summaryrefslogtreecommitdiffstats
path: root/src/network/networkd-nexthop.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/network/networkd-nexthop.c')
-rw-r--r--src/network/networkd-nexthop.c1406
1 files changed, 1406 insertions, 0 deletions
diff --git a/src/network/networkd-nexthop.c b/src/network/networkd-nexthop.c
new file mode 100644
index 0000000..d827667
--- /dev/null
+++ b/src/network/networkd-nexthop.c
@@ -0,0 +1,1406 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later
+ * Copyright © 2019 VMware, Inc.
+ */
+
+#include <net/if.h>
+#include <linux/nexthop.h>
+
+#include "alloc-util.h"
+#include "netlink-util.h"
+#include "networkd-link.h"
+#include "networkd-manager.h"
+#include "networkd-network.h"
+#include "networkd-nexthop.h"
+#include "networkd-queue.h"
+#include "networkd-route-util.h"
+#include "parse-util.h"
+#include "set.h"
+#include "stdio-util.h"
+#include "string-util.h"
+
+NextHop *nexthop_free(NextHop *nexthop) {
+ if (!nexthop)
+ return NULL;
+
+ if (nexthop->network) {
+ assert(nexthop->section);
+ hashmap_remove(nexthop->network->nexthops_by_section, nexthop->section);
+ }
+
+ config_section_free(nexthop->section);
+
+ if (nexthop->link) {
+ set_remove(nexthop->link->nexthops, nexthop);
+
+ if (nexthop->link->manager && nexthop->id > 0)
+ hashmap_remove(nexthop->link->manager->nexthops_by_id, UINT32_TO_PTR(nexthop->id));
+ }
+
+ if (nexthop->manager) {
+ set_remove(nexthop->manager->nexthops, nexthop);
+
+ if (nexthop->id > 0)
+ hashmap_remove(nexthop->manager->nexthops_by_id, UINT32_TO_PTR(nexthop->id));
+ }
+
+ hashmap_free_free(nexthop->group);
+
+ return mfree(nexthop);
+}
+
+DEFINE_SECTION_CLEANUP_FUNCTIONS(NextHop, nexthop_free);
+
+static int nexthop_new(NextHop **ret) {
+ _cleanup_(nexthop_freep) NextHop *nexthop = NULL;
+
+ nexthop = new(NextHop, 1);
+ if (!nexthop)
+ return -ENOMEM;
+
+ *nexthop = (NextHop) {
+ .family = AF_UNSPEC,
+ .onlink = -1,
+ };
+
+ *ret = TAKE_PTR(nexthop);
+
+ return 0;
+}
+
+static int nexthop_new_static(Network *network, const char *filename, unsigned section_line, NextHop **ret) {
+ _cleanup_(config_section_freep) ConfigSection *n = NULL;
+ _cleanup_(nexthop_freep) NextHop *nexthop = NULL;
+ int r;
+
+ assert(network);
+ assert(ret);
+ assert(filename);
+ assert(section_line > 0);
+
+ r = config_section_new(filename, section_line, &n);
+ if (r < 0)
+ return r;
+
+ nexthop = hashmap_get(network->nexthops_by_section, n);
+ if (nexthop) {
+ *ret = TAKE_PTR(nexthop);
+ return 0;
+ }
+
+ r = nexthop_new(&nexthop);
+ if (r < 0)
+ return r;
+
+ nexthop->protocol = RTPROT_STATIC;
+ nexthop->network = network;
+ nexthop->section = TAKE_PTR(n);
+ nexthop->source = NETWORK_CONFIG_SOURCE_STATIC;
+
+ r = hashmap_ensure_put(&network->nexthops_by_section, &config_section_hash_ops, nexthop->section, nexthop);
+ if (r < 0)
+ return r;
+
+ *ret = TAKE_PTR(nexthop);
+ return 0;
+}
+
+static void nexthop_hash_func(const NextHop *nexthop, struct siphash *state) {
+ assert(nexthop);
+
+ siphash24_compress(&nexthop->protocol, sizeof(nexthop->protocol), state);
+ siphash24_compress(&nexthop->id, sizeof(nexthop->id), state);
+ siphash24_compress(&nexthop->blackhole, sizeof(nexthop->blackhole), state);
+ siphash24_compress(&nexthop->family, sizeof(nexthop->family), state);
+
+ switch (nexthop->family) {
+ case AF_INET:
+ case AF_INET6:
+ siphash24_compress(&nexthop->gw, FAMILY_ADDRESS_SIZE(nexthop->family), state);
+
+ break;
+ default:
+ /* treat any other address family as AF_UNSPEC */
+ break;
+ }
+}
+
+static int nexthop_compare_func(const NextHop *a, const NextHop *b) {
+ int r;
+
+ r = CMP(a->protocol, b->protocol);
+ if (r != 0)
+ return r;
+
+ r = CMP(a->id, b->id);
+ if (r != 0)
+ return r;
+
+ r = CMP(a->blackhole, b->blackhole);
+ if (r != 0)
+ return r;
+
+ r = CMP(a->family, b->family);
+ if (r != 0)
+ return r;
+
+ if (IN_SET(a->family, AF_INET, AF_INET6))
+ return memcmp(&a->gw, &b->gw, FAMILY_ADDRESS_SIZE(a->family));
+
+ return 0;
+}
+
+DEFINE_PRIVATE_HASH_OPS_WITH_KEY_DESTRUCTOR(
+ nexthop_hash_ops,
+ NextHop,
+ nexthop_hash_func,
+ nexthop_compare_func,
+ nexthop_free);
+
+static bool nexthop_equal(const NextHop *a, const NextHop *b) {
+ if (a == b)
+ return true;
+
+ if (!a || !b)
+ return false;
+
+ return nexthop_compare_func(a, b) == 0;
+}
+
+static int nexthop_dup(const NextHop *src, NextHop **ret) {
+ _cleanup_(nexthop_freep) NextHop *dest = NULL;
+ struct nexthop_grp *nhg;
+ int r;
+
+ assert(src);
+ assert(ret);
+
+ dest = newdup(NextHop, src, 1);
+ if (!dest)
+ return -ENOMEM;
+
+ /* unset all pointers */
+ dest->manager = NULL;
+ dest->link = NULL;
+ dest->network = NULL;
+ dest->section = NULL;
+ dest->group = NULL;
+
+ HASHMAP_FOREACH(nhg, src->group) {
+ _cleanup_free_ struct nexthop_grp *g = NULL;
+
+ g = newdup(struct nexthop_grp, nhg, 1);
+ if (!g)
+ return -ENOMEM;
+
+ r = hashmap_ensure_put(&dest->group, NULL, UINT32_TO_PTR(g->id), g);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ TAKE_PTR(g);
+ }
+
+ *ret = TAKE_PTR(dest);
+ return 0;
+}
+
+int manager_get_nexthop_by_id(Manager *manager, uint32_t id, NextHop **ret) {
+ NextHop *nh;
+
+ assert(manager);
+
+ if (id == 0)
+ return -EINVAL;
+
+ nh = hashmap_get(manager->nexthops_by_id, UINT32_TO_PTR(id));
+ if (!nh)
+ return -ENOENT;
+
+ if (ret)
+ *ret = nh;
+ return 0;
+}
+
+static bool nexthop_owned_by_link(const NextHop *nexthop) {
+ return !nexthop->blackhole && hashmap_isempty(nexthop->group);
+}
+
+static int nexthop_get(Manager *manager, Link *link, NextHop *in, NextHop **ret) {
+ NextHop *nexthop;
+ Set *nexthops;
+
+ assert(in);
+
+ if (nexthop_owned_by_link(in)) {
+ if (!link)
+ return -ENOENT;
+
+ nexthops = link->nexthops;
+ } else {
+ if (!manager)
+ return -ENOENT;
+
+ nexthops = manager->nexthops;
+ }
+
+ nexthop = set_get(nexthops, in);
+ if (nexthop) {
+ if (ret)
+ *ret = nexthop;
+ return 0;
+ }
+
+ if (in->id > 0)
+ return -ENOENT;
+
+ /* Also find nexthop configured without ID. */
+ SET_FOREACH(nexthop, nexthops) {
+ uint32_t id;
+ bool found;
+
+ id = nexthop->id;
+ nexthop->id = 0;
+ found = nexthop_equal(nexthop, in);
+ nexthop->id = id;
+
+ if (!found)
+ continue;
+
+ if (ret)
+ *ret = nexthop;
+ return 0;
+ }
+
+ return -ENOENT;
+}
+
+static int nexthop_add(Manager *manager, Link *link, NextHop *nexthop) {
+ int r;
+
+ assert(nexthop);
+ assert(nexthop->id > 0);
+
+ if (nexthop_owned_by_link(nexthop)) {
+ assert(link);
+
+ r = set_ensure_put(&link->nexthops, &nexthop_hash_ops, nexthop);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ return -EEXIST;
+
+ nexthop->link = link;
+
+ manager = link->manager;
+ } else {
+ assert(manager);
+
+ r = set_ensure_put(&manager->nexthops, &nexthop_hash_ops, nexthop);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ return -EEXIST;
+
+ nexthop->manager = manager;
+ }
+
+ return hashmap_ensure_put(&manager->nexthops_by_id, NULL, UINT32_TO_PTR(nexthop->id), nexthop);
+}
+
+static int nexthop_acquire_id(Manager *manager, NextHop *nexthop) {
+ _cleanup_set_free_ Set *ids = NULL;
+ Network *network;
+ uint32_t id;
+ int r;
+
+ assert(manager);
+ assert(nexthop);
+
+ if (nexthop->id > 0)
+ return 0;
+
+ /* Find the lowest unused ID. */
+
+ ORDERED_HASHMAP_FOREACH(network, manager->networks) {
+ NextHop *tmp;
+
+ HASHMAP_FOREACH(tmp, network->nexthops_by_section) {
+ if (tmp->id == 0)
+ continue;
+
+ r = set_ensure_put(&ids, NULL, UINT32_TO_PTR(tmp->id));
+ if (r < 0)
+ return r;
+ }
+ }
+
+ for (id = 1; id < UINT32_MAX; id++) {
+ if (manager_get_nexthop_by_id(manager, id, NULL) >= 0)
+ continue;
+ if (set_contains(ids, UINT32_TO_PTR(id)))
+ continue;
+ break;
+ }
+
+ nexthop->id = id;
+ return 0;
+}
+
+static void log_nexthop_debug(const NextHop *nexthop, const char *str, const Link *link) {
+ _cleanup_free_ char *state = NULL, *group = NULL, *flags = NULL;
+ struct nexthop_grp *nhg;
+
+ assert(nexthop);
+ assert(str);
+
+ /* link may be NULL. */
+
+ if (!DEBUG_LOGGING)
+ return;
+
+ (void) network_config_state_to_string_alloc(nexthop->state, &state);
+ (void) route_flags_to_string_alloc(nexthop->flags, &flags);
+
+ HASHMAP_FOREACH(nhg, nexthop->group)
+ (void) strextendf_with_separator(&group, ",", "%"PRIu32":%"PRIu32, nhg->id, nhg->weight+1u);
+
+ log_link_debug(link, "%s %s nexthop (%s): id: %"PRIu32", gw: %s, blackhole: %s, group: %s, flags: %s",
+ str, strna(network_config_source_to_string(nexthop->source)), strna(state),
+ nexthop->id,
+ IN_ADDR_TO_STRING(nexthop->family, &nexthop->gw),
+ yes_no(nexthop->blackhole), strna(group), strna(flags));
+}
+
+static int nexthop_remove_handler(sd_netlink *rtnl, sd_netlink_message *m, Link *link) {
+ int r;
+
+ assert(m);
+
+ /* link may be NULL. */
+
+ if (link && IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
+ return 1;
+
+ r = sd_netlink_message_get_errno(m);
+ if (r < 0 && r != -ENOENT)
+ log_link_message_warning_errno(link, m, r, "Could not drop nexthop, ignoring");
+
+ return 1;
+}
+
+static int nexthop_remove(NextHop *nexthop) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *req = NULL;
+ Manager *manager;
+ Link *link;
+ int r;
+
+ assert(nexthop);
+ assert(nexthop->manager || (nexthop->link && nexthop->link->manager));
+
+ /* link may be NULL. */
+ link = nexthop->link;
+ manager = nexthop->manager ?: nexthop->link->manager;
+
+ if (nexthop->id == 0) {
+ log_link_debug(link, "Cannot remove nexthop without valid ID, ignoring.");
+ return 0;
+ }
+
+ log_nexthop_debug(nexthop, "Removing", link);
+
+ r = sd_rtnl_message_new_nexthop(manager->rtnl, &req, RTM_DELNEXTHOP, AF_UNSPEC, RTPROT_UNSPEC);
+ if (r < 0)
+ return log_link_error_errno(link, r, "Could not create RTM_DELNEXTHOP message: %m");
+
+ r = sd_netlink_message_append_u32(req, NHA_ID, nexthop->id);
+ if (r < 0)
+ return log_link_error_errno(link, r, "Could not append NHA_ID attribute: %m");
+
+ r = netlink_call_async(manager->rtnl, NULL, req, nexthop_remove_handler,
+ link ? link_netlink_destroy_callback : NULL, link);
+ if (r < 0)
+ return log_link_error_errno(link, r, "Could not send rtnetlink message: %m");
+
+ link_ref(link); /* link may be NULL, link_ref() is OK with that */
+
+ nexthop_enter_removing(nexthop);
+ return 0;
+}
+
+static int nexthop_configure(NextHop *nexthop, Link *link, Request *req) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ int r;
+
+ assert(nexthop);
+ assert(IN_SET(nexthop->family, AF_UNSPEC, AF_INET, AF_INET6));
+ assert(link);
+ assert(link->manager);
+ assert(link->manager->rtnl);
+ assert(link->ifindex > 0);
+ assert(req);
+
+ log_nexthop_debug(nexthop, "Configuring", link);
+
+ r = sd_rtnl_message_new_nexthop(link->manager->rtnl, &m, RTM_NEWNEXTHOP, nexthop->family, nexthop->protocol);
+ if (r < 0)
+ return r;
+
+ if (nexthop->id > 0) {
+ r = sd_netlink_message_append_u32(m, NHA_ID, nexthop->id);
+ if (r < 0)
+ return r;
+ }
+
+ if (!hashmap_isempty(nexthop->group)) {
+ _cleanup_free_ struct nexthop_grp *group = NULL;
+ struct nexthop_grp *p, *nhg;
+
+ group = new(struct nexthop_grp, hashmap_size(nexthop->group));
+ if (!group)
+ return log_oom();
+
+ p = group;
+ HASHMAP_FOREACH(nhg, nexthop->group)
+ *p++ = *nhg;
+
+ r = sd_netlink_message_append_data(m, NHA_GROUP, group, sizeof(struct nexthop_grp) * hashmap_size(nexthop->group));
+ if (r < 0)
+ return r;
+
+ } else if (nexthop->blackhole) {
+ r = sd_netlink_message_append_flag(m, NHA_BLACKHOLE);
+ if (r < 0)
+ return r;
+ } else {
+ r = sd_netlink_message_append_u32(m, NHA_OIF, link->ifindex);
+ if (r < 0)
+ return r;
+
+ if (in_addr_is_set(nexthop->family, &nexthop->gw)) {
+ r = netlink_message_append_in_addr_union(m, NHA_GATEWAY, nexthop->family, &nexthop->gw);
+ if (r < 0)
+ return r;
+
+ r = sd_rtnl_message_nexthop_set_flags(m, nexthop->flags & RTNH_F_ONLINK);
+ if (r < 0)
+ return r;
+ }
+ }
+
+ return request_call_netlink_async(link->manager->rtnl, m, req);
+}
+
+static int static_nexthop_handler(sd_netlink *rtnl, sd_netlink_message *m, Request *req, Link *link, NextHop *nexthop) {
+ int r;
+
+ assert(m);
+ assert(link);
+
+ r = sd_netlink_message_get_errno(m);
+ if (r < 0 && r != -EEXIST) {
+ log_link_message_warning_errno(link, m, r, "Could not set nexthop");
+ link_enter_failed(link);
+ return 1;
+ }
+
+ if (link->static_nexthop_messages == 0) {
+ log_link_debug(link, "Nexthops set");
+ link->static_nexthops_configured = true;
+ link_check_ready(link);
+ }
+
+ return 1;
+}
+
+static bool nexthop_is_ready_to_configure(Link *link, const NextHop *nexthop) {
+ struct nexthop_grp *nhg;
+
+ assert(link);
+ assert(nexthop);
+
+ if (!link_is_ready_to_configure(link, false))
+ return false;
+
+ if (nexthop_owned_by_link(nexthop)) {
+ /* TODO: fdb nexthop does not require IFF_UP. The conditions below needs to be updated
+ * when fdb nexthop support is added. See rtm_to_nh_config() in net/ipv4/nexthop.c of
+ * kernel. */
+ if (link->set_flags_messages > 0)
+ return false;
+ if (!FLAGS_SET(link->flags, IFF_UP))
+ return false;
+ }
+
+ /* All group members must be configured first. */
+ HASHMAP_FOREACH(nhg, nexthop->group) {
+ NextHop *g;
+
+ if (manager_get_nexthop_by_id(link->manager, nhg->id, &g) < 0)
+ return false;
+
+ if (!nexthop_exists(g))
+ return false;
+ }
+
+ if (nexthop->id == 0) {
+ Request *req;
+
+ ORDERED_SET_FOREACH(req, link->manager->request_queue) {
+ if (req->type != REQUEST_TYPE_NEXTHOP)
+ continue;
+ if (((NextHop*) req->userdata)->id != 0)
+ return false; /* first configure nexthop with id. */
+ }
+ }
+
+ return gateway_is_ready(link, FLAGS_SET(nexthop->flags, RTNH_F_ONLINK), nexthop->family, &nexthop->gw);
+}
+
+static int nexthop_process_request(Request *req, Link *link, NextHop *nexthop) {
+ int r;
+
+ assert(req);
+ assert(link);
+ assert(nexthop);
+
+ if (!nexthop_is_ready_to_configure(link, nexthop))
+ return 0;
+
+ r = nexthop_configure(nexthop, link, req);
+ if (r < 0)
+ return log_link_warning_errno(link, r, "Failed to configure nexthop");
+
+ nexthop_enter_configuring(nexthop);
+ return 1;
+}
+
+static int link_request_nexthop(Link *link, NextHop *nexthop) {
+ NextHop *existing;
+ int r;
+
+ assert(link);
+ assert(nexthop);
+ assert(nexthop->source != NETWORK_CONFIG_SOURCE_FOREIGN);
+
+ if (nexthop_get(link->manager, link, nexthop, &existing) < 0) {
+ _cleanup_(nexthop_freep) NextHop *tmp = NULL;
+
+ r = nexthop_dup(nexthop, &tmp);
+ if (r < 0)
+ return r;
+
+ r = nexthop_acquire_id(link->manager, tmp);
+ if (r < 0)
+ return r;
+
+ r = nexthop_add(link->manager, link, tmp);
+ if (r < 0)
+ return r;
+
+ existing = TAKE_PTR(tmp);
+ } else
+ existing->source = nexthop->source;
+
+ log_nexthop_debug(existing, "Requesting", link);
+ r = link_queue_request_safe(link, REQUEST_TYPE_NEXTHOP,
+ existing, NULL,
+ nexthop_hash_func,
+ nexthop_compare_func,
+ nexthop_process_request,
+ &link->static_nexthop_messages,
+ static_nexthop_handler,
+ NULL);
+ if (r <= 0)
+ return r;
+
+ nexthop_enter_requesting(existing);
+ return 1;
+}
+
+int link_request_static_nexthops(Link *link, bool only_ipv4) {
+ NextHop *nh;
+ int r;
+
+ assert(link);
+ assert(link->network);
+
+ link->static_nexthops_configured = false;
+
+ HASHMAP_FOREACH(nh, link->network->nexthops_by_section) {
+ if (only_ipv4 && nh->family != AF_INET)
+ continue;
+
+ r = link_request_nexthop(link, nh);
+ if (r < 0)
+ return log_link_warning_errno(link, r, "Could not request nexthop: %m");
+ }
+
+ if (link->static_nexthop_messages == 0) {
+ link->static_nexthops_configured = true;
+ link_check_ready(link);
+ } else {
+ log_link_debug(link, "Requesting nexthops");
+ link_set_state(link, LINK_STATE_CONFIGURING);
+ }
+
+ return 0;
+}
+
+static void manager_mark_nexthops(Manager *manager, bool foreign, const Link *except) {
+ NextHop *nexthop;
+ Link *link;
+
+ assert(manager);
+
+ /* First, mark all nexthops. */
+ SET_FOREACH(nexthop, manager->nexthops) {
+ /* do not touch nexthop created by the kernel */
+ if (nexthop->protocol == RTPROT_KERNEL)
+ continue;
+
+ /* When 'foreign' is true, mark only foreign nexthops, and vice versa. */
+ if (foreign != (nexthop->source == NETWORK_CONFIG_SOURCE_FOREIGN))
+ continue;
+
+ /* Ignore nexthops not assigned yet or already removed. */
+ if (!nexthop_exists(nexthop))
+ continue;
+
+ nexthop_mark(nexthop);
+ }
+
+ /* Then, unmark all nexthops requested by active links. */
+ HASHMAP_FOREACH(link, manager->links_by_index) {
+ if (link == except)
+ continue;
+
+ if (!IN_SET(link->state, LINK_STATE_CONFIGURING, LINK_STATE_CONFIGURED))
+ continue;
+
+ HASHMAP_FOREACH(nexthop, link->network->nexthops_by_section) {
+ NextHop *existing;
+
+ if (nexthop_get(manager, NULL, nexthop, &existing) >= 0)
+ nexthop_unmark(existing);
+ }
+ }
+}
+
+static int manager_drop_marked_nexthops(Manager *manager) {
+ NextHop *nexthop;
+ int k, r = 0;
+
+ assert(manager);
+
+ SET_FOREACH(nexthop, manager->nexthops) {
+ if (!nexthop_is_marked(nexthop))
+ continue;
+
+ k = nexthop_remove(nexthop);
+ if (k < 0 && r >= 0)
+ r = k;
+ }
+
+ return r;
+}
+
+int link_drop_foreign_nexthops(Link *link) {
+ NextHop *nexthop;
+ int k, r = 0;
+
+ assert(link);
+ assert(link->manager);
+ assert(link->network);
+
+ /* First, mark all nexthops. */
+ SET_FOREACH(nexthop, link->nexthops) {
+ /* do not touch nexthop created by the kernel */
+ if (nexthop->protocol == RTPROT_KERNEL)
+ continue;
+
+ /* Do not remove nexthops we configured. */
+ if (nexthop->source != NETWORK_CONFIG_SOURCE_FOREIGN)
+ continue;
+
+ /* Ignore nexthops not assigned yet or already removed. */
+ if (!nexthop_exists(nexthop))
+ continue;
+
+ nexthop_mark(nexthop);
+ }
+
+ /* Then, unmark all nexthops requested by active links. */
+ HASHMAP_FOREACH(nexthop, link->network->nexthops_by_section) {
+ NextHop *existing;
+
+ if (nexthop_get(NULL, link, nexthop, &existing) >= 0)
+ nexthop_unmark(existing);
+ }
+
+ /* Finally, remove all marked rules. */
+ SET_FOREACH(nexthop, link->nexthops) {
+ if (!nexthop_is_marked(nexthop))
+ continue;
+
+ k = nexthop_remove(nexthop);
+ if (k < 0 && r >= 0)
+ r = k;
+ }
+
+ manager_mark_nexthops(link->manager, /* foreign = */ true, NULL);
+
+ k = manager_drop_marked_nexthops(link->manager);
+ if (k < 0 && r >= 0)
+ r = k;
+
+ return r;
+}
+
+int link_drop_managed_nexthops(Link *link) {
+ NextHop *nexthop;
+ int k, r = 0;
+
+ assert(link);
+ assert(link->manager);
+
+ SET_FOREACH(nexthop, link->nexthops) {
+ /* do not touch nexthop created by the kernel */
+ if (nexthop->protocol == RTPROT_KERNEL)
+ continue;
+
+ /* Do not touch addresses managed by kernel or other tools. */
+ if (nexthop->source == NETWORK_CONFIG_SOURCE_FOREIGN)
+ continue;
+
+ /* Ignore nexthops not assigned yet or already removing. */
+ if (!nexthop_exists(nexthop))
+ continue;
+
+ k = nexthop_remove(nexthop);
+ if (k < 0 && r >= 0)
+ r = k;
+ }
+
+ manager_mark_nexthops(link->manager, /* foreign = */ false, link);
+
+ k = manager_drop_marked_nexthops(link->manager);
+ if (k < 0 && r >= 0)
+ r = k;
+
+ return r;
+}
+
+void link_foreignize_nexthops(Link *link) {
+ NextHop *nexthop;
+
+ assert(link);
+
+ SET_FOREACH(nexthop, link->nexthops)
+ nexthop->source = NETWORK_CONFIG_SOURCE_FOREIGN;
+
+ manager_mark_nexthops(link->manager, /* foreign = */ false, link);
+
+ SET_FOREACH(nexthop, link->manager->nexthops) {
+ if (!nexthop_is_marked(nexthop))
+ continue;
+
+ nexthop->source = NETWORK_CONFIG_SOURCE_FOREIGN;
+ }
+}
+
+int manager_rtnl_process_nexthop(sd_netlink *rtnl, sd_netlink_message *message, Manager *m) {
+ _cleanup_(nexthop_freep) NextHop *tmp = NULL;
+ _cleanup_free_ void *raw_group = NULL;
+ NextHop *nexthop = NULL;
+ size_t raw_group_size;
+ uint32_t ifindex;
+ uint16_t type;
+ Link *link = NULL;
+ int r;
+
+ assert(rtnl);
+ assert(message);
+ assert(m);
+
+ if (sd_netlink_message_is_error(message)) {
+ r = sd_netlink_message_get_errno(message);
+ if (r < 0)
+ log_message_warning_errno(message, r, "rtnl: failed to receive rule message, ignoring");
+
+ return 0;
+ }
+
+ r = sd_netlink_message_get_type(message, &type);
+ if (r < 0) {
+ log_warning_errno(r, "rtnl: could not get message type, ignoring: %m");
+ return 0;
+ } else if (!IN_SET(type, RTM_NEWNEXTHOP, RTM_DELNEXTHOP)) {
+ log_warning("rtnl: received unexpected message type %u when processing nexthop, ignoring.", type);
+ return 0;
+ }
+
+ r = sd_netlink_message_read_u32(message, NHA_OIF, &ifindex);
+ if (r < 0 && r != -ENODATA) {
+ log_warning_errno(r, "rtnl: could not get NHA_OIF attribute, ignoring: %m");
+ return 0;
+ } else if (r >= 0) {
+ if (ifindex <= 0) {
+ log_warning("rtnl: received nexthop message with invalid ifindex %"PRIu32", ignoring.", ifindex);
+ return 0;
+ }
+
+ r = link_get_by_index(m, ifindex, &link);
+ if (r < 0 || !link) {
+ if (!m->enumerating)
+ log_warning("rtnl: received nexthop message for link (%"PRIu32") we do not know about, ignoring", ifindex);
+ return 0;
+ }
+ }
+
+ r = nexthop_new(&tmp);
+ if (r < 0)
+ return log_oom();
+
+ r = sd_rtnl_message_get_family(message, &tmp->family);
+ if (r < 0) {
+ log_link_warning_errno(link, r, "rtnl: could not get nexthop family, ignoring: %m");
+ return 0;
+ } else if (!IN_SET(tmp->family, AF_UNSPEC, AF_INET, AF_INET6)) {
+ log_link_debug(link, "rtnl: received nexthop message with invalid family %d, ignoring.", tmp->family);
+ return 0;
+ }
+
+ r = sd_rtnl_message_nexthop_get_protocol(message, &tmp->protocol);
+ if (r < 0) {
+ log_link_warning_errno(link, r, "rtnl: could not get nexthop protocol, ignoring: %m");
+ return 0;
+ }
+
+ r = sd_rtnl_message_nexthop_get_flags(message, &tmp->flags);
+ if (r < 0) {
+ log_link_warning_errno(link, r, "rtnl: could not get nexthop flags, ignoring: %m");
+ return 0;
+ }
+
+ r = sd_netlink_message_read_data(message, NHA_GROUP, &raw_group_size, &raw_group);
+ if (r < 0 && r != -ENODATA) {
+ log_link_warning_errno(link, r, "rtnl: could not get NHA_GROUP attribute, ignoring: %m");
+ return 0;
+ } else if (r >= 0) {
+ struct nexthop_grp *group = raw_group;
+ size_t n_group;
+
+ if (raw_group_size == 0 || raw_group_size % sizeof(struct nexthop_grp) != 0) {
+ log_link_warning(link, "rtnl: received nexthop message with invalid nexthop group size, ignoring.");
+ return 0;
+ }
+
+ assert((uintptr_t) group % __alignof__(struct nexthop_grp) == 0);
+
+ n_group = raw_group_size / sizeof(struct nexthop_grp);
+ for (size_t i = 0; i < n_group; i++) {
+ _cleanup_free_ struct nexthop_grp *nhg = NULL;
+
+ if (group[i].id == 0) {
+ log_link_warning(link, "rtnl: received nexthop message with invalid ID in group, ignoring.");
+ return 0;
+ }
+ if (group[i].weight > 254) {
+ log_link_warning(link, "rtnl: received nexthop message with invalid weight in group, ignoring.");
+ return 0;
+ }
+
+ nhg = newdup(struct nexthop_grp, group + i, 1);
+ if (!nhg)
+ return log_oom();
+
+ r = hashmap_ensure_put(&tmp->group, NULL, UINT32_TO_PTR(nhg->id), nhg);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0) {
+ log_link_warning_errno(link, r, "Failed to store nexthop group, ignoring: %m");
+ return 0;
+ }
+ if (r > 0)
+ TAKE_PTR(nhg);
+ }
+ }
+
+ if (tmp->family != AF_UNSPEC) {
+ r = netlink_message_read_in_addr_union(message, NHA_GATEWAY, tmp->family, &tmp->gw);
+ if (r < 0 && r != -ENODATA) {
+ log_link_warning_errno(link, r, "rtnl: could not get NHA_GATEWAY attribute, ignoring: %m");
+ return 0;
+ }
+ }
+
+ r = sd_netlink_message_has_flag(message, NHA_BLACKHOLE);
+ if (r < 0) {
+ log_link_warning_errno(link, r, "rtnl: could not get NHA_BLACKHOLE attribute, ignoring: %m");
+ return 0;
+ }
+ tmp->blackhole = r;
+
+ r = sd_netlink_message_read_u32(message, NHA_ID, &tmp->id);
+ if (r == -ENODATA) {
+ log_link_warning_errno(link, r, "rtnl: received nexthop message without NHA_ID attribute, ignoring: %m");
+ return 0;
+ } else if (r < 0) {
+ log_link_warning_errno(link, r, "rtnl: could not get NHA_ID attribute, ignoring: %m");
+ return 0;
+ } else if (tmp->id == 0) {
+ log_link_warning(link, "rtnl: received nexthop message with invalid nexthop ID, ignoring: %m");
+ return 0;
+ }
+
+ /* All blackhole or group nexthops are managed by Manager. Note that the linux kernel does not
+ * set NHA_OID attribute when NHA_BLACKHOLE or NHA_GROUP is set. Just for safety. */
+ if (!nexthop_owned_by_link(tmp))
+ link = NULL;
+
+ (void) nexthop_get(m, link, tmp, &nexthop);
+
+ switch (type) {
+ case RTM_NEWNEXTHOP:
+ if (nexthop) {
+ nexthop->flags = tmp->flags;
+ nexthop_enter_configured(nexthop);
+ log_nexthop_debug(tmp, "Received remembered", link);
+ } else {
+ nexthop_enter_configured(tmp);
+ log_nexthop_debug(tmp, "Remembering", link);
+
+ r = nexthop_add(m, link, tmp);
+ if (r < 0) {
+ log_link_warning_errno(link, r, "Could not remember foreign nexthop, ignoring: %m");
+ return 0;
+ }
+
+ TAKE_PTR(tmp);
+ }
+
+ break;
+ case RTM_DELNEXTHOP:
+ if (nexthop) {
+ nexthop_enter_removed(nexthop);
+ if (nexthop->state == 0) {
+ log_nexthop_debug(nexthop, "Forgetting", link);
+ nexthop_free(nexthop);
+ } else
+ log_nexthop_debug(nexthop, "Removed", link);
+ } else
+ log_nexthop_debug(tmp, "Kernel removed unknown", link);
+ break;
+
+ default:
+ assert_not_reached();
+ }
+
+ return 1;
+}
+
+static int nexthop_section_verify(NextHop *nh) {
+ if (section_is_invalid(nh->section))
+ return -EINVAL;
+
+ if (!hashmap_isempty(nh->group)) {
+ if (in_addr_is_set(nh->family, &nh->gw))
+ return log_warning_errno(SYNTHETIC_ERRNO(EINVAL),
+ "%s: nexthop group cannot have gateway address. "
+ "Ignoring [NextHop] section from line %u.",
+ nh->section->filename, nh->section->line);
+
+ if (nh->family != AF_UNSPEC)
+ return log_warning_errno(SYNTHETIC_ERRNO(EINVAL),
+ "%s: nexthop group cannot have Family= setting. "
+ "Ignoring [NextHop] section from line %u.",
+ nh->section->filename, nh->section->line);
+
+ if (nh->blackhole && in_addr_is_set(nh->family, &nh->gw))
+ return log_warning_errno(SYNTHETIC_ERRNO(EINVAL),
+ "%s: nexthop group cannot be a blackhole. "
+ "Ignoring [NextHop] section from line %u.",
+ nh->section->filename, nh->section->line);
+ } else if (nh->family == AF_UNSPEC)
+ /* When neither Family=, Gateway=, nor Group= is specified, assume IPv4. */
+ nh->family = AF_INET;
+
+ if (nh->blackhole && in_addr_is_set(nh->family, &nh->gw))
+ return log_warning_errno(SYNTHETIC_ERRNO(EINVAL),
+ "%s: blackhole nexthop cannot have gateway address. "
+ "Ignoring [NextHop] section from line %u.",
+ nh->section->filename, nh->section->line);
+
+ if (nh->onlink < 0 && in_addr_is_set(nh->family, &nh->gw) &&
+ ordered_hashmap_isempty(nh->network->addresses_by_section)) {
+ /* If no address is configured, in most cases the gateway cannot be reachable.
+ * TODO: we may need to improve the condition above. */
+ log_warning("%s: Gateway= without static address configured. "
+ "Enabling OnLink= option.",
+ nh->section->filename);
+ nh->onlink = true;
+ }
+
+ if (nh->onlink >= 0)
+ SET_FLAG(nh->flags, RTNH_F_ONLINK, nh->onlink);
+
+ return 0;
+}
+
+void network_drop_invalid_nexthops(Network *network) {
+ NextHop *nh;
+
+ assert(network);
+
+ HASHMAP_FOREACH(nh, network->nexthops_by_section)
+ if (nexthop_section_verify(nh) < 0)
+ nexthop_free(nh);
+}
+
+int config_parse_nexthop_id(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ _cleanup_(nexthop_free_or_set_invalidp) NextHop *n = NULL;
+ Network *network = userdata;
+ uint32_t id;
+ int r;
+
+ assert(filename);
+ assert(section);
+ assert(lvalue);
+ assert(rvalue);
+ assert(data);
+
+ r = nexthop_new_static(network, filename, section_line, &n);
+ if (r < 0)
+ return log_oom();
+
+ if (isempty(rvalue)) {
+ n->id = 0;
+ TAKE_PTR(n);
+ return 0;
+ }
+
+ r = safe_atou32(rvalue, &id);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Could not parse nexthop id \"%s\", ignoring assignment: %m", rvalue);
+ return 0;
+ }
+ if (id == 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0,
+ "Invalid nexthop id \"%s\", ignoring assignment: %m", rvalue);
+ return 0;
+ }
+
+ n->id = id;
+ TAKE_PTR(n);
+ return 0;
+}
+
+int config_parse_nexthop_gateway(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ _cleanup_(nexthop_free_or_set_invalidp) NextHop *n = NULL;
+ Network *network = userdata;
+ int r;
+
+ assert(filename);
+ assert(section);
+ assert(lvalue);
+ assert(rvalue);
+ assert(data);
+
+ r = nexthop_new_static(network, filename, section_line, &n);
+ if (r < 0)
+ return log_oom();
+
+ if (isempty(rvalue)) {
+ n->family = AF_UNSPEC;
+ n->gw = IN_ADDR_NULL;
+
+ TAKE_PTR(n);
+ return 0;
+ }
+
+ r = in_addr_from_string_auto(rvalue, &n->family, &n->gw);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Invalid %s='%s', ignoring assignment: %m", lvalue, rvalue);
+ return 0;
+ }
+
+ TAKE_PTR(n);
+ return 0;
+}
+
+int config_parse_nexthop_family(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ _cleanup_(nexthop_free_or_set_invalidp) NextHop *n = NULL;
+ Network *network = userdata;
+ AddressFamily a;
+ int r;
+
+ assert(filename);
+ assert(section);
+ assert(lvalue);
+ assert(rvalue);
+ assert(data);
+
+ r = nexthop_new_static(network, filename, section_line, &n);
+ if (r < 0)
+ return log_oom();
+
+ if (isempty(rvalue) &&
+ !in_addr_is_set(n->family, &n->gw)) {
+ /* Accept an empty string only when Gateway= is null or not specified. */
+ n->family = AF_UNSPEC;
+ TAKE_PTR(n);
+ return 0;
+ }
+
+ a = nexthop_address_family_from_string(rvalue);
+ if (a < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0,
+ "Invalid %s='%s', ignoring assignment: %m", lvalue, rvalue);
+ return 0;
+ }
+
+ if (in_addr_is_set(n->family, &n->gw) &&
+ ((a == ADDRESS_FAMILY_IPV4 && n->family == AF_INET6) ||
+ (a == ADDRESS_FAMILY_IPV6 && n->family == AF_INET))) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0,
+ "Specified family '%s' conflicts with the family of the previously specified Gateway=, "
+ "ignoring assignment.", rvalue);
+ return 0;
+ }
+
+ switch (a) {
+ case ADDRESS_FAMILY_IPV4:
+ n->family = AF_INET;
+ break;
+ case ADDRESS_FAMILY_IPV6:
+ n->family = AF_INET6;
+ break;
+ default:
+ assert_not_reached();
+ }
+
+ TAKE_PTR(n);
+ return 0;
+}
+
+int config_parse_nexthop_onlink(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ _cleanup_(nexthop_free_or_set_invalidp) NextHop *n = NULL;
+ Network *network = userdata;
+ int r;
+
+ assert(filename);
+ assert(section);
+ assert(lvalue);
+ assert(rvalue);
+ assert(data);
+
+ r = nexthop_new_static(network, filename, section_line, &n);
+ if (r < 0)
+ return log_oom();
+
+ if (isempty(rvalue)) {
+ n->onlink = -1;
+ TAKE_PTR(n);
+ return 0;
+ }
+
+ r = parse_boolean(rvalue);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Failed to parse %s=, ignoring assignment: %s", lvalue, rvalue);
+ return 0;
+ }
+
+ n->onlink = r;
+
+ TAKE_PTR(n);
+ return 0;
+}
+
+int config_parse_nexthop_blackhole(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ _cleanup_(nexthop_free_or_set_invalidp) NextHop *n = NULL;
+ Network *network = userdata;
+ int r;
+
+ assert(filename);
+ assert(section);
+ assert(lvalue);
+ assert(rvalue);
+ assert(data);
+
+ r = nexthop_new_static(network, filename, section_line, &n);
+ if (r < 0)
+ return log_oom();
+
+ r = parse_boolean(rvalue);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Failed to parse %s=, ignoring assignment: %s", lvalue, rvalue);
+ return 0;
+ }
+
+ n->blackhole = r;
+
+ TAKE_PTR(n);
+ return 0;
+}
+
+int config_parse_nexthop_group(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ _cleanup_(nexthop_free_or_set_invalidp) NextHop *n = NULL;
+ Network *network = userdata;
+ int r;
+
+ assert(filename);
+ assert(section);
+ assert(lvalue);
+ assert(rvalue);
+ assert(data);
+
+ r = nexthop_new_static(network, filename, section_line, &n);
+ if (r < 0)
+ return log_oom();
+
+ if (isempty(rvalue)) {
+ n->group = hashmap_free_free(n->group);
+ TAKE_PTR(n);
+ return 0;
+ }
+
+ for (const char *p = rvalue;;) {
+ _cleanup_free_ struct nexthop_grp *nhg = NULL;
+ _cleanup_free_ char *word = NULL;
+ uint32_t w;
+ char *sep;
+
+ r = extract_first_word(&p, &word, NULL, 0);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Invalid %s=, ignoring assignment: %s", lvalue, rvalue);
+ return 0;
+ }
+ if (r == 0)
+ break;
+
+ nhg = new0(struct nexthop_grp, 1);
+ if (!nhg)
+ return log_oom();
+
+ sep = strchr(word, ':');
+ if (sep) {
+ *sep++ = '\0';
+ r = safe_atou32(sep, &w);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Failed to parse weight for nexthop group, ignoring assignment: %s:%s",
+ word, sep);
+ continue;
+ }
+ if (w == 0 || w > 256) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0,
+ "Invalid weight for nexthop group, ignoring assignment: %s:%s",
+ word, sep);
+ continue;
+ }
+ /* See comments in config_parse_multipath_route(). */
+ nhg->weight = w - 1;
+ }
+
+ r = safe_atou32(word, &nhg->id);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Failed to parse nexthop ID in %s=, ignoring assignment: %s%s%s",
+ lvalue, word, sep ? ":" : "", strempty(sep));
+ continue;
+ }
+ if (nhg->id == 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0,
+ "Nexthop ID in %s= must be positive, ignoring assignment: %s%s%s",
+ lvalue, word, sep ? ":" : "", strempty(sep));
+ continue;
+ }
+
+ r = hashmap_ensure_put(&n->group, NULL, UINT32_TO_PTR(nhg->id), nhg);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r == -EEXIST) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Nexthop ID %"PRIu32" is specified multiple times in %s=, ignoring assignment: %s%s%s",
+ nhg->id, lvalue, word, sep ? ":" : "", strempty(sep));
+ continue;
+ }
+ assert(r > 0);
+ TAKE_PTR(nhg);
+ }
+
+ TAKE_PTR(n);
+ return 0;
+}