/* * netifd - network interface daemon * Copyright (C) 2012 Felix Fietkau * Copyright (C) 2012 Steven Barth * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 * as published by the Free Software Foundation * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. */ #include #include #include #include #include #include #include "netifd.h" #include "device.h" #include "interface.h" #include "interface-ip.h" #include "proto.h" #include "ubus.h" #include "system.h" enum { ROUTE_INTERFACE, ROUTE_TARGET, ROUTE_MASK, ROUTE_GATEWAY, ROUTE_METRIC, ROUTE_MTU, ROUTE_VALID, ROUTE_TABLE, ROUTE_SOURCE, ROUTE_ONLINK, ROUTE_TYPE, __ROUTE_MAX }; static const struct blobmsg_policy route_attr[__ROUTE_MAX] = { [ROUTE_INTERFACE] = { .name = "interface", .type = BLOBMSG_TYPE_STRING }, [ROUTE_TARGET] = { .name = "target", .type = BLOBMSG_TYPE_STRING }, [ROUTE_MASK] = { .name = "netmask", .type = BLOBMSG_TYPE_STRING }, [ROUTE_GATEWAY] = { .name = "gateway", .type = BLOBMSG_TYPE_STRING }, [ROUTE_METRIC] = { .name = "metric", .type = BLOBMSG_TYPE_INT32 }, [ROUTE_MTU] = { .name = "mtu", .type = BLOBMSG_TYPE_INT32 }, [ROUTE_TABLE] = { .name = "table", .type = BLOBMSG_TYPE_STRING }, [ROUTE_VALID] = { .name = "valid", .type = BLOBMSG_TYPE_INT32 }, [ROUTE_SOURCE] = { .name = "source", .type = BLOBMSG_TYPE_STRING }, [ROUTE_ONLINK] = { .name = "onlink", .type = BLOBMSG_TYPE_BOOL }, [ROUTE_TYPE] = { .name = "type", .type = BLOBMSG_TYPE_STRING } }; const struct uci_blob_param_list route_attr_list = { .n_params = __ROUTE_MAX, .params = route_attr, }; struct list_head prefixes = LIST_HEAD_INIT(prefixes); static struct device_prefix *ula_prefix = NULL; static struct uloop_timeout valid_until_timeout; static void clear_if_addr(union if_addr *a, int mask) { int m_bytes = (mask + 7) / 8; uint8_t m_clear = (1 << (m_bytes * 8 - mask)) - 1; uint8_t *p = (uint8_t *) a; if (m_bytes < sizeof(a)) memset(p + m_bytes, 0, sizeof(a) - m_bytes); p[m_bytes - 1] &= ~m_clear; } static bool match_if_addr(union if_addr *a1, union if_addr *a2, int mask) { union if_addr *p1, *p2; p1 = alloca(sizeof(*a1)); p2 = alloca(sizeof(*a2)); memcpy(p1, a1, sizeof(*a1)); clear_if_addr(p1, mask); memcpy(p2, a2, sizeof(*a2)); clear_if_addr(p2, mask); return !memcmp(p1, p2, sizeof(*p1)); } static int set_ip_source_policy(bool add, bool v6, unsigned int priority, const union if_addr *addr, uint8_t mask, unsigned int table, struct interface *in_iface, const char *action) { struct iprule rule = { .flags = IPRULE_PRIORITY, .priority = priority }; if (addr) { rule.flags |= IPRULE_SRC; rule.src_addr = *addr; rule.src_mask = mask; } if (table) { rule.flags |= IPRULE_LOOKUP; rule.lookup = table; if (!rule.lookup) return 0; } else if (action) { rule.flags |= IPRULE_ACTION; system_resolve_iprule_action(action, &rule.action); } if (in_iface && in_iface->l3_dev.dev) { rule.flags |= IPRULE_IN; strcpy(rule.in_dev, in_iface->l3_dev.dev->ifname); } rule.flags |= (v6) ? IPRULE_INET6 : IPRULE_INET4; return (add) ? system_add_iprule(&rule) : system_del_iprule(&rule); } static int set_ip_lo_policy(bool add, bool v6, struct interface *iface) { struct iprule rule = { .flags = IPRULE_IN | IPRULE_LOOKUP | IPRULE_PRIORITY, .priority = IPRULE_PRIORITY_NW + iface->l3_dev.dev->ifindex, .lookup = (v6) ? iface->ip6table : iface->ip4table, .in_dev = "lo" }; if (!rule.lookup) return 0; rule.flags |= (v6) ? IPRULE_INET6 : IPRULE_INET4; return (add) ? system_add_iprule(&rule) : system_del_iprule(&rule); } static bool __find_ip_addr_target(struct interface_ip_settings *ip, union if_addr *a, bool v6) { struct device_addr *addr; vlist_for_each_element(&ip->addr, addr, node) { if (!addr->enabled) continue; if (v6 != ((addr->flags & DEVADDR_FAMILY) == DEVADDR_INET6)) continue; // Handle offlink addresses correctly unsigned int mask = addr->mask; if ((addr->flags & DEVADDR_FAMILY) == DEVADDR_INET6 && (addr->flags & DEVADDR_OFFLINK)) mask = 128; if (!match_if_addr(&addr->addr, a, mask)) continue; return true; } return false; } static void __find_ip_route_target(struct interface_ip_settings *ip, union if_addr *a, bool v6, struct device_route **res) { struct device_route *route; vlist_for_each_element(&ip->route, route, node) { if (!route->enabled) continue; if (v6 != ((route->flags & DEVADDR_FAMILY) == DEVADDR_INET6)) continue; if (!match_if_addr(&route->addr, a, route->mask)) continue; if (route->flags & DEVROUTE_TABLE) continue; if (!*res || route->mask < (*res)->mask) *res = route; } } static bool interface_ip_find_addr_target(struct interface *iface, union if_addr *a, bool v6) { return __find_ip_addr_target(&iface->proto_ip, a, v6) || __find_ip_addr_target(&iface->config_ip, a, v6); } static void interface_ip_find_route_target(struct interface *iface, union if_addr *a, bool v6, struct device_route **route) { __find_ip_route_target(&iface->proto_ip, a, v6, route); __find_ip_route_target(&iface->config_ip, a, v6, route); } struct interface * interface_ip_add_target_route(union if_addr *addr, bool v6, struct interface *iface) { struct device_route *route, *r_next = NULL; bool defaultroute_target = false; int addrsize = v6 ? sizeof(addr->in6) : sizeof(addr->in); route = calloc(1, sizeof(*route)); if (!route) return NULL; route->flags = v6 ? DEVADDR_INET6 : DEVADDR_INET4; route->mask = v6 ? 128 : 32; if (memcmp(&route->addr, addr, addrsize) == 0) defaultroute_target = true; else memcpy(&route->addr, addr, addrsize); if (iface) { /* look for locally addressable target first */ if (interface_ip_find_addr_target(iface, addr, v6)) goto done; /* do not stop at the first route, let the lookup compare * masks to find the best match */ interface_ip_find_route_target(iface, addr, v6, &r_next); } else { vlist_for_each_element(&interfaces, iface, node) { /* look for locally addressable target first */ if (interface_ip_find_addr_target(iface, addr, v6)) goto done; /* do not stop at the first route, let the lookup compare * masks to find the best match */ interface_ip_find_route_target(iface, addr, v6, &r_next); } } if (!r_next) { free(route); return NULL; } iface = r_next->iface; memcpy(&route->nexthop, &r_next->nexthop, sizeof(route->nexthop)); route->mtu = r_next->mtu; route->metric = r_next->metric; route->table = r_next->table; done: route->iface = iface; if (defaultroute_target) free(route); else vlist_add(&iface->host_routes, &route->node, route); return iface; } void interface_ip_add_route(struct interface *iface, struct blob_attr *attr, bool v6) { struct interface_ip_settings *ip; struct blob_attr *tb[__ROUTE_MAX], *cur; struct device_route *route; int af = v6 ? AF_INET6 : AF_INET; bool is_proto_route = !!iface; blobmsg_parse(route_attr, __ROUTE_MAX, tb, blobmsg_data(attr), blobmsg_data_len(attr)); if (!iface) { if ((cur = tb[ROUTE_INTERFACE]) == NULL) return; iface = vlist_find(&interfaces, blobmsg_data(cur), iface, node); if (!iface) return; ip = &iface->config_ip; } else { ip = &iface->proto_ip; } route = calloc(1, sizeof(*route)); if (!route) return; route->flags = v6 ? DEVADDR_INET6 : DEVADDR_INET4; route->mask = v6 ? 128 : 32; if ((cur = tb[ROUTE_MASK]) != NULL) { route->mask = parse_netmask_string(blobmsg_data(cur), v6); if (route->mask > (v6 ? 128 : 32)) goto error; } if ((cur = tb[ROUTE_TARGET]) != NULL) { if (!parse_ip_and_netmask(af, blobmsg_data(cur), &route->addr, &route->mask)) { DPRINTF("Failed to parse route target: %s\n", (char *) blobmsg_data(cur)); goto error; } } if ((cur = tb[ROUTE_GATEWAY]) != NULL) { if (!inet_pton(af, blobmsg_data(cur), &route->nexthop)) { DPRINTF("Failed to parse route gateway: %s\n", (char *) blobmsg_data(cur)); goto error; } } if ((cur = tb[ROUTE_METRIC]) != NULL) { route->metric = blobmsg_get_u32(cur); route->flags |= DEVROUTE_METRIC; } else route->metric = iface->metric; if ((cur = tb[ROUTE_MTU]) != NULL) { route->mtu = blobmsg_get_u32(cur); route->flags |= DEVROUTE_MTU; } // Use source-based routing if ((cur = tb[ROUTE_SOURCE]) != NULL) { char *saveptr, *source = alloca(blobmsg_data_len(cur)); memcpy(source, blobmsg_data(cur), blobmsg_data_len(cur)); const char *addr = strtok_r(source, "/", &saveptr); const char *mask = strtok_r(NULL, "/", &saveptr); if (!addr || inet_pton(af, addr, &route->source) < 1) { DPRINTF("Failed to parse route source: %s\n", addr); goto error; } route->sourcemask = (mask) ? atoi(mask) : ((af == AF_INET6) ? 128 : 32); } if ((cur = tb[ROUTE_ONLINK]) != NULL && blobmsg_get_bool(cur)) route->flags |= DEVROUTE_ONLINK; if (is_proto_route) { route->table = (v6) ? iface->ip6table : iface->ip4table; route->flags |= DEVROUTE_SRCTABLE; } if ((cur = tb[ROUTE_TABLE]) != NULL) { if (!system_resolve_rt_table(blobmsg_data(cur), &route->table)) { DPRINTF("Failed to resolve routing table: %s\n", (char *) blobmsg_data(cur)); goto error; } if (route->table) route->flags |= DEVROUTE_TABLE; } if ((cur = tb[ROUTE_VALID]) != NULL) { int64_t valid = blobmsg_get_u32(cur); int64_t valid_until = valid + (int64_t)system_get_rtime(); if (valid_until <= LONG_MAX && valid != 0xffffffffLL) // Catch overflow route->valid_until = valid_until; } if ((cur = tb[ROUTE_TYPE]) != NULL) { if (!system_resolve_rt_type(blobmsg_data(cur), &route->type)) { DPRINTF("Failed to resolve routing type: %s\n", (char *) blobmsg_data(cur)); goto error; } route->flags |= DEVROUTE_TYPE; } vlist_add(&ip->route, &route->node, route); return; error: free(route); } static int addr_cmp(const void *k1, const void *k2, void *ptr) { return memcmp(k1, k2, sizeof(struct device_addr) - offsetof(struct device_addr, flags)); } static int route_cmp(const void *k1, const void *k2, void *ptr) { const struct device_route *r1 = k1, *r2 = k2; if (r1->mask != r2->mask) return r2->mask - r1->mask; if (r1->metric != r2->metric) return r1->metric - r2->metric; if (r1->flags != r2->flags) return r2->flags - r1->flags; if (r1->sourcemask != r2->sourcemask) return r1->sourcemask - r2->sourcemask; if (r1->table != r2->table) return r1->table - r2->table; int maskcmp = memcmp(&r1->source, &r2->source, sizeof(r1->source)); if (maskcmp) return maskcmp; return memcmp(&r1->addr, &r2->addr, sizeof(r1->addr)); } static int prefix_cmp(const void *k1, const void *k2, void *ptr) { return memcmp(k1, k2, offsetof(struct device_prefix, pclass) - offsetof(struct device_prefix, addr)); } static void interface_handle_subnet_route(struct interface *iface, struct device_addr *addr, bool add) { struct device *dev = iface->l3_dev.dev; struct device_route route; memset(&route, 0, sizeof(route)); route.iface = iface; route.flags = addr->flags; route.mask = addr->mask; memcpy(&route.addr, &addr->addr, sizeof(route.addr)); clear_if_addr(&route.addr, route.mask); if (add) { route.flags |= DEVADDR_KERNEL; system_del_route(dev, &route); if (!(addr->flags & DEVADDR_OFFLINK)) { route.flags &= ~DEVADDR_KERNEL; route.metric = iface->metric; system_add_route(dev, &route); } } else { if (!(addr->flags & DEVADDR_OFFLINK)) system_del_route(dev, &route); } } static void interface_update_proto_addr(struct vlist_tree *tree, struct vlist_node *node_new, struct vlist_node *node_old) { struct interface_ip_settings *ip; struct interface *iface; struct device *dev; struct device_addr *a_new = NULL, *a_old = NULL; bool replace = false; bool keep = false; bool v6 = false; ip = container_of(tree, struct interface_ip_settings, addr); iface = ip->iface; dev = iface->l3_dev.dev; if (!node_new || !node_old) iface->updated |= IUF_ADDRESS; if (node_new) { a_new = container_of(node_new, struct device_addr, node); if ((a_new->flags & DEVADDR_FAMILY) == DEVADDR_INET4 && !a_new->broadcast) { uint32_t mask = ~0; uint32_t *a = (uint32_t *) &a_new->addr; mask >>= a_new->mask; a_new->broadcast = *a | htonl(mask); } } if (node_old) a_old = container_of(node_old, struct device_addr, node); if (a_new && a_old) { keep = true; if (a_old->flags != a_new->flags || a_old->failed) keep = false; if (a_old->valid_until != a_new->valid_until || a_old->preferred_until != a_new->preferred_until) replace = true; if ((a_new->flags & DEVADDR_FAMILY) == DEVADDR_INET4 && a_new->broadcast != a_old->broadcast) keep = false; } if (node_old) { if (!(a_old->flags & DEVADDR_EXTERNAL) && a_old->enabled && !keep) { interface_handle_subnet_route(iface, a_old, false); if ((a_old->flags & DEVADDR_FAMILY) == DEVADDR_INET6) v6 = true; unsigned int table = (v6) ? iface->ip6table : iface->ip4table; //This is needed for source routing to work correctly. If a device //has two connections to a network using the same subnet, adding //only the network-rule will cause packets to be routed through the //first matching network (source IP matches both masks). if (table) { set_ip_source_policy(false, v6, IPRULE_PRIORITY_ADDR, &a_old->addr, (v6) ? 128 : 32, table, NULL, NULL); set_ip_source_policy(false, v6, IPRULE_PRIORITY_NW, &a_old->addr, a_old->mask, table, NULL, NULL); } system_del_address(dev, a_old); } free(a_old->pclass); free(a_old); } if (node_new) { a_new->enabled = true; if (!(a_new->flags & DEVADDR_EXTERNAL) && (!keep || replace)) { if (system_add_address(dev, a_new)) a_new->failed = true; if (!keep) { if ((a_new->flags & DEVADDR_FAMILY) == DEVADDR_INET6) v6 = true; unsigned int table = (v6) ? iface->ip6table : iface->ip4table; if (table) { set_ip_source_policy(true, v6, IPRULE_PRIORITY_ADDR, &a_new->addr, (v6) ? 128 : 32, table, NULL, NULL); set_ip_source_policy(true, v6, IPRULE_PRIORITY_NW, &a_new->addr, a_new->mask, table, NULL, NULL); } } if ((a_new->flags & DEVADDR_OFFLINK) || iface->metric) interface_handle_subnet_route(iface, a_new, true); } } } static bool enable_route(struct interface_ip_settings *ip, struct device_route *route) { if (ip->no_defaultroute && !route->mask) return false; return ip->enabled; } static void interface_update_proto_route(struct vlist_tree *tree, struct vlist_node *node_new, struct vlist_node *node_old) { struct interface_ip_settings *ip; struct interface *iface; struct device *dev; struct device_route *route_old, *route_new; bool keep = false; ip = container_of(tree, struct interface_ip_settings, route); iface = ip->iface; dev = iface->l3_dev.dev; if (!node_new || !node_old) iface->updated |= IUF_ROUTE; route_old = container_of(node_old, struct device_route, node); route_new = container_of(node_new, struct device_route, node); if (node_old && node_new) keep = !memcmp(&route_old->nexthop, &route_new->nexthop, sizeof(route_old->nexthop)) && (route_old->mtu == route_new->mtu) && (route_old->type == route_new->type) && !route_old->failed; if (node_old) { if (!(route_old->flags & DEVADDR_EXTERNAL) && route_old->enabled && !keep) system_del_route(dev, route_old); free(route_old); } if (node_new) { bool _enabled = enable_route(ip, route_new); if (!(route_new->flags & DEVADDR_EXTERNAL) && !keep && _enabled) if (system_add_route(dev, route_new)) route_new->failed = true; route_new->iface = iface; route_new->enabled = _enabled; } } static void interface_update_host_route(struct vlist_tree *tree, struct vlist_node *node_new, struct vlist_node *node_old) { struct interface *iface; struct device *dev; struct device_route *route_old, *route_new; iface = container_of(tree, struct interface, host_routes); dev = iface->l3_dev.dev; route_old = container_of(node_old, struct device_route, node); route_new = container_of(node_new, struct device_route, node); if (node_old) { system_del_route(dev, route_old); free(route_old); } if (node_new) { if (system_add_route(dev, route_new)) route_new->failed = true; } } static void interface_set_prefix_address(struct device_prefix_assignment *assignment, const struct device_prefix *prefix, struct interface *iface, bool add) { const struct interface *uplink = prefix->iface; if (!iface->l3_dev.dev) return; struct device *l3_downlink = iface->l3_dev.dev; struct device_addr addr; memset(&addr, 0, sizeof(addr)); addr.addr.in6 = prefix->addr; addr.addr.in6.s6_addr32[1] |= htonl(assignment->assigned); addr.addr.in6.s6_addr[15] += 1; addr.mask = assignment->length; addr.flags = DEVADDR_INET6; addr.preferred_until = prefix->preferred_until; addr.valid_until = prefix->valid_until; if (!add && assignment->enabled) { time_t now = system_get_rtime(); addr.preferred_until = now; if (!addr.valid_until || addr.valid_until - now > 7200) addr.valid_until = now + 7200; system_del_address(l3_downlink, &addr); // Work around dangling prefix routes system_add_address(l3_downlink, &addr); if (prefix->iface) { if (prefix->iface->ip6table) set_ip_source_policy(false, true, IPRULE_PRIORITY_NW, &addr.addr, addr.mask, prefix->iface->ip6table, iface, NULL); set_ip_source_policy(false, true, IPRULE_PRIORITY_REJECT, &addr.addr, addr.mask, 0, iface, "unreachable"); } assignment->enabled = false; } else if (add && (iface->state == IFS_UP || iface->state == IFS_SETUP) && !system_add_address(l3_downlink, &addr)) { if (prefix->iface && !assignment->enabled) { set_ip_source_policy(true, true, IPRULE_PRIORITY_REJECT, &addr.addr, addr.mask, 0, iface, "unreachable"); if (prefix->iface->ip6table) set_ip_source_policy(true, true, IPRULE_PRIORITY_NW, &addr.addr, addr.mask, prefix->iface->ip6table, iface, NULL); } if (uplink && uplink->l3_dev.dev) { int mtu = system_update_ipv6_mtu( uplink->l3_dev.dev, 0); if (mtu > 0) system_update_ipv6_mtu(l3_downlink, mtu); } assignment->enabled = true; } } static bool interface_prefix_assign(struct list_head *list, struct device_prefix_assignment *assign) { int32_t current = 0, asize = (1 << (64 - assign->length)) - 1; struct device_prefix_assignment *c; list_for_each_entry(c, list, head) { if (assign->assigned != -1) { if (assign->assigned >= current && assign->assigned + asize < c->assigned) { list_add_tail(&assign->head, &c->head); return true; } } else if (assign->assigned == -1) { current = (current + asize) & (~asize); if (current + asize < c->assigned) { assign->assigned = current; list_add_tail(&assign->head, &c->head); return true; } } current = (c->assigned + (1 << (64 - c->length))); } return false; } static void interface_update_prefix_assignments(struct device_prefix *prefix, bool setup) { struct device_prefix_assignment *c; struct interface *iface; // Delete all assignments while (!list_empty(&prefix->assignments)) { c = list_first_entry(&prefix->assignments, struct device_prefix_assignment, head); if ((iface = vlist_find(&interfaces, c->name, iface, node))) interface_set_prefix_address(c, prefix, iface, false); list_del(&c->head); free(c); } if (!setup) return; // End-of-assignment sentinel c = malloc(sizeof(*c) + 1); c->assigned = 1 << (64 - prefix->length); c->length = 64; c->name[0] = 0; list_add(&c->head, &prefix->assignments); // Excluded prefix if (prefix->excl_length > 0) { const char name[] = "!excluded"; c = malloc(sizeof(*c) + sizeof(name)); c->assigned = ntohl(prefix->excl_addr.s6_addr32[1]) & ((1 << (64 - prefix->length)) - 1); c->length = prefix->excl_length; memcpy(c->name, name, sizeof(name)); list_add(&c->head, &prefix->assignments); } bool assigned_any = false; struct list_head assign_later = LIST_HEAD_INIT(assign_later); vlist_for_each_element(&interfaces, iface, node) { if (iface->assignment_length < 48 || iface->assignment_length > 64) continue; // Test whether there is a matching class if (!list_empty(&iface->assignment_classes)) { bool found = false; struct interface_assignment_class *c; list_for_each_entry(c, &iface->assignment_classes, head) { if (!strcmp(c->name, prefix->pclass)) { found = true; break; } } if (!found) continue; } size_t namelen = strlen(iface->name) + 1; c = malloc(sizeof(*c) + namelen); c->length = iface->assignment_length; c->assigned = iface->assignment_hint; c->enabled = false; memcpy(c->name, iface->name, namelen); // First process all custom assignments, put all others in later-list if (c->assigned == -1 || !interface_prefix_assign(&prefix->assignments, c)) { if (c->assigned != -1) { c->assigned = -1; netifd_log_message(L_WARNING, "Failed to assign requested subprefix " "of size %hhu for %s, trying other\n", c->length, c->name); } struct list_head *next = &assign_later; struct device_prefix_assignment *n; list_for_each_entry(n, &assign_later, head) { if (n->length < c->length) { next = &n->head; break; } } list_add_tail(&c->head, next); } if (c->assigned != -1) assigned_any = true; } // Then try to assign all other + failed custom assignments while (!list_empty(&assign_later)) { c = list_first_entry(&assign_later, struct device_prefix_assignment, head); list_del(&c->head); bool assigned = false; do { assigned = interface_prefix_assign(&prefix->assignments, c); } while (!assigned && ++c->length <= 64); if (!assigned) { netifd_log_message(L_WARNING, "Failed to assign subprefix " "of size %hhu for %s\n", c->length, c->name); free(c); } else { assigned_any = true; } } list_for_each_entry(c, &prefix->assignments, head) if ((iface = vlist_find(&interfaces, c->name, iface, node))) interface_set_prefix_address(c, prefix, iface, true); if (!assigned_any) netifd_log_message(L_WARNING, "You have delegated IPv6-prefixes but haven't assigned them " "to any interface. Did you forget to set option ip6assign on your lan-interfaces?"); } void interface_refresh_assignments(bool hint) { static bool refresh = false; if (!hint && refresh) { struct device_prefix *p; list_for_each_entry(p, &prefixes, head) interface_update_prefix_assignments(p, true); } refresh = hint; } static void interface_update_prefix(struct vlist_tree *tree, struct vlist_node *node_new, struct vlist_node *node_old) { struct device_prefix *prefix_old, *prefix_new; prefix_old = container_of(node_old, struct device_prefix, node); prefix_new = container_of(node_new, struct device_prefix, node); struct interface_ip_settings *ip = container_of(tree, struct interface_ip_settings, prefix); if (tree && (!node_new || !node_old)) ip->iface->updated |= IUF_PREFIX; struct device_route route; memset(&route, 0, sizeof(route)); route.flags = DEVADDR_INET6; route.metric = INT32_MAX; route.mask = (node_new) ? prefix_new->length : prefix_old->length; route.addr.in6 = (node_new) ? prefix_new->addr : prefix_old->addr; struct device_prefix_assignment *c; struct interface *iface; if (node_old && node_new) { // Move assignments and refresh addresses to update valid times list_splice(&prefix_old->assignments, &prefix_new->assignments); list_for_each_entry(c, &prefix_new->assignments, head) if ((iface = vlist_find(&interfaces, c->name, iface, node))) interface_set_prefix_address(c, prefix_new, iface, true); } else if (node_new) { // Set null-route to avoid routing loops system_add_route(NULL, &route); if (!prefix_new->iface || !prefix_new->iface->proto_ip.no_delegation) interface_update_prefix_assignments(prefix_new, true); } else if (node_old) { // Remove null-route interface_update_prefix_assignments(prefix_old, false); system_del_route(NULL, &route); } if (node_old) { if (prefix_old->head.next) list_del(&prefix_old->head); free(prefix_old); } if (node_new && (!prefix_new->iface || !prefix_new->iface->proto_ip.no_delegation)) list_add(&prefix_new->head, &prefixes); } struct device_prefix* interface_ip_add_device_prefix(struct interface *iface, struct in6_addr *addr, uint8_t length, time_t valid_until, time_t preferred_until, struct in6_addr *excl_addr, uint8_t excl_length, const char *pclass) { if (!pclass) pclass = (iface) ? iface->name : "local"; struct device_prefix *prefix = calloc(1, sizeof(*prefix) + strlen(pclass) + 1); prefix->length = length; prefix->addr = *addr; prefix->preferred_until = preferred_until; prefix->valid_until = valid_until; prefix->iface = iface; INIT_LIST_HEAD(&prefix->assignments); if (excl_addr) { prefix->excl_addr = *excl_addr; prefix->excl_length = excl_length; } strcpy(prefix->pclass, pclass); if (iface) vlist_add(&iface->proto_ip.prefix, &prefix->node, &prefix->addr); else interface_update_prefix(NULL, &prefix->node, NULL); return prefix; } void interface_ip_set_ula_prefix(const char *prefix) { char buf[INET6_ADDRSTRLEN + 4] = {0}, *saveptr; if (prefix) strncpy(buf, prefix, sizeof(buf) - 1); char *prefixaddr = strtok_r(buf, "/", &saveptr); struct in6_addr addr; if (!prefixaddr || inet_pton(AF_INET6, prefixaddr, &addr) < 1) { if (ula_prefix) { interface_update_prefix(NULL, NULL, &ula_prefix->node); ula_prefix = NULL; } return; } int length; char *prefixlen = strtok_r(NULL, ",", &saveptr); if (!prefixlen || (length = atoi(prefixlen)) < 1 || length > 64) return; if (!ula_prefix || !IN6_ARE_ADDR_EQUAL(&addr, &ula_prefix->addr) || ula_prefix->length != length) { if (ula_prefix) interface_update_prefix(NULL, NULL, &ula_prefix->node); ula_prefix = interface_ip_add_device_prefix(NULL, &addr, length, 0, 0, NULL, 0, NULL); } } void interface_add_dns_server(struct interface_ip_settings *ip, const char *str) { struct dns_server *s; s = calloc(1, sizeof(*s)); if (!s) return; s->af = AF_INET; if (inet_pton(s->af, str, &s->addr.in)) goto add; s->af = AF_INET6; if (inet_pton(s->af, str, &s->addr.in)) goto add; free(s); return; add: D(INTERFACE, "Add IPv%c DNS server: %s\n", s->af == AF_INET6 ? '6' : '4', str); vlist_simple_add(&ip->dns_servers, &s->node); } void interface_add_dns_server_list(struct interface_ip_settings *ip, struct blob_attr *list) { struct blob_attr *cur; int rem; blobmsg_for_each_attr(cur, list, rem) { if (blobmsg_type(cur) != BLOBMSG_TYPE_STRING) continue; if (!blobmsg_check_attr(cur, NULL)) continue; interface_add_dns_server(ip, blobmsg_data(cur)); } } static void interface_add_dns_search_domain(struct interface_ip_settings *ip, const char *str) { struct dns_search_domain *s; int len = strlen(str); s = calloc(1, sizeof(*s) + len + 1); if (!s) return; D(INTERFACE, "Add DNS search domain: %s\n", str); memcpy(s->name, str, len); vlist_simple_add(&ip->dns_search, &s->node); } void interface_add_dns_search_list(struct interface_ip_settings *ip, struct blob_attr *list) { struct blob_attr *cur; int rem; blobmsg_for_each_attr(cur, list, rem) { if (blobmsg_type(cur) != BLOBMSG_TYPE_STRING) continue; if (!blobmsg_check_attr(cur, NULL)) continue; interface_add_dns_search_domain(ip, blobmsg_data(cur)); } } static void write_resolv_conf_entries(FILE *f, struct interface_ip_settings *ip, const char *dev) { struct dns_server *s; struct dns_search_domain *d; const char *str; char buf[INET6_ADDRSTRLEN]; vlist_simple_for_each_element(&ip->dns_servers, s, node) { str = inet_ntop(s->af, &s->addr, buf, sizeof(buf)); if (!str) continue; if (s->af == AF_INET6 && IN6_IS_ADDR_LINKLOCAL(&s->addr.in6)) fprintf(f, "nameserver %s%%%s\n", str, dev); else fprintf(f, "nameserver %s\n", str); } vlist_simple_for_each_element(&ip->dns_search, d, node) { fprintf(f, "search %s\n", d->name); } } void interface_write_resolv_conf(void) { struct interface *iface; char *path = alloca(strlen(resolv_conf) + 5); FILE *f; uint32_t crcold, crcnew; sprintf(path, "%s.tmp", resolv_conf); unlink(path); f = fopen(path, "w+"); if (!f) { D(INTERFACE, "Failed to open %s for writing\n", path); return; } vlist_for_each_element(&interfaces, iface, node) { if (iface->state != IFS_UP) continue; if (vlist_simple_empty(&iface->proto_ip.dns_search) && vlist_simple_empty(&iface->proto_ip.dns_servers) && vlist_simple_empty(&iface->config_ip.dns_search) && vlist_simple_empty(&iface->config_ip.dns_servers)) continue; fprintf(f, "# Interface %s\n", iface->name); write_resolv_conf_entries(f, &iface->config_ip, iface->ifname); if (!iface->proto_ip.no_dns) write_resolv_conf_entries(f, &iface->proto_ip, iface->ifname); } fflush(f); rewind(f); crcnew = crc32_file(f); fclose(f); crcold = crcnew + 1; f = fopen(resolv_conf, "r"); if (f) { crcold = crc32_file(f); fclose(f); } if (crcold == crcnew) { unlink(path); } else if (rename(path, resolv_conf) < 0) { D(INTERFACE, "Failed to replace %s\n", resolv_conf); unlink(path); } } void interface_ip_set_enabled(struct interface_ip_settings *ip, bool enabled) { struct device_addr *addr; struct device_route *route; struct device *dev; ip->enabled = enabled; dev = ip->iface->l3_dev.dev; if (!dev) return; vlist_for_each_element(&ip->addr, addr, node) { if (addr->enabled == enabled) continue; if (enabled) system_add_address(dev, addr); else system_del_address(dev, addr); addr->enabled = enabled; } vlist_for_each_element(&ip->route, route, node) { bool _enabled = enabled; if (!enable_route(ip, route)) _enabled = false; if (route->enabled == _enabled) continue; if (_enabled) { if (!(route->flags & DEVROUTE_METRIC)) route->metric = ip->iface->metric; if (system_add_route(dev, route)) route->failed = true; } else system_del_route(dev, route); route->enabled = _enabled; } struct device_prefix *c; struct device_prefix_assignment *a; list_for_each_entry(c, &prefixes, head) list_for_each_entry(a, &c->assignments, head) if (!strcmp(a->name, ip->iface->name)) interface_set_prefix_address(a, c, ip->iface, enabled); if (ip->iface && ip->iface->l3_dev.dev) { set_ip_lo_policy(enabled, true, ip->iface); set_ip_lo_policy(enabled, false, ip->iface); set_ip_source_policy(enabled, true, IPRULE_PRIORITY_REJECT + ip->iface->l3_dev.dev->ifindex, NULL, 0, 0, ip->iface, "failed_policy"); } } void interface_ip_update_start(struct interface_ip_settings *ip) { if (ip != &ip->iface->config_ip) { vlist_simple_update(&ip->dns_servers); vlist_simple_update(&ip->dns_search); } vlist_update(&ip->route); vlist_update(&ip->addr); vlist_update(&ip->prefix); } void interface_ip_update_complete(struct interface_ip_settings *ip) { vlist_simple_flush(&ip->dns_servers); vlist_simple_flush(&ip->dns_search); vlist_flush(&ip->route); vlist_flush(&ip->addr); vlist_flush(&ip->prefix); interface_write_resolv_conf(); } void interface_ip_flush(struct interface_ip_settings *ip) { if (ip == &ip->iface->proto_ip) vlist_flush_all(&ip->iface->host_routes); vlist_simple_flush_all(&ip->dns_servers); vlist_simple_flush_all(&ip->dns_search); vlist_flush_all(&ip->route); vlist_flush_all(&ip->addr); vlist_flush_all(&ip->prefix); } static void __interface_ip_init(struct interface_ip_settings *ip, struct interface *iface) { ip->iface = iface; ip->enabled = true; vlist_simple_init(&ip->dns_search, struct dns_search_domain, node); vlist_simple_init(&ip->dns_servers, struct dns_server, node); vlist_init(&ip->route, route_cmp, interface_update_proto_route); vlist_init(&ip->addr, addr_cmp, interface_update_proto_addr); vlist_init(&ip->prefix, prefix_cmp, interface_update_prefix); } void interface_ip_init(struct interface *iface) { __interface_ip_init(&iface->proto_ip, iface); __interface_ip_init(&iface->config_ip, iface); vlist_init(&iface->host_routes, route_cmp, interface_update_host_route); } static void interface_ip_valid_until_handler(struct uloop_timeout *t) { time_t now = system_get_rtime(); struct interface *iface; vlist_for_each_element(&interfaces, iface, node) { if (iface->state != IFS_UP) continue; struct device_addr *addr, *addrp; struct device_route *route, *routep; struct device_prefix *pref, *prefp; vlist_for_each_element_safe(&iface->proto_ip.addr, addr, node, addrp) if (addr->valid_until && addr->valid_until < now) vlist_delete(&iface->proto_ip.addr, &addr->node); vlist_for_each_element_safe(&iface->proto_ip.route, route, node, routep) if (route->valid_until && route->valid_until < now) vlist_delete(&iface->proto_ip.route, &route->node); vlist_for_each_element_safe(&iface->proto_ip.prefix, pref, node, prefp) if (pref->valid_until && pref->valid_until < now) vlist_delete(&iface->proto_ip.prefix, &pref->node); } uloop_timeout_set(t, 1000); } static void __init interface_ip_init_worker(void) { valid_until_timeout.cb = interface_ip_valid_until_handler; uloop_timeout_set(&valid_until_timeout, 1000); }