond) {
r = link_set_bond(link);
if (r < 0)
log_link_error_errno(link, r, "Could not set bond message: %m");
r = link_append_to_master(link, link->network->bond);
if (r < 0)
log_link_error_errno(link, r, "Failed to add to bond master's slave list: %m");
}
if (link->network->use_br_vlan &&
(link->network->bridge || streq_ptr("bridge", link->kind))) {
r = link_set_bridge_vlan(link);
if (r < 0)
log_link_error_errno(link, r, "Could not set bridge vlan: %m");
}
/* Skip setting up addresses until it gets carrier,
or it would try to set addresses twice,
which is bad for non-idempotent steps. */
if (!link_has_carrier(link) && !link->network->configure_without_carrier)
return 0;
link_set_state(link, LINK_STATE_CONFIGURING);
return link_request_set_addresses(link);
}
static int netdev_join_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
assert(link);
assert(link->network);
assert(link->enslaving > 0);
link->enslaving--;
if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
return 1;
r = sd_netlink_message_get_errno(m);
if (r < 0 && r != -EEXIST) {
log_link_error_errno(link, r, "Could not join netdev: %m");
link_enter_failed(link);
return 1;
}
log_link_debug(link, "Joined netdev");
if (link->enslaving == 0) {
r = link_joined(link);
if (r < 0)
link_enter_failed(link);
}
return 1;
}
static int link_enter_join_netdev(Link *link) {
NetDev *netdev;
Iterator i;
int r;
assert(link);
assert(link->network);
assert(link->state == LINK_STATE_INITIALIZED);
link_set_state(link, LINK_STATE_CONFIGURING);
link_dirty(link);
link->enslaving = 0;
if (link->network->bond) {
if (link->network->bond->state == NETDEV_STATE_READY &&
link->network->bond->ifindex == link->master_ifindex)
return link_joined(link);
log_struct(LOG_DEBUG,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(link->network->bond),
LOG_LINK_MESSAGE(link, "Enslaving by '%s'", link->network->bond->ifname));
link->enslaving++;
r = netdev_join(link->network->bond, link, netdev_join_handler);
if (r < 0) {
log_struct_errno(LOG_WARNING, r,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(link->network->bond),
LOG_LINK_MESSAGE(link, "Could not join netdev '%s': %m", link->network->bond->ifname));
link_enter_failed(link);
return r;
}
}
if (link->network->bridge) {
log_struct(LOG_DEBUG,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(link->network->bridge),
LOG_LINK_MESSAGE(link, "Enslaving by '%s'", link->network->bridge->ifname));
link->enslaving++;
r = netdev_join(link->network->bridge, link, netdev_join_handler);
if (r < 0) {
log_struct_errno(LOG_WARNING, r,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(link->network->bridge),
LOG_LINK_MESSAGE(link, "Could not join netdev '%s': %m", link->network->bridge->ifname));
link_enter_failed(link);
return r;
}
}
if (link->network->vrf) {
log_struct(LOG_DEBUG,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(link->network->vrf),
LOG_LINK_MESSAGE(link, "Enslaving by '%s'", link->network->vrf->ifname));
link->enslaving++;
r = netdev_join(link->network->vrf, link, netdev_join_handler);
if (r < 0) {
log_struct_errno(LOG_WARNING, r,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(link->network->vrf),
LOG_LINK_MESSAGE(link, "Could not join netdev '%s': %m", link->network->vrf->ifname));
link_enter_failed(link);
return r;
}
}
HASHMAP_FOREACH(netdev, link->network->stacked_netdevs, i) {
if (netdev->ifindex > 0)
/* Assume already enslaved. */
continue;
if (netdev_get_create_type(netdev) != NETDEV_CREATE_STACKED)
continue;
log_struct(LOG_DEBUG,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(netdev),
LOG_LINK_MESSAGE(link, "Enslaving by '%s'", netdev->ifname));
link->enslaving++;
r = netdev_join(netdev, link, netdev_join_handler);
if (r < 0) {
log_struct_errno(LOG_WARNING, r,
LOG_LINK_INTERFACE(link),
LOG_NETDEV_INTERFACE(netdev),
LOG_LINK_MESSAGE(link, "Could not join netdev '%s': %m", netdev->ifname));
link_enter_failed(link);
return r;
}
}
if (link->enslaving == 0)
return link_joined(link);
return 0;
}
static int link_set_ipv4_forward(Link *link) {
int r;
if (!link_ipv4_forward_enabled(link))
return 0;
/* We propagate the forwarding flag from one interface to the
* global setting one way. This means: as long as at least one
* interface was configured at any time that had IP forwarding
* enabled the setting will stay on for good. We do this
* primarily to keep IPv4 and IPv6 packet forwarding behaviour
* somewhat in sync (see below). */
r = sysctl_write_ip_property(AF_INET, NULL, "ip_forward", "1");
if (r < 0)
log_link_warning_errno(link, r, "Cannot turn on IPv4 packet forwarding, ignoring: %m");
return 0;
}
static int link_set_ipv6_forward(Link *link) {
int r;
if (!link_ipv6_forward_enabled(link))
return 0;
/* On Linux, the IPv6 stack does not know a per-interface
* packet forwarding setting: either packet forwarding is on
* for all, or off for all. We hence don't bother with a
* per-interface setting, but simply propagate the interface
* flag, if it is set, to the global flag, one-way. Note that
* while IPv4 would allow a per-interface flag, we expose the
* same behaviour there and also propagate the setting from
* one to all, to keep things simple (see above). */
r = sysctl_write_ip_property(AF_INET6, "all", "forwarding", "1");
if (r < 0)
log_link_warning_errno(link, r, "Cannot configure IPv6 packet forwarding, ignoring: %m");
return 0;
}
static int link_set_ipv6_privacy_extensions(Link *link) {
IPv6PrivacyExtensions s;
int r;
s = link_ipv6_privacy_extensions(link);
if (s < 0)
return 0;
r = sysctl_write_ip_property_int(AF_INET6, link->ifname, "use_tempaddr", (int) link->network->ipv6_privacy_extensions);
if (r < 0)
log_link_warning_errno(link, r, "Cannot configure IPv6 privacy extension for interface: %m");
return 0;
}
static int link_set_ipv6_accept_ra(Link *link) {
int r;
/* Make this a NOP if IPv6 is not available */
if (!socket_ipv6_is_supported())
return 0;
if (link->flags & IFF_LOOPBACK)
return 0;
if (!link->network)
return 0;
r = sysctl_write_ip_property(AF_INET6, link->ifname, "accept_ra", "0");
if (r < 0)
log_link_warning_errno(link, r, "Cannot disable kernel IPv6 accept_ra for interface: %m");
return 0;
}
static int link_set_ipv6_dad_transmits(Link *link) {
int r;
/* Make this a NOP if IPv6 is not available */
if (!socket_ipv6_is_supported())
return 0;
if (link->flags & IFF_LOOPBACK)
return 0;
if (!link->network)
return 0;
if (link->network->ipv6_dad_transmits < 0)
return 0;
r = sysctl_write_ip_property_int(AF_INET6, link->ifname, "dad_transmits", link->network->ipv6_dad_transmits);
if (r < 0)
log_link_warning_errno(link, r, "Cannot set IPv6 dad transmits for interface: %m");
return 0;
}
static int link_set_ipv6_hop_limit(Link *link) {
int r;
/* Make this a NOP if IPv6 is not available */
if (!socket_ipv6_is_supported())
return 0;
if (link->flags & IFF_LOOPBACK)
return 0;
if (!link->network)
return 0;
if (link->network->ipv6_hop_limit < 0)
return 0;
r = sysctl_write_ip_property_int(AF_INET6, link->ifname, "hop_limit", link->network->ipv6_hop_limit);
if (r < 0)
log_link_warning_errno(link, r, "Cannot set IPv6 hop limit for interface: %m");
return 0;
}
static int link_set_ipv6_mtu(Link *link) {
int r;
/* Make this a NOP if IPv6 is not available */
if (!socket_ipv6_is_supported())
return 0;
if (link->flags & IFF_LOOPBACK)
return 0;
if (link->network->ipv6_mtu == 0)
return 0;
r = sysctl_write_ip_property_uint32(AF_INET6, link->ifname, "mtu", link->network->ipv6_mtu);
if (r < 0)
log_link_warning_errno(link, r, "Cannot set IPv6 MTU for interface: %m");
return 0;
}
static bool link_is_static_address_configured(Link *link, Address *address) {
Address *net_address;
assert(link);
assert(address);
if (!link->network)
return false;
LIST_FOREACH(addresses, net_address, link->network->static_addresses)
if (address_equal(net_address, address))
return true;
return false;
}
static bool link_is_static_route_configured(Link *link, Route *route) {
Route *net_route;
assert(link);
assert(route);
if (!link->network)
return false;
LIST_FOREACH(routes, net_route, link->network->static_routes)
if (route_equal(net_route, route))
return true;
return false;
}
static bool link_address_is_dynamic(Link *link, Address *address) {
Route *route;
Iterator i;
assert(link);
assert(address);
if (address->cinfo.ifa_prefered != CACHE_INFO_INFINITY_LIFE_TIME)
return true;
/* Even when the address is leased from a DHCP server, networkd assign the address
* without lifetime when KeepConfiguration=dhcp. So, let's check that we have
* corresponding routes with RTPROT_DHCP. */
SET_FOREACH(route, link->routes_foreign, i) {
if (route->protocol != RTPROT_DHCP)
continue;
if (address->family != route->family)
continue;
if (in_addr_equal(address->family, &address->in_addr, &route->prefsrc))
return true;
}
return false;
}
static int link_drop_foreign_config(Link *link) {
Address *address;
Route *route;
Iterator i;
int r;
SET_FOREACH(address, link->addresses_foreign, i) {
/* we consider IPv6LL addresses to be managed by the kernel */
if (address->family == AF_INET6 && in_addr_is_link_local(AF_INET6, &address->in_addr) == 1)
continue;
if (link_address_is_dynamic(link, address)) {
if (FLAGS_SET(link->network->keep_configuration, KEEP_CONFIGURATION_DHCP))
continue;
} else if (FLAGS_SET(link->network->keep_configuration, KEEP_CONFIGURATION_STATIC))
continue;
if (link_is_static_address_configured(link, address)) {
r = address_add(link, address->family, &address->in_addr, address->prefixlen, NULL);
if (r < 0)
return log_link_error_errno(link, r, "Failed to add address: %m");
} else {
r = address_remove(address, link, NULL);
if (r < 0)
return r;
}
}
SET_FOREACH(route, link->routes_foreign, i) {
/* do not touch routes managed by the kernel */
if (route->protocol == RTPROT_KERNEL)
continue;
/* do not touch multicast route added by kernel */
/* FIXME: Why the kernel adds this route with protocol RTPROT_BOOT??? We need to investigate that.
* https://tools.ietf.org/html/rfc4862#section-5.4 may explain why. */
if (route->protocol == RTPROT_BOOT &&
route->family == AF_INET6 &&
route->dst_prefixlen == 8 &&
in_addr_equal(AF_INET6, &route->dst, &(union in_addr_union) { .in6 = {{{ 0xff,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0 }}} }))
continue;
if (route->protocol == RTPROT_STATIC &&
FLAGS_SET(link->network->keep_configuration, KEEP_CONFIGURATION_STATIC))
continue;
if (route->protocol == RTPROT_DHCP &&
FLAGS_SET(link->network->keep_configuration, KEEP_CONFIGURATION_DHCP))
continue;
if (link_is_static_route_configured(link, route)) {
r = route_add(link, route->family, &route->dst, route->dst_prefixlen, route->tos, route->priority, route->table, NULL);
if (r < 0)
return r;
} else {
r = route_remove(route, link, NULL);
if (r < 0)
return r;
}
}
return 0;
}
static int link_drop_config(Link *link) {
Address *address, *pool_address;
Route *route;
Iterator i;
int r;
SET_FOREACH(address, link->addresses, i) {
/* we consider IPv6LL addresses to be managed by the kernel */
if (address->family == AF_INET6 && in_addr_is_link_local(AF_INET6, &address->in_addr) == 1)
continue;
r = address_remove(address, link, NULL);
if (r < 0)
return r;
/* If this address came from an address pool, clean up the pool */
LIST_FOREACH(addresses, pool_address, link->pool_addresses) {
if (address_equal(address, pool_address)) {
LIST_REMOVE(addresses, link->pool_addresses, pool_address);
address_free(pool_address);
break;
}
}
}
SET_FOREACH(route, link->routes, i) {
/* do not touch routes managed by the kernel */
if (route->protocol == RTPROT_KERNEL)
continue;
r = route_remove(route, link, NULL);
if (r < 0)
return r;
}
ndisc_flush(link);
return 0;
}
static int link_configure(Link *link) {
int r;
assert(link);
assert(link->network);
assert(link->state == LINK_STATE_INITIALIZED);
if (link->iftype == ARPHRD_CAN)
return link_configure_can(link);
/* Drop foreign config, but ignore loopback or critical devices.
* We do not want to remove loopback address or addresses used for root NFS. */
if (!(link->flags & IFF_LOOPBACK) &&
link->network->keep_configuration != KEEP_CONFIGURATION_YES) {
r = link_drop_foreign_config(link);
if (r < 0)
return r;
}
r = link_set_proxy_arp(link);
if (r < 0)
return r;
r = ipv6_proxy_ndp_addresses_configure(link);
if (r < 0)
return r;
r = link_set_ipv4_forward(link);
if (r < 0)
return r;
r = link_set_ipv6_forward(link);
if (r < 0)
return r;
r = link_set_ipv6_privacy_extensions(link);
if (r < 0)
return r;
r = link_set_ipv6_accept_ra(link);
if (r < 0)
return r;
r = link_set_ipv6_dad_transmits(link);
if (r < 0)
return r;
r = link_set_ipv6_hop_limit(link);
if (r < 0)
return r;
r = link_set_flags(link);
if (r < 0)
return r;
r = link_set_ipv6_mtu(link);
if (r < 0)
return r;
if (link_ipv4ll_enabled(link, ADDRESS_FAMILY_IPV4 | ADDRESS_FAMILY_FALLBACK_IPV4)) {
r = ipv4ll_configure(link);
if (r < 0)
return r;
}
if (link_dhcp4_enabled(link)) {
r = dhcp4_set_promote_secondaries(link);
if (r < 0)
return r;
r = dhcp4_configure(link);
if (r < 0)
return r;
}
if (link_dhcp4_server_enabled(link)) {
r = sd_dhcp_server_new(&link->dhcp_server, link->ifindex);
if (r < 0)
return r;
r = sd_dhcp_server_attach_event(link->dhcp_server, NULL, 0);
if (r < 0)
return r;
}
if (link_dhcp6_enabled(link) ||
link_ipv6_accept_ra_enabled(link)) {
r = dhcp6_configure(link);
if (r < 0)
return r;
}
if (link_ipv6_accept_ra_enabled(link)) {
r = ndisc_configure(link);
if (r < 0)
return r;
}
if (link_radv_enabled(link)) {
r = radv_configure(link);
if (r < 0)
return r;
}
if (link_lldp_rx_enabled(link)) {
r = link_lldp_rx_configure(link);
if (r < 0)
return r;
}
r = link_configure_mtu(link);
if (r < 0)
return r;
r = link_configure_addrgen_mode(link);
if (r < 0)
return r;
return link_configure_after_setting_mtu(link);
}
static int link_configure_after_setting_mtu(Link *link) {
int r;
assert(link);
assert(link->network);
assert(link->state == LINK_STATE_INITIALIZED);
if (link->setting_mtu)
return 0;
if (link_has_carrier(link) || link->network->configure_without_carrier) {
r = link_acquire_conf(link);
if (r < 0)
return r;
}
return link_enter_join_netdev(link);
}
static int duid_set_uuid(DUID *duid, sd_id128_t uuid) {
assert(duid);
if (duid->raw_data_len > 0)
return 0;
if (duid->type != DUID_TYPE_UUID)
return -EINVAL;
memcpy(&duid->raw_data, &uuid, sizeof(sd_id128_t));
duid->raw_data_len = sizeof(sd_id128_t);
return 1;
}
int get_product_uuid_handler(sd_bus_message *m, void *userdata, sd_bus_error *ret_error) {
Manager *manager = userdata;
const sd_bus_error *e;
const void *a;
size_t sz;
DUID *duid;
Link *link;
int r;
assert(m);
assert(manager);
e = sd_bus_message_get_error(m);
if (e) {
log_error_errno(sd_bus_error_get_errno(e),
"Could not get product UUID. Falling back to use machine-app-specific ID as DUID-UUID: %s",
e->message);
goto configure;
}
r = sd_bus_message_read_array(m, 'y', &a, &sz);
if (r < 0)
goto configure;
if (sz != sizeof(sd_id128_t)) {
log_error("Invalid product UUID. Falling back to use machine-app-specific ID as DUID-UUID.");
goto configure;
}
memcpy(&manager->product_uuid, a, sz);
while ((duid = set_steal_first(manager->duids_requesting_uuid)))
(void) duid_set_uuid(duid, manager->product_uuid);
manager->duids_requesting_uuid = set_free(manager->duids_requesting_uuid);
configure:
while ((link = set_steal_first(manager->links_requesting_uuid))) {
r = link_configure(link);
if (r < 0)
link_enter_failed(link);
}
manager->links_requesting_uuid = set_free(manager->links_requesting_uuid);
/* To avoid calling GetProductUUID() bus method so frequently, set the flag below
* even if the method fails. */
manager->has_product_uuid = true;
return 1;
}
static bool link_requires_uuid(Link *link) {
const DUID *duid;
assert(link);
assert(link->manager);
assert(link->network);
duid = link_get_duid(link);
if (duid->type != DUID_TYPE_UUID || duid->raw_data_len != 0)
return false;
if (link_dhcp4_enabled(link) && IN_SET(link->network->dhcp_client_identifier, DHCP_CLIENT_ID_DUID, DHCP_CLIENT_ID_DUID_ONLY))
return true;
if (link_dhcp6_enabled(link) || link_ipv6_accept_ra_enabled(link))
return true;
return false;
}
static int link_configure_duid(Link *link) {
Manager *m;
DUID *duid;
int r;
assert(link);
assert(link->manager);
assert(link->network);
m = link->manager;
duid = link_get_duid(link);
if (!link_requires_uuid(link))
return 1;
if (m->has_product_uuid) {
(void) duid_set_uuid(duid, m->product_uuid);
return 1;
}
if (!m->links_requesting_uuid) {
r = manager_request_product_uuid(m, link);
if (r < 0) {
if (r == -ENOMEM)
return r;
log_link_warning_errno(link, r,
"Failed to get product UUID. Falling back to use machine-app-specific ID as DUID-UUID: %m");
return 1;
}
} else {
r = set_put(m->links_requesting_uuid, link);
if (r < 0)
return log_oom();
r = set_put(m->duids_requesting_uuid, duid);
if (r < 0)
return log_oom();
}
return 0;
}
static int link_initialized_and_synced(Link *link) {
Network *network;
int r;
assert(link);
assert(link->ifname);
assert(link->manager);
/* We may get called either from the asynchronous netlink callback,
* or directly for link_add() if running in a container. See link_add(). */
if (!IN_SET(link->state, LINK_STATE_PENDING, LINK_STATE_INITIALIZED))
return 0;
log_link_debug(link, "Link state is up-to-date");
link_set_state(link, LINK_STATE_INITIALIZED);
r = link_new_bound_by_list(link);
if (r < 0)
return r;
r = link_handle_bound_by_list(link);
if (r < 0)
return r;
if (!link->network) {
r = network_get(link->manager, link->sd_device, link->ifname,
&link->mac, &network);
if (r == -ENOENT) {
link_enter_unmanaged(link);
return 0;
} else if (r == 0 && network->unmanaged) {
link_enter_unmanaged(link);
return 0;
} else if (r < 0)
return r;
if (link->flags & IFF_LOOPBACK) {
if (network->link_local != ADDRESS_FAMILY_NO)
log_link_debug(link, "Ignoring link-local autoconfiguration for loopback link");
if (network->dhcp != ADDRESS_FAMILY_NO)
log_link_debug(link, "Ignoring DHCP clients for loopback link");
if (network->dhcp_server)
log_link_debug(link, "Ignoring DHCP server for loopback link");
}
r = network_apply(network, link);
if (r < 0)
return r;
}
r = link_new_bound_to_list(link);
if (r < 0)
return r;
/* link_configure_duid() returns 0 if it requests product UUID. In that case,
* link_configure() is called later asynchronously. */
r = link_configure_duid(link);
if (r <= 0)
return r;
r = link_configure(link);
if (r < 0)
return r;
return 0;
}
static int link_initialized_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
int r;
r = link_initialized_and_synced(link);
if (r < 0)
link_enter_failed(link);
return 1;
}
int link_initialized(Link *link, sd_device *device) {
_cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
int r;
assert(link);
assert(link->manager);
assert(link->manager->rtnl);
assert(device);
if (link->state != LINK_STATE_PENDING)
return 0;
if (link->sd_device)
return 0;
log_link_debug(link, "udev initialized link");
link_set_state(link, LINK_STATE_INITIALIZED);
link->sd_device = sd_device_ref(device);
/* udev has initialized the link, but we don't know if we have yet
* processed the NEWLINK messages with the latest state. Do a GETLINK,
* when it returns we know that the pending NEWLINKs have already been
* processed and that we are up-to-date */
r = sd_rtnl_message_new_link(link->manager->rtnl, &req, RTM_GETLINK,
link->ifindex);
if (r < 0)
return r;
r = netlink_call_async(link->manager->rtnl, NULL, req, link_initialized_handler,
link_netlink_destroy_callback, link);
if (r < 0)
return r;
link_ref(link);
return 0;
}
static int link_load(Link *link) {
_cleanup_free_ char *network_file = NULL,
*addresses = NULL,
*routes = NULL,
*dhcp4_address = NULL,
*ipv4ll_address = NULL;
union in_addr_union address;
union in_addr_union route_dst;
const char *p;
int r;
assert(link);
r = parse_env_file(NULL, link->state_file,
"NETWORK_FILE", &network_file,
"ADDRESSES", &addresses,
"ROUTES", &routes,
"DHCP4_ADDRESS", &dhcp4_address,
"IPV4LL_ADDRESS", &ipv4ll_address);
if (r < 0 && r != -ENOENT)
return log_link_error_errno(link, r, "Failed to read %s: %m", link->state_file);
if (network_file) {
Network *network;
char *suffix;
/* drop suffix */
suffix = strrchr(network_file, '.');
if (!suffix) {
log_link_debug(link, "Failed to get network name from %s", network_file);
goto network_file_fail;
}
*suffix = '\0';
r = network_get_by_name(link->manager, basename(network_file), &network);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to get network %s: %m", basename(network_file));
goto network_file_fail;
}
r = network_apply(network, link);
if (r < 0)
return log_link_error_errno(link, r, "Failed to apply network %s: %m", basename(network_file));
}
network_file_fail:
if (addresses) {
p = addresses;
for (;;) {
_cleanup_free_ char *address_str = NULL;
char *prefixlen_str;
int family;
unsigned char prefixlen;
r = extract_first_word(&p, &address_str, NULL, 0);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to extract next address string: %m");
continue;
}
if (r == 0)
break;
prefixlen_str = strchr(address_str, '/');
if (!prefixlen_str) {
log_link_debug(link, "Failed to parse address and prefix length %s", address_str);
continue;
}
*prefixlen_str++ = '\0';
r = sscanf(prefixlen_str, "%hhu", &prefixlen);
if (r != 1) {
log_link_error(link, "Failed to parse prefixlen %s", prefixlen_str);
continue;
}
r = in_addr_from_string_auto(address_str, &family, &address);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to parse address %s: %m", address_str);
continue;
}
r = address_add(link, family, &address, prefixlen, NULL);
if (r < 0)
return log_link_error_errno(link, r, "Failed to add address: %m");
}
}
if (routes) {
p = routes;
for (;;) {
Route *route;
_cleanup_free_ char *route_str = NULL;
_cleanup_(sd_event_source_unrefp) sd_event_source *expire = NULL;
usec_t lifetime;
char *prefixlen_str;
int family;
unsigned char prefixlen, tos, table;
uint32_t priority;
r = extract_first_word(&p, &route_str, NULL, 0);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to extract next route string: %m");
continue;
}
if (r == 0)
break;
prefixlen_str = strchr(route_str, '/');
if (!prefixlen_str) {
log_link_debug(link, "Failed to parse route %s", route_str);
continue;
}
*prefixlen_str++ = '\0';
r = sscanf(prefixlen_str, "%hhu/%hhu/%"SCNu32"/%hhu/"USEC_FMT, &prefixlen, &tos, &priority, &table, &lifetime);
if (r != 5) {
log_link_debug(link,
"Failed to parse destination prefix length, tos, priority, table or expiration %s",
prefixlen_str);
continue;
}
r = in_addr_from_string_auto(route_str, &family, &route_dst);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to parse route destination %s: %m", route_str);
continue;
}
r = route_add(link, family, &route_dst, prefixlen, tos, priority, table, &route);
if (r < 0)
return log_link_error_errno(link, r, "Failed to add route: %m");
if (lifetime != USEC_INFINITY && !kernel_route_expiration_supported()) {
r = sd_event_add_time(link->manager->event, &expire, clock_boottime_or_monotonic(), lifetime,
0, route_expire_handler, route);
if (r < 0)
log_link_warning_errno(link, r, "Could not arm route expiration handler: %m");
}
route->lifetime = lifetime;
sd_event_source_unref(route->expire);
route->expire = TAKE_PTR(expire);
}
}
if (dhcp4_address) {
r = in_addr_from_string(AF_INET, dhcp4_address, &address);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to parse DHCPv4 address %s: %m", dhcp4_address);
goto dhcp4_address_fail;
}
r = sd_dhcp_client_new(&link->dhcp_client, link->network ? link->network->dhcp_anonymize : 0);
if (r < 0)
return log_link_error_errno(link, r, "Failed to create DHCPv4 client: %m");
r = sd_dhcp_client_set_request_address(link->dhcp_client, &address.in);
if (r < 0)
return log_link_error_errno(link, r, "Failed to set initial DHCPv4 address %s: %m", dhcp4_address);
}
dhcp4_address_fail:
if (ipv4ll_address) {
r = in_addr_from_string(AF_INET, ipv4ll_address, &address);
if (r < 0) {
log_link_debug_errno(link, r, "Failed to parse IPv4LL address %s: %m", ipv4ll_address);
goto ipv4ll_address_fail;
}
r = sd_ipv4ll_new(&link->ipv4ll);
if (r < 0)
return log_link_error_errno(link, r, "Failed to create IPv4LL client: %m");
r = sd_ipv4ll_set_address(link->ipv4ll, &address.in);
if (r < 0)
return log_link_error_errno(link, r, "Failed to set initial IPv4LL address %s: %m", ipv4ll_address);
}
ipv4ll_address_fail:
return 0;
}
int link_add(Manager *m, sd_netlink_message *message, Link **ret) {
_cleanup_(sd_device_unrefp) sd_device *device = NULL;
char ifindex_str[2 + DECIMAL_STR_MAX(int)];
Link *link;
int r;
assert(m);
assert(m->rtnl);
assert(message);
assert(ret);
r = link_new(m, message, ret);
if (r < 0)
return r;
link = *ret;
log_link_debug(link, "Link %d added", link->ifindex);
r = link_load(link);
if (r < 0)
return r;
if (detect_container() <= 0) {
/* not in a container, udev will be around */
sprintf(ifindex_str, "n%d", link->ifindex);
r = sd_device_new_from_device_id(&device, ifindex_str);
if (r < 0) {
log_link_warning_errno(link, r, "Could not find device: %m");
goto failed;
}
r = sd_device_get_is_initialized(device);
if (r < 0) {
log_link_warning_errno(link, r, "Could not determine whether the device is initialized or not: %m");
goto failed;
}
if (r == 0) {
/* not yet ready */
log_link_debug(link, "link pending udev initialization...");
return 0;
}
r = device_is_renaming(device);
if (r < 0) {
log_link_warning_errno(link, r, "Failed to determine the device is renamed or not: %m");
goto failed;
}
if (r > 0) {
log_link_debug(link, "Interface is under renaming, pending initialization.");
return 0;
}
r = link_initialized(link, device);
if (r < 0)
goto failed;
} else {
r = link_initialized_and_synced(link);
if (r < 0)
goto failed;
}
return 0;
failed:
link_enter_failed(link);
return r;
}
int link_ipv6ll_gained(Link *link, const struct in6_addr *address) {
int r;
assert(link);
log_link_info(link, "Gained IPv6LL");
link->ipv6ll_address = *address;
link_check_ready(link);
if (IN_SET(link->state, LINK_STATE_CONFIGURING, LINK_STATE_CONFIGURED)) {
r = link_acquire_ipv6_conf(link);
if (r < 0) {
link_enter_failed(link);
return r;
}
}
return 0;
}
static int link_carrier_gained(Link *link) {
int r;
assert(link);
if (IN_SET(link->state, LINK_STATE_CONFIGURING, LINK_STATE_CONFIGURED)) {
r = link_acquire_conf(link);
if (r < 0) {
link_enter_failed(link);
return r;
}
link_set_state(link, LINK_STATE_CONFIGURING);
r = link_request_set_addresses(link);
if (r < 0)
return r;
}
r = link_handle_bound_by_list(link);
if (r < 0)
return r;
return 0;
}
static int link_carrier_lost(Link *link) {
int r;
assert(link);
if (link->network && link->network->ignore_carrier_loss)
return 0;
/* Some devices reset itself while setting the MTU. This causes the DHCP client fall into a loop.
* setting_mtu keep track whether the device got reset because of setting MTU and does not drop the
* configuration and stop the clients as well. */
if (link->setting_mtu)
return 0;
r = link_stop_clients(link, false);
if (r < 0) {
link_enter_failed(link);
return r;
}
if (link_dhcp4_server_enabled(link))
(void) sd_dhcp_server_stop(link->dhcp_server);
r = link_drop_config(link);
if (r < 0)
return r;
if (!IN_SET(link->state, LINK_STATE_UNMANAGED, LINK_STATE_PENDING)) {
log_link_debug(link, "State is %s, dropping config", link_state_to_string(link->state));
r = link_drop_foreign_config(link);
if (r < 0)
return r;
}
r = link_handle_bound_by_list(link);
if (r < 0)
return r;
return 0;
}
int link_carrier_reset(Link *link) {
int r;
assert(link);
if (link_has_carrier(link)) {
r = link_carrier_lost(link);
if (r < 0)
return r;
r = link_carrier_gained(link);
if (r < 0)
return r;
log_link_info(link, "Reset carrier");
}
return 0;
}
int link_update(Link *link, sd_netlink_message *m) {
struct ether_addr mac;
const char *ifname;
uint32_t mtu;
bool had_carrier, carrier_gained, carrier_lost;
int old_master, r;
assert(link);
assert(link->ifname);
assert(m);
if (link->state == LINK_STATE_LINGER) {
log_link_info(link, "Link re-added");
link_set_state(link, LINK_STATE_CONFIGURING);
r = link_new_carrier_maps(link);
if (r < 0)
return r;
}
r = sd_netlink_message_read_string(m, IFLA_IFNAME, &ifname);
if (r >= 0 && !streq(ifname, link->ifname)) {
Manager *manager = link->manager;
log_link_info(link, "Interface name change detected, %s has been renamed to %s.", link->ifname, ifname);
link_drop(link);
r = link_add(manager, m, &link);
if (r < 0)
return r;
}
r = sd_netlink_message_read_u32(m, IFLA_MTU, &mtu);
if (r >= 0 && mtu > 0) {
link->mtu = mtu;
if (link->original_mtu == 0) {
link->original_mtu = mtu;
log_link_debug(link, "Saved original MTU: %" PRIu32, link->original_mtu);
}
if (link->dhcp_client) {
r = sd_dhcp_client_set_mtu(link->dhcp_client,
link->mtu);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update MTU in DHCP client: %m");
}
if (link->radv) {
r = sd_radv_set_mtu(link->radv, link->mtu);
if (r < 0)
return log_link_warning_errno(link, r, "Could not set MTU for Router Advertisement: %m");
}
}
/* The kernel may broadcast NEWLINK messages without the MAC address
set, simply ignore them. */
r = sd_netlink_message_read_ether_addr(m, IFLA_ADDRESS, &mac);
if (r >= 0) {
if (memcmp(link->mac.ether_addr_octet, mac.ether_addr_octet,
ETH_ALEN)) {
memcpy(link->mac.ether_addr_octet, mac.ether_addr_octet,
ETH_ALEN);
log_link_debug(link, "MAC address: "
"%02hhx:%02hhx:%02hhx:%02hhx:%02hhx:%02hhx",
mac.ether_addr_octet[0],
mac.ether_addr_octet[1],
mac.ether_addr_octet[2],
mac.ether_addr_octet[3],
mac.ether_addr_octet[4],
mac.ether_addr_octet[5]);
if (link->ipv4ll) {
r = sd_ipv4ll_set_mac(link->ipv4ll, &link->mac);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update MAC address in IPv4LL client: %m");
}
if (link->dhcp_client) {
r = sd_dhcp_client_set_mac(link->dhcp_client,
(const uint8_t *) &link->mac,
sizeof (link->mac),
ARPHRD_ETHER);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update MAC address in DHCP client: %m");
r = dhcp4_set_client_identifier(link);
if (r < 0)
return r;
}
if (link->dhcp6_client) {
const DUID* duid = link_get_duid(link);
r = sd_dhcp6_client_set_mac(link->dhcp6_client,
(const uint8_t *) &link->mac,
sizeof (link->mac),
ARPHRD_ETHER);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update MAC address in DHCPv6 client: %m");
if (link->network->iaid_set) {
r = sd_dhcp6_client_set_iaid(link->dhcp6_client,
link->network->iaid);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update DHCPv6 IAID: %m");
}
r = sd_dhcp6_client_set_duid(link->dhcp6_client,
duid->type,
duid->raw_data_len > 0 ? duid->raw_data : NULL,
duid->raw_data_len);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update DHCPv6 DUID: %m");
}
if (link->radv) {
r = sd_radv_set_mac(link->radv, &link->mac);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update MAC for Router Advertisement: %m");
}
if (link->ndisc) {
r = sd_ndisc_set_mac(link->ndisc, &link->mac);
if (r < 0)
return log_link_warning_errno(link, r, "Could not update MAC for ndisc: %m");
}
}
}
old_master = link->master_ifindex;
(void) sd_netlink_message_read_u32(m, IFLA_MASTER, (uint32_t *) &link->master_ifindex);
had_carrier = link_has_carrier(link);
r = link_update_flags(link, m, old_master != link->master_ifindex);
if (r < 0)
return r;
r = link_update_lldp(link);
if (r < 0)
return r;
carrier_gained = !had_carrier && link_has_carrier(link);
carrier_lost = had_carrier && !link_has_carrier(link);
if (carrier_gained) {
log_link_info(link, "Gained carrier");
r = link_carrier_gained(link);
if (r < 0)
return r;
} else if (carrier_lost) {
log_link_info(link, "Lost carrier");
r = link_carrier_lost(link);
if (r < 0)
return r;
}
return 0;
}
static void print_link_hashmap(FILE *f, const char *prefix, Hashmap* h) {
bool space = false;
Iterator i;
Link *link;
assert(f);
assert(prefix);
if (hashmap_isempty(h))
return;
fputs(prefix, f);
HASHMAP_FOREACH(link, h, i) {
if (space)
fputc(' ', f);
fprintf(f, "%i", link->ifindex);
space = true;
}
fputc('\n', f);
}
static void link_save_dns(FILE *f, struct in_addr_data *dns, unsigned n_dns, bool *space) {
unsigned j;
int r;
for (j = 0; j < n_dns; j++) {
_cleanup_free_ char *b = NULL;
r = in_addr_to_string(dns[j].family, &dns[j].address, &b);
if (r < 0) {
log_debug_errno(r, "Failed to format address, ignoring: %m");
continue;
}
if (*space)
fputc(' ', f);
fputs(b, f);
*space = true;
}
}
int link_save(Link *link) {
_cleanup_free_ char *temp_path = NULL;
_cleanup_fclose_ FILE *f = NULL;
const char *admin_state, *oper_state, *carrier_state, *address_state;
Address *a;
Route *route;
Iterator i;
int r;
assert(link);
assert(link->state_file);
assert(link->lease_file);
assert(link->manager);
if (link->state == LINK_STATE_LINGER) {
(void) unlink(link->state_file);
return 0;
}
link_lldp_save(link);
admin_state = link_state_to_string(link->state);
assert(admin_state);
oper_state = link_operstate_to_string(link->operstate);
assert(oper_state);
carrier_state = link_carrier_state_to_string(link->carrier_state);
assert(carrier_state);
address_state = link_address_state_to_string(link->address_state);
assert(address_state);
r = fopen_temporary(link->state_file, &f, &temp_path);
if (r < 0)
goto fail;
(void) fchmod(fileno(f), 0644);
fprintf(f,
"# This is private data. Do not parse.\n"
"ADMIN_STATE=%s\n"
"OPER_STATE=%s\n"
"CARRIER_STATE=%s\n"
"ADDRESS_STATE=%s\n",
admin_state, oper_state, carrier_state, address_state);
if (link->network) {
char **dhcp6_domains = NULL, **dhcp_domains = NULL;
const char *dhcp_domainname = NULL, *p;
sd_dhcp6_lease *dhcp6_lease = NULL;
bool space;
fprintf(f, "REQUIRED_FOR_ONLINE=%s\n",
yes_no(link->network->required_for_online));
fprintf(f, "REQUIRED_OPER_STATE_FOR_ONLINE=%s\n",
strempty(link_operstate_to_string(link->network->required_operstate_for_online)));
if (link->dhcp6_client) {
r = sd_dhcp6_client_get_lease(link->dhcp6_client, &dhcp6_lease);
if (r < 0 && r != -ENOMSG)
log_link_debug(link, "No DHCPv6 lease");
}
fprintf(f, "NETWORK_FILE=%s\n", link->network->filename);
fputs("DNS=", f);
space = false;
if (link->n_dns != (unsigned) -1)
link_save_dns(f, link->dns, link->n_dns, &space);
else
link_save_dns(f, link->network->dns, link->network->n_dns, &space);
if (link->network->dhcp_use_dns &&
link->dhcp_lease) {
const struct in_addr *addresses;
r = sd_dhcp_lease_get_dns(link->dhcp_lease, &addresses);
if (r > 0)
if (serialize_in_addrs(f, addresses, r, space, in4_addr_is_non_local) > 0)
space = true;
}
if (link->network->dhcp6_use_dns && dhcp6_lease) {
struct in6_addr *in6_addrs;
r = sd_dhcp6_lease_get_dns(dhcp6_lease, &in6_addrs);
if (r > 0) {
if (space)
fputc(' ', f);
serialize_in6_addrs(f, in6_addrs, r);
space = true;
}
}
/* Make sure to flush out old entries before we use the NDISC data */
ndisc_vacuum(link);
if (link->network->ipv6_accept_ra_use_dns && link->ndisc_rdnss) {
NDiscRDNSS *dd;
SET_FOREACH(dd, link->ndisc_rdnss, i) {
if (space)
fputc(' ', f);
serialize_in6_addrs(f, &dd->address, 1);
space = true;
}
}
fputc('\n', f);
fputs("NTP=", f);
space = false;
fputstrv(f, link->ntp ?: link->network->ntp, NULL, &space);
if (link->network->dhcp_use_ntp &&
link->dhcp_lease) {
const struct in_addr *addresses;
r = sd_dhcp_lease_get_ntp(link->dhcp_lease, &addresses);
if (r > 0)
if (serialize_in_addrs(f, addresses, r, space, in4_addr_is_non_local) > 0)
space = true;
}
if (link->network->dhcp6_use_ntp && dhcp6_lease) {
struct in6_addr *in6_addrs;
char **hosts;
r = sd_dhcp6_lease_get_ntp_addrs(dhcp6_lease,
&in6_addrs);
if (r > 0) {
if (space)
fputc(' ', f);
serialize_in6_addrs(f, in6_addrs, r);
space = true;
}
r = sd_dhcp6_lease_get_ntp_fqdn(dhcp6_lease, &hosts);
if (r > 0)
fputstrv(f, hosts, NULL, &space);
}
fputc('\n', f);
if (link->network->dhcp_use_domains != DHCP_USE_DOMAINS_NO) {
if (link->dhcp_lease) {
(void) sd_dhcp_lease_get_domainname(link->dhcp_lease, &dhcp_domainname);
(void) sd_dhcp_lease_get_search_domains(link->dhcp_lease, &dhcp_domains);
}
if (dhcp6_lease)
(void) sd_dhcp6_lease_get_domains(dhcp6_lease, &dhcp6_domains);
}
fputs("DOMAINS=", f);
space = false;
ORDERED_SET_FOREACH(p, link->search_domains ?: link->network->search_domains, i)
fputs_with_space(f, p, NULL, &space);
if (link->network->dhcp_use_domains == DHCP_USE_DOMAINS_YES) {
if (dhcp_domainname)
fputs_with_space(f, dhcp_domainname, NULL, &space);
if (dhcp_domains)
fputstrv(f, dhcp_domains, NULL, &space);
if (dhcp6_domains)
fputstrv(f, dhcp6_domains, NULL, &space);
}
if (link->network->ipv6_accept_ra_use_domains == DHCP_USE_DOMAINS_YES) {
NDiscDNSSL *dd;
SET_FOREACH(dd, link->ndisc_dnssl, i)
fputs_with_space(f, NDISC_DNSSL_DOMAIN(dd), NULL, &space);
}
fputc('\n', f);
fputs("ROUTE_DOMAINS=", f);
space = false;
ORDERED_SET_FOREACH(p, link->route_domains ?: link->network->route_domains, i)
fputs_with_space(f, p, NULL, &space);
if (link->network->dhcp_use_domains == DHCP_USE_DOMAINS_ROUTE) {
if (dhcp_domainname)
fputs_with_space(f, dhcp_domainname, NULL, &space);
if (dhcp_domains)
fputstrv(f, dhcp_domains, NULL, &space);
if (dhcp6_domains)
fputstrv(f, dhcp6_domains, NULL, &space);
}
if (link->network->ipv6_accept_ra_use_domains == DHCP_USE_DOMAINS_ROUTE) {
NDiscDNSSL *dd;
SET_FOREACH(dd, link->ndisc_dnssl, i)
fputs_with_space(f, NDISC_DNSSL_DOMAIN(dd), NULL, &space);
}
fputc('\n', f);
fprintf(f, "LLMNR=%s\n",
resolve_support_to_string(link->llmnr >= 0 ? link->llmnr : link->network->llmnr));
fprintf(f, "MDNS=%s\n",
resolve_support_to_string(link->mdns >= 0 ? link->mdns : link->network->mdns));
if (link->dns_default_route >= 0)
fprintf(f, "DNS_DEFAULT_ROUTE=%s\n", yes_no(link->dns_default_route));
else if (link->network->dns_default_route >= 0)
fprintf(f, "DNS_DEFAULT_ROUTE=%s\n", yes_no(link->network->dns_default_route));
if (link->dns_over_tls_mode != _DNS_OVER_TLS_MODE_INVALID)
fprintf(f, "DNS_OVER_TLS=%s\n",
dns_over_tls_mode_to_string(link->dns_over_tls_mode));
else if (link->network->dns_over_tls_mode != _DNS_OVER_TLS_MODE_INVALID)
fprintf(f, "DNS_OVER_TLS=%s\n",
dns_over_tls_mode_to_string(link->network->dns_over_tls_mode));
if (link->dnssec_mode != _DNSSEC_MODE_INVALID)
fprintf(f, "DNSSEC=%s\n",
dnssec_mode_to_string(link->dnssec_mode));
else if (link->network->dnssec_mode != _DNSSEC_MODE_INVALID)
fprintf(f, "DNSSEC=%s\n",
dnssec_mode_to_string(link->network->dnssec_mode));
if (!set_isempty(link->dnssec_negative_trust_anchors)) {
const char *n;
fputs("DNSSEC_NTA=", f);
space = false;
SET_FOREACH(n, link->dnssec_negative_trust_anchors, i)
fputs_with_space(f, n, NULL, &space);
fputc('\n', f);
} else if (!set_isempty(link->network->dnssec_negative_trust_anchors)) {
const char *n;
fputs("DNSSEC_NTA=", f);
space = false;
SET_FOREACH(n, link->network->dnssec_negative_trust_anchors, i)
fputs_with_space(f, n, NULL, &space);
fputc('\n', f);
}
fputs("ADDRESSES=", f);
space = false;
SET_FOREACH(a, link->addresses, i) {
_cleanup_free_ char *address_str = NULL;
r = in_addr_to_string(a->family, &a->in_addr, &address_str);
if (r < 0)
goto fail;
fprintf(f, "%s%s/%u", space ? " " : "", address_str, a->prefixlen);
space = true;
}
fputc('\n', f);
fputs("ROUTES=", f);
space = false;
SET_FOREACH(route, link->routes, i) {
_cleanup_free_ char *route_str = NULL;
r = in_addr_to_string(route->family, &route->dst, &route_str);
if (r < 0)
goto fail;
fprintf(f, "%s%s/%hhu/%hhu/%"PRIu32"/%"PRIu32"/"USEC_FMT,
space ? " " : "", route_str,
route->dst_prefixlen, route->tos, route->priority, route->table, route->lifetime);
space = true;
}
fputc('\n', f);
}
print_link_hashmap(f, "CARRIER_BOUND_TO=", link->bound_to_links);
print_link_hashmap(f, "CARRIER_BOUND_BY=", link->bound_by_links);
if (link->dhcp_lease) {
struct in_addr address;
const char *tz = NULL;
assert(link->network);
r = sd_dhcp_lease_get_timezone(link->dhcp_lease, &tz);
if (r >= 0)
fprintf(f, "TIMEZONE=%s\n", tz);
r = sd_dhcp_lease_get_address(link->dhcp_lease, &address);
if (r >= 0) {
fputs("DHCP4_ADDRESS=", f);
serialize_in_addrs(f, &address, 1, false, NULL);
fputc('\n', f);
}
r = dhcp_lease_save(link->dhcp_lease, link->lease_file);
if (r < 0)
goto fail;
fprintf(f,
"DHCP_LEASE=%s\n",
link->lease_file);
} else
(void) unlink(link->lease_file);
if (link->ipv4ll) {
struct in_addr address;
r = sd_ipv4ll_get_address(link->ipv4ll, &address);
if (r >= 0) {
fputs("IPV4LL_ADDRESS=", f);
serialize_in_addrs(f, &address, 1, false, NULL);
fputc('\n', f);
}
}
r = fflush_and_check(f);
if (r < 0)
goto fail;
if (rename(temp_path, link->state_file) < 0) {
r = -errno;
goto fail;
}
return 0;
fail:
(void) unlink(link->state_file);
if (temp_path)
(void) unlink(temp_path);
return log_link_error_errno(link, r, "Failed to save link data to %s: %m", link->state_file);
}
/* The serialized state in /run is no longer up-to-date. */
void link_dirty(Link *link) {
int r;
assert(link);
/* mark manager dirty as link is dirty */
manager_dirty(link->manager);
r = set_ensure_allocated(&link->manager->dirty_links, NULL);
if (r < 0)
/* allocation errors are ignored */
return;
r = set_put(link->manager->dirty_links, link);
if (r <= 0)
/* don't take another ref if the link was already dirty */
return;
link_ref(link);
}
/* The serialized state in /run is up-to-date */
void link_clean(Link *link) {
assert(link);
assert(link->manager);
link_unref(set_remove(link->manager->dirty_links, link));
}
static const char* const link_state_table[_LINK_STATE_MAX] = {
[LINK_STATE_PENDING] = "pending",
[LINK_STATE_INITIALIZED] = "initialized",
[LINK_STATE_CONFIGURING] = "configuring",
[LINK_STATE_CONFIGURED] = "configured",
[LINK_STATE_UNMANAGED] = "unmanaged",
[LINK_STATE_FAILED] = "failed",
[LINK_STATE_LINGER] = "linger",
};
DEFINE_STRING_TABLE_LOOKUP(link_state, LinkState);