networkd: add minimal IP forwarding and masquerading support to .network files

This adds two new settings to networkd's .network files:
IPForwarding=yes and IPMasquerade=yes. The former controls the
"forwarding" sysctl setting of the interface, thus controlling whether
IP forwarding shall be enabled on the specific interface. The latter
controls whether a firewall rule shall be installed that exposes traffic
coming from the interface as coming from the local host to all other
interfaces.

This also enables both options by default for container network
interfaces, thus making "systemd-nspawn --network-veth" have network
connectivity out of the box.

7 files changed, 194 insertions, 68 deletions

diff --git a/src/network/networkd-address.c b/src/network/networkd-address.c
index b28c2e08f2..b4eb91ebb6 100644
--- a/src/network/networkd-address.c
+++ b/src/network/networkd-address.c
@@ -21,13 +21,13 @@
 
 #include <net/if.h>
 
-#include "networkd.h"
-#include "networkd-link.h"
-
 #include "utf8.h"
 #include "util.h"
 #include "conf-parser.h"
+#include "fw-util.h"
 #include "network-internal.h"
+#include "networkd.h"
+#include "networkd-link.h"
 
 static void address_init(Address *address) {
         assert(address);
@@ -103,6 +103,54 @@ void address_free(Address *address) {
         free(address);
 }
 
+int address_establish(Address *address, Link *link) {
+        bool masq;
+        int r;
+
+        assert(address);
+        assert(link);
+
+        masq = link->network &&
+                link->network->ip_masquerade &&
+                address->family == AF_INET &&
+                address->scope < RT_SCOPE_LINK;
+
+        /* Add firewall entry if this is requested */
+        if (address->ip_forward_done != masq) {
+                union in_addr_union masked = address->in_addr;
+                in_addr_mask(address->family, &masked, address->prefixlen);
+
+                r = fw_add_masquerade(masq, AF_INET, 0, &masked, address->prefixlen, NULL, NULL, 0);
+                if (r < 0)
+                        log_link_warning_errno(link, r, "Could not enable IP masquerading: %m");
+
+                address->ip_forward_done = masq;
+        }
+
+        return 0;
+}
+
+int address_release(Address *address, Link *link) {
+        int r;
+
+        assert(address);
+        assert(link);
+
+        /* Remove masquerading firewall entry if it was added */
+        if (address->ip_forward_done) {
+                union in_addr_union masked = address->in_addr;
+                in_addr_mask(address->family, &masked, address->prefixlen);
+
+                r = fw_add_masquerade(false, AF_INET, 0, &masked, address->prefixlen, NULL, NULL, 0);
+                if (r < 0)
+                        log_link_warning_errno(link, r, "Failed to disable IP masquerading: %m");
+
+                address->ip_forward_done = false;
+        }
+
+        return 0;
+}
+
 int address_drop(Address *address, Link *link,
                  sd_rtnl_message_handler_t callback) {
         _cleanup_rtnl_message_unref_ sd_rtnl_message *req = NULL;
@@ -115,6 +163,8 @@ int address_drop(Address *address, Link *link,
         assert(link->manager);
         assert(link->manager->rtnl);
 
+        address_release(address, link);
+
         r = sd_rtnl_message_new_addr(link->manager->rtnl, &req, RTM_DELADDR,
                                      link->ifindex, address->family);
         if (r < 0)
@@ -333,6 +383,8 @@ int address_configure(Address *address, Link *link,
 
         link_ref(link);
 
+        address_establish(address, link);
+
         return 0;
 }
 
@@ -549,8 +601,7 @@ bool address_equal(Address *a1, Address *a2) {
                         return (b1 >> (32 - a1->prefixlen)) == (b2 >> (32 - a1->prefixlen));
                 }
 
-        case AF_INET6:
-        {
+        case AF_INET6: {
                 uint64_t *b1, *b2;
 
                 b1 = (uint64_t*)&a1->in_addr.in6;
@@ -558,6 +609,7 @@ bool address_equal(Address *a1, Address *a2) {
 
                 return (((b1[0] ^ b2[0]) | (b1[1] ^ b2[1])) == 0UL);
         }
+
         default:
                 assert_not_reached("Invalid address family");
         }
diff --git a/src/network/networkd-link.c b/src/network/networkd-link.c
index 63c7a8bf24..12944a03a3 100644
--- a/src/network/networkd-link.c
+++ b/src/network/networkd-link.c
@@ -23,16 +23,16 @@
 #include <linux/if.h>
 #include <unistd.h>
 
-#include "networkd-link.h"
-#include "networkd-netdev.h"
-#include "libudev-private.h"
-#include "udev-util.h"
 #include "util.h"
 #include "virt.h"
+#include "fileio.h"
 #include "bus-util.h"
+#include "udev-util.h"
+#include "libudev-private.h"
 #include "network-internal.h"
+#include "networkd-link.h"
+#include "networkd-netdev.h"
 #include "conf-parser.h"
-
 #include "dhcp-lease-internal.h"
 
 static bool link_dhcp6_enabled(Link *link) {
@@ -82,12 +82,22 @@ static bool link_lldp_enabled(Link *link) {
         if (!link->network)
                 return false;
 
-        if(link->network->bridge)
+        if (link->network->bridge)
                 return false;
 
         return link->network->lldp;
 }
 
+static bool link_ip_forward_enabled(Link *link) {
+        if (link->flags & IFF_LOOPBACK)
+                return false;
+
+        if (!link->network)
+                return false;
+
+        return link->network->ip_forward;
+}
+
 #define FLAG_STRING(string, flag, old, new) \
         (((old ^ new) & flag) \
                 ? ((old & flag) ? (" -" string) : (" +" string)) \
@@ -653,9 +663,7 @@ static int link_enter_set_addresses(Link *link) {
         LIST_FOREACH(addresses, ad, link->network->static_addresses) {
                 r = address_configure(ad, link, &address_handler);
                 if (r < 0) {
-                        log_link_warning(link,
-                                         "could not set addresses: %s",
-                                         strerror(-r));
+                        log_link_warning_errno(link, r, "Could not set addresses: %m");
                         link_enter_failed(link);
                         return r;
                 }
@@ -1217,6 +1225,18 @@ static int link_enter_join_netdev(Link *link) {
         return 0;
 }
 
+static int link_set_ip_forward(Link *link) {
+        const char *p = NULL;
+        int r;
+
+        p = strappenda("/proc/sys/net/ipv4/conf/", link->ifname, "/forwarding");
+        r = write_string_file_no_create(p, link_ip_forward_enabled(link) ? "1" : "0");
+        if (r < 0)
+                log_link_warning_errno(link, r, "Cannot configure IP forwarding for interface: %m");
+
+        return 0;
+}
+
 static int link_configure(Link *link) {
         int r;
 
@@ -1228,6 +1248,10 @@ static int link_configure(Link *link) {
         if (r < 0)
                 return r;
 
+        r = link_set_ip_forward(link);
+        if (r < 0)
+                return r;
+
         if (link_ipv4ll_enabled(link)) {
                 r = ipv4ll_configure(link);
                 if (r < 0)
@@ -1364,16 +1388,27 @@ int link_initialized(Link *link, struct udev_device *device) {
         return 0;
 }
 
+static Address* link_get_equal_address(Link *link, Address *needle) {
+        Address *i;
+
+        assert(link);
+        assert(needle);
+
+        LIST_FOREACH(addresses, i, link->addresses)
+                if (address_equal(i, needle))
+                        return i;
+
+        return NULL;
+}
+
 int link_rtnl_process_address(sd_rtnl *rtnl, sd_rtnl_message *message, void *userdata) {
         Manager *m = userdata;
         Link *link = NULL;
         uint16_t type;
         _cleanup_address_free_ Address *address = NULL;
-        Address *ad;
-        char buf[INET6_ADDRSTRLEN];
-        char valid_buf[FORMAT_TIMESPAN_MAX];
+        Address *existing;
+        char buf[INET6_ADDRSTRLEN], valid_buf[FORMAT_TIMESPAN_MAX];
         const char *valid_str = NULL;
-        bool address_dropped = false;
         int r, ifindex;
 
         assert(rtnl);
@@ -1415,50 +1450,42 @@ int link_rtnl_process_address(sd_rtnl *rtnl, sd_rtnl_message *message, void *use
 
         r = sd_rtnl_message_addr_get_family(message, &address->family);
         if (r < 0 || !IN_SET(address->family, AF_INET, AF_INET6)) {
-                log_link_warning(link,
-                                 "rtnl: received address with invalid family, ignoring");
+                log_link_warning(link, "rtnl: received address with invalid family, ignoring");
                 return 0;
         }
 
         r = sd_rtnl_message_addr_get_prefixlen(message, &address->prefixlen);
         if (r < 0) {
-                log_link_warning(link,
-                                 "rtnl: received address with invalid prefixlen, ignoring");
+                log_link_warning(link, "rtnl: received address with invalid prefixlen, ignoring");
                 return 0;
         }
 
         r = sd_rtnl_message_addr_get_scope(message, &address->scope);
         if (r < 0) {
-                log_link_warning(link,
-                                 "rtnl: received address with invalid scope, ignoring");
+                log_link_warning(link, "rtnl: received address with invalid scope, ignoring");
                 return 0;
         }
 
         r = sd_rtnl_message_addr_get_flags(message, &address->flags);
         if (r < 0) {
-                log_link_warning(link,
-                                 "rtnl: received address with invalid flags, ignoring");
+                log_link_warning(link, "rtnl: received address with invalid flags, ignoring");
                 return 0;
         }
 
         switch (address->family) {
         case AF_INET:
-                r = sd_rtnl_message_read_in_addr(message, IFA_LOCAL,
-                                                 &address->in_addr.in);
+                r = sd_rtnl_message_read_in_addr(message, IFA_LOCAL, &address->in_addr.in);
                 if (r < 0) {
-                        log_link_warning(link,
-                                         "rtnl: received address without valid address, ignoring");
+                        log_link_warning(link, "rtnl: received address without valid address, ignoring");
                         return 0;
                 }
 
                 break;
 
         case AF_INET6:
-                r = sd_rtnl_message_read_in6_addr(message, IFA_ADDRESS,
-                                                  &address->in_addr.in6);
+                r = sd_rtnl_message_read_in6_addr(message, IFA_ADDRESS, &address->in_addr.in6);
                 if (r < 0) {
-                        log_link_warning(link,
-                                         "rtnl: received address without valid address, ignoring");
+                        log_link_warning(link, "rtnl: received address without valid address, ignoring");
                         return 0;
                 }
 
@@ -1468,14 +1495,12 @@ int link_rtnl_process_address(sd_rtnl *rtnl, sd_rtnl_message *message, void *use
                 assert_not_reached("invalid address family");
         }
 
-        if (!inet_ntop(address->family, &address->in_addr, buf,
-                       INET6_ADDRSTRLEN)) {
+        if (!inet_ntop(address->family, &address->in_addr, buf, INET6_ADDRSTRLEN)) {
                 log_link_warning(link, "could not print address");
                 return 0;
         }
 
-        r = sd_rtnl_message_read_cache_info(message, IFA_CACHEINFO,
-                                            &address->cinfo);
+        r = sd_rtnl_message_read_cache_info(message, IFA_CACHEINFO, &address->cinfo);
         if (r >= 0) {
                 if (address->cinfo.ifa_valid == CACHE_INFO_INFINITY_LIFE_TIME)
                         valid_str = "ever";
@@ -1485,43 +1510,40 @@ int link_rtnl_process_address(sd_rtnl *rtnl, sd_rtnl_message *message, void *use
                                                     USEC_PER_SEC);
         }
 
-        LIST_FOREACH(addresses, ad, link->addresses) {
-                if (address_equal(ad, address)) {
-                        LIST_REMOVE(addresses, link->addresses, ad);
+        existing = link_get_equal_address(link, address);
+
+        switch (type) {
+        case RTM_NEWADDR:
+                if (existing) {
+                        log_link_debug(link, "Updating address: %s/%u (valid for %s)", buf, address->prefixlen, valid_str);
 
-                        address_free(ad);
 
-                        address_dropped = true;
+                        existing->scope = address->scope;
+                        existing->flags = address->flags;
+                        existing->cinfo = address->cinfo;
 
-                        break;
-                }
-        }
+                } else {
+                        log_link_debug(link, "Adding address: %s/%u (valid for %s)", buf, address->prefixlen, valid_str);
 
-        switch (type) {
-        case RTM_NEWADDR:
-                if (!address_dropped)
-                        log_link_debug(link, "added address: %s/%u (valid for %s)",
-                                       buf, address->prefixlen, valid_str);
-                else
-                        log_link_debug(link, "updated address: %s/%u (valid for %s)",
-                                       buf, address->prefixlen, valid_str);
+                        LIST_PREPEND(addresses, link->addresses, address);
+                        address_establish(address, link);
 
-                LIST_PREPEND(addresses, link->addresses, address);
-                address = NULL;
+                        address = NULL;
 
-                link_save(link);
+                        link_save(link);
+                }
 
                 break;
+
         case RTM_DELADDR:
-                if (address_dropped) {
-                        log_link_debug(link, "removed address: %s/%u (valid for %s)",
-                                       buf, address->prefixlen, valid_str);
 
-                        link_save(link);
+                if (existing) {
+                        log_link_debug(link, "Removing address: %s/%u (valid for %s)", buf, address->prefixlen, valid_str);
+                        address_release(existing, link);
+                        LIST_REMOVE(addresses, link->addresses, existing);
+                        address_free(existing);
                 } else
-                        log_link_warning(link,
-                                         "removing non-existent address: %s/%u (valid for %s)",
-                                         buf, address->prefixlen, valid_str);
+                        log_link_warning(link, "Removing non-existent address: %s/%u (valid for %s)", buf, address->prefixlen, valid_str);
 
                 break;
         default:
diff --git a/src/network/networkd-network-gperf.gperf b/src/network/networkd-network-gperf.gperf
index 5094b488f3..3eb37b4d24 100644
--- a/src/network/networkd-network-gperf.gperf
+++ b/src/network/networkd-network-gperf.gperf
@@ -44,6 +44,8 @@ Network.Domains,             config_parse_domains,               0,
 Network.DNS,                 config_parse_strv,                  0,                             offsetof(Network, dns)
 Network.LLMNR,               config_parse_llmnr,                 0,                             offsetof(Network, llmnr)
 Network.NTP,                 config_parse_strv,                  0,                             offsetof(Network, ntp)
+Network.IPForward,           config_parse_bool,                  0,                             offsetof(Network, ip_forward)
+Network.IPMasquerade,        config_parse_bool,                  0,                             offsetof(Network, ip_masquerade)
 Address.Address,             config_parse_address,               0,                             0
 Address.Peer,                config_parse_address,               0,                             0
 Address.Broadcast,           config_parse_broadcast,             0,                             0
diff --git a/src/network/networkd-network.c b/src/network/networkd-network.c
index 966b59b878..d6504cc178 100644
--- a/src/network/networkd-network.c
+++ b/src/network/networkd-network.c
@@ -22,14 +22,14 @@
 #include <ctype.h>
 #include <net/if.h>
 
-#include "networkd.h"
-#include "networkd-netdev.h"
-#include "networkd-link.h"
-#include "network-internal.h"
 #include "path-util.h"
 #include "conf-files.h"
 #include "conf-parser.h"
 #include "util.h"
+#include "networkd.h"
+#include "networkd-netdev.h"
+#include "networkd-link.h"
+#include "network-internal.h"
 
 static int network_load_one(Manager *manager, const char *filename) {
         _cleanup_network_free_ Network *network = NULL;
@@ -109,6 +109,10 @@ static int network_load_one(Manager *manager, const char *filename) {
         if (r < 0)
                 return r;
 
+        /* IPMasquerade=yes implies IPForward=yes */
+        if (network->ip_masquerade)
+                network->ip_forward = true;
+
         LIST_PREPEND(networks, manager->networks, network);
 
         LIST_FOREACH(routes, route, network->static_routes) {
diff --git a/src/network/networkd.h b/src/network/networkd.h
index 7107c5f932..39b2d2bec9 100644
--- a/src/network/networkd.h
+++ b/src/network/networkd.h
@@ -120,6 +120,9 @@ struct Network {
 
         unsigned cost;
 
+        bool ip_masquerade;
+        bool ip_forward;
+
         struct ether_addr *mac;
         unsigned mtu;
 
@@ -157,6 +160,8 @@ struct Address {
         union in_addr_union in_addr;
         union in_addr_union in_addr_peer;
 
+        bool ip_forward_done;
+
         LIST_FIELDS(Address, addresses);
 };
 
@@ -326,6 +331,8 @@ void address_free(Address *address);
 int address_configure(Address *address, Link *link, sd_rtnl_message_handler_t callback);
 int address_update(Address *address, Link *link, sd_rtnl_message_handler_t callback);
 int address_drop(Address *address, Link *link, sd_rtnl_message_handler_t callback);
+int address_establish(Address *address, Link *link);
+int address_release(Address *address, Link *link);
 bool address_equal(Address *a1, Address *a2);
 
 DEFINE_TRIVIAL_CLEANUP_FUNC(Address*, address_free);
diff --git a/src/shared/in-addr-util.c b/src/shared/in-addr-util.c
index b02e7516ca..d88864b598 100644
--- a/src/shared/in-addr-util.c
+++ b/src/shared/in-addr-util.c
@@ -300,3 +300,39 @@ int in_addr_default_subnet_mask(const struct in_addr *addr, struct in_addr *mask
         in_addr_prefixlen_to_netmask(mask, prefixlen);
         return 0;
 }
+
+int in_addr_mask(int family, union in_addr_union *addr, unsigned char prefixlen) {
+        assert(addr);
+
+        if (family == AF_INET) {
+                struct in_addr mask;
+
+                if (!in_addr_prefixlen_to_netmask(&mask, prefixlen))
+                        return -EINVAL;
+
+                addr->in.s_addr &= mask.s_addr;
+                return 0;
+        }
+
+        if (family == AF_INET6) {
+                unsigned i;
+
+                for (i = 0; i < 16; i++) {
+                        uint8_t mask;
+
+                        if (prefixlen >= 8) {
+                                mask = 0xFF;
+                                prefixlen -= 8;
+                        } else {
+                                mask = 0xFF << (8 - prefixlen);
+                                prefixlen = 0;
+                        }
+
+                        addr->in6.s6_addr[i] &= mask;
+                }
+
+                return 0;
+        }
+
+        return -EAFNOSUPPORT;
+}
diff --git a/src/shared/in-addr-util.h b/src/shared/in-addr-util.h
index 4cf4418b4b..51af08868c 100644
--- a/src/shared/in-addr-util.h
+++ b/src/shared/in-addr-util.h
@@ -43,8 +43,11 @@ unsigned char in_addr_netmask_to_prefixlen(const struct in_addr *addr);
 struct in_addr* in_addr_prefixlen_to_netmask(struct in_addr *addr, unsigned char prefixlen);
 int in_addr_default_prefixlen(const struct in_addr *addr, unsigned char *prefixlen);
 int in_addr_default_subnet_mask(const struct in_addr *addr, struct in_addr *mask);
+int in_addr_mask(int family, union in_addr_union *addr, unsigned char prefixlen);
 
 static inline size_t FAMILY_ADDRESS_SIZE(int family) {
         assert(family == AF_INET || family == AF_INET6);
         return family == AF_INET6 ? 16 : 4;
 }
+
+#define IN_ADDR_NULL ((union in_addr_union) {})