diff --git a/sys/compat/linux/linux_netlink.c b/sys/compat/linux/linux_netlink.c index 816d9f34f19f..7744fee094e0 100644 --- a/sys/compat/linux/linux_netlink.c +++ b/sys/compat/linux/linux_netlink.c @@ -1,573 +1,573 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2022 Alexander V. Chernikov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define DEBUG_MOD_NAME nl_linux #define DEBUG_MAX_LEVEL LOG_DEBUG3 #include _DECLARE_DEBUG(LOG_DEBUG); static bool valid_rta_size(const struct rtattr *rta, int sz) { return (NL_RTA_DATA_LEN(rta) == sz); } static bool valid_rta_u32(const struct rtattr *rta) { return (valid_rta_size(rta, sizeof(uint32_t))); } static uint32_t _rta_get_uint32(const struct rtattr *rta) { return (*((const uint32_t *)NL_RTA_DATA_CONST(rta))); } static struct nlmsghdr * rtnl_neigh_from_linux(struct nlmsghdr *hdr, struct nl_pstate *npt) { struct ndmsg *ndm = (struct ndmsg *)(hdr + 1); if (hdr->nlmsg_len >= sizeof(struct nlmsghdr) + sizeof(struct ndmsg)) ndm->ndm_family = linux_to_bsd_domain(ndm->ndm_family); return (hdr); } static struct nlmsghdr * rtnl_ifaddr_from_linux(struct nlmsghdr *hdr, struct nl_pstate *npt) { struct ifaddrmsg *ifam = (struct ifaddrmsg *)(hdr + 1); if (hdr->nlmsg_len >= sizeof(struct nlmsghdr) + sizeof(struct ifaddrmsg)) ifam->ifa_family = linux_to_bsd_domain(ifam->ifa_family); return (hdr); } static struct nlmsghdr * rtnl_route_from_linux(struct nlmsghdr *hdr, struct nl_pstate *npt) { /* Tweak address families and default fib only */ struct rtmsg *rtm = (struct rtmsg *)(hdr + 1); struct nlattr *nla, *nla_head; int attrs_len; rtm->rtm_family = linux_to_bsd_domain(rtm->rtm_family); if (rtm->rtm_table == 254) rtm->rtm_table = 0; attrs_len = hdr->nlmsg_len - sizeof(struct nlmsghdr); attrs_len -= NETLINK_ALIGN(sizeof(struct rtmsg)); nla_head = (struct nlattr *)((char *)rtm + NETLINK_ALIGN(sizeof(struct rtmsg))); NLA_FOREACH(nla, nla_head, attrs_len) { RT_LOG(LOG_DEBUG3, "GOT type %d len %d total %d", nla->nla_type, nla->nla_len, attrs_len); struct rtattr *rta = (struct rtattr *)nla; if (rta->rta_len < sizeof(struct rtattr)) { break; } switch (rta->rta_type) { case NL_RTA_TABLE: if (!valid_rta_u32(rta)) goto done; rtm->rtm_table = 0; uint32_t fibnum = _rta_get_uint32(rta); RT_LOG(LOG_DEBUG3, "GET RTABLE: %u", fibnum); if (fibnum == 254) { *((uint32_t *)NL_RTA_DATA(rta)) = 0; } break; } } done: return (hdr); } static struct nlmsghdr * rtnl_from_linux(struct nlmsghdr *hdr, struct nl_pstate *npt) { switch (hdr->nlmsg_type) { case NL_RTM_GETROUTE: case NL_RTM_NEWROUTE: case NL_RTM_DELROUTE: return (rtnl_route_from_linux(hdr, npt)); case NL_RTM_GETNEIGH: return (rtnl_neigh_from_linux(hdr, npt)); case NL_RTM_GETADDR: return (rtnl_ifaddr_from_linux(hdr, npt)); /* Silence warning for the messages where no translation is required */ case NL_RTM_NEWLINK: case NL_RTM_DELLINK: case NL_RTM_GETLINK: break; default: RT_LOG(LOG_DEBUG, "Passing message type %d untranslated", hdr->nlmsg_type); } return (hdr); } static struct nlmsghdr * nlmsg_from_linux(int netlink_family, struct nlmsghdr *hdr, struct nl_pstate *npt) { switch (netlink_family) { case NETLINK_ROUTE: return (rtnl_from_linux(hdr, npt)); } return (hdr); } /************************************************************ * Kernel -> Linux ************************************************************/ static bool handle_default_out(struct nlmsghdr *hdr, struct nl_writer *nw) { char *out_hdr; out_hdr = nlmsg_reserve_data(nw, NLMSG_ALIGN(hdr->nlmsg_len), char); if (out_hdr != NULL) { memcpy(out_hdr, hdr, hdr->nlmsg_len); return (true); } return (false); } static bool nlmsg_copy_header(struct nlmsghdr *hdr, struct nl_writer *nw) { return (nlmsg_add(nw, hdr->nlmsg_pid, hdr->nlmsg_seq, hdr->nlmsg_type, hdr->nlmsg_flags, 0)); } static void * _nlmsg_copy_next_header(struct nlmsghdr *hdr, struct nl_writer *nw, int sz) { void *next_hdr = nlmsg_reserve_data(nw, sz, void); memcpy(next_hdr, hdr + 1, NLMSG_ALIGN(sz)); return (next_hdr); } #define nlmsg_copy_next_header(_hdr, _ns, _t) \ ((_t *)(_nlmsg_copy_next_header(_hdr, _ns, sizeof(_t)))) static bool nlmsg_copy_nla(const struct nlattr *nla_orig, struct nl_writer *nw) { struct nlattr *nla = nlmsg_reserve_data(nw, nla_orig->nla_len, struct nlattr); if (nla != NULL) { memcpy(nla, nla_orig, nla_orig->nla_len); return (true); } return (false); } static bool nlmsg_copy_all_nla(struct nlmsghdr *hdr, int raw_hdrlen, struct nl_writer *nw) { struct nlattr *nla; int hdrlen = NETLINK_ALIGN(raw_hdrlen); int attrs_len = hdr->nlmsg_len - sizeof(struct nlmsghdr) - hdrlen; struct nlattr *nla_head = (struct nlattr *)((char *)(hdr + 1) + hdrlen); NLA_FOREACH(nla, nla_head, attrs_len) { RT_LOG(LOG_DEBUG3, "reading attr %d len %d", nla->nla_type, nla->nla_len); if (nla->nla_len < sizeof(struct nlattr)) { return (false); } if (!nlmsg_copy_nla(nla, nw)) return (false); } return (true); } static unsigned int rtnl_if_flags_to_linux(unsigned int if_flags) { unsigned int result = 0; for (int i = 0; i < 31; i++) { unsigned int flag = 1 << i; if (!(flag & if_flags)) continue; switch (flag) { case IFF_UP: case IFF_BROADCAST: case IFF_DEBUG: case IFF_LOOPBACK: case IFF_POINTOPOINT: case IFF_DRV_RUNNING: case IFF_NOARP: case IFF_PROMISC: case IFF_ALLMULTI: result |= flag; break; case IFF_KNOWSEPOCH: case IFF_DRV_OACTIVE: case IFF_SIMPLEX: case IFF_LINK0: case IFF_LINK1: case IFF_LINK2: case IFF_CANTCONFIG: case IFF_PPROMISC: case IFF_MONITOR: case IFF_STATICARP: case IFF_STICKYARP: case IFF_DYING: case IFF_RENAMING: case IFF_NOGROUP: /* No Linux analogue */ break; case IFF_MULTICAST: result |= 1 << 12; } } return (result); } static bool rtnl_newlink_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { if (!nlmsg_copy_header(hdr, nw)) return (false); struct ifinfomsg *ifinfo; ifinfo = nlmsg_copy_next_header(hdr, nw, struct ifinfomsg); ifinfo->ifi_family = bsd_to_linux_domain(ifinfo->ifi_family); /* Convert interface type */ switch (ifinfo->ifi_type) { case IFT_ETHER: ifinfo->ifi_type = 1; // ARPHRD_ETHER break; } ifinfo->ifi_flags = rtnl_if_flags_to_linux(ifinfo->ifi_flags); /* Copy attributes unchanged */ if (!nlmsg_copy_all_nla(hdr, sizeof(struct ifinfomsg), nw)) return (false); /* make ip(8) happy */ if (!nlattr_add_string(nw, IFLA_QDISC, "noqueue")) return (false); if (!nlattr_add_u32(nw, IFLA_TXQLEN, 1000)) return (false); nlmsg_end(nw); RT_LOG(LOG_DEBUG2, "done processing nw %p", nw); return (true); } static bool rtnl_newaddr_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { if (!nlmsg_copy_header(hdr, nw)) return (false); struct ifaddrmsg *ifamsg; ifamsg = nlmsg_copy_next_header(hdr, nw, struct ifaddrmsg); ifamsg->ifa_family = bsd_to_linux_domain(ifamsg->ifa_family); /* XXX: fake ifa_flags? */ /* Copy attributes unchanged */ if (!nlmsg_copy_all_nla(hdr, sizeof(struct ifaddrmsg), nw)) return (false); nlmsg_end(nw); RT_LOG(LOG_DEBUG2, "done processing nw %p", nw); return (true); } static bool rtnl_newneigh_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { if (!nlmsg_copy_header(hdr, nw)) return (false); struct ndmsg *ndm; ndm = nlmsg_copy_next_header(hdr, nw, struct ndmsg); ndm->ndm_family = bsd_to_linux_domain(ndm->ndm_family); /* Copy attributes unchanged */ if (!nlmsg_copy_all_nla(hdr, sizeof(struct ndmsg), nw)) return (false); nlmsg_end(nw); RT_LOG(LOG_DEBUG2, "done processing nw %p", nw); return (true); } static bool rtnl_newroute_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { if (!nlmsg_copy_header(hdr, nw)) return (false); struct rtmsg *rtm; rtm = nlmsg_copy_next_header(hdr, nw, struct rtmsg); rtm->rtm_family = bsd_to_linux_domain(rtm->rtm_family); struct nlattr *nla; int hdrlen = NETLINK_ALIGN(sizeof(struct rtmsg)); int attrs_len = hdr->nlmsg_len - sizeof(struct nlmsghdr) - hdrlen; struct nlattr *nla_head = (struct nlattr *)((char *)(hdr + 1) + hdrlen); NLA_FOREACH(nla, nla_head, attrs_len) { struct rtattr *rta = (struct rtattr *)nla; //RT_LOG(LOG_DEBUG, "READING attr %d len %d", nla->nla_type, nla->nla_len); if (rta->rta_len < sizeof(struct rtattr)) { break; } switch (rta->rta_type) { case NL_RTA_TABLE: { uint32_t fibnum; fibnum = _rta_get_uint32(rta); if (fibnum == 0) fibnum = 254; RT_LOG(LOG_DEBUG3, "XFIBNUM %u", fibnum); if (!nlattr_add_u32(nw, NL_RTA_TABLE, fibnum)) return (false); } break; default: if (!nlmsg_copy_nla(nla, nw)) return (false); break; } } nlmsg_end(nw); RT_LOG(LOG_DEBUG2, "done processing nw %p", nw); return (true); } static bool rtnl_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { RT_LOG(LOG_DEBUG2, "Got message type %d", hdr->nlmsg_type); switch (hdr->nlmsg_type) { case NL_RTM_NEWLINK: case NL_RTM_DELLINK: case NL_RTM_GETLINK: return (rtnl_newlink_to_linux(hdr, nlp, nw)); case NL_RTM_NEWADDR: case NL_RTM_DELADDR: return (rtnl_newaddr_to_linux(hdr, nlp, nw)); case NL_RTM_NEWROUTE: case NL_RTM_DELROUTE: return (rtnl_newroute_to_linux(hdr, nlp, nw)); case NL_RTM_NEWNEIGH: case NL_RTM_DELNEIGH: case NL_RTM_GETNEIGH: return (rtnl_newneigh_to_linux(hdr, nlp, nw)); default: RT_LOG(LOG_DEBUG, "[WARN] Passing message type %d untranslated", hdr->nlmsg_type); return (handle_default_out(hdr, nw)); } } static bool nlmsg_error_to_linux(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { if (!nlmsg_copy_header(hdr, nw)) return (false); struct nlmsgerr *nlerr; nlerr = nlmsg_copy_next_header(hdr, nw, struct nlmsgerr); nlerr->error = bsd_to_linux_errno(nlerr->error); int copied_len = sizeof(struct nlmsghdr) + sizeof(struct nlmsgerr); if (hdr->nlmsg_len == copied_len) { nlmsg_end(nw); return (true); } /* * CAP_ACK was not set. Original request needs to be translated. * XXX: implement translation of the original message */ RT_LOG(LOG_DEBUG, "[WARN] Passing ack message type %d untranslated", nlerr->msg.nlmsg_type); char *dst_payload, *src_payload; int copy_len = hdr->nlmsg_len - copied_len; dst_payload = nlmsg_reserve_data(nw, NLMSG_ALIGN(copy_len), char); src_payload = (char *)hdr + copied_len; memcpy(dst_payload, src_payload, copy_len); nlmsg_end(nw); return (true); } static bool nlmsg_to_linux(int netlink_family, struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_writer *nw) { if (hdr->nlmsg_type < NLMSG_MIN_TYPE) { switch (hdr->nlmsg_type) { case NLMSG_ERROR: return (nlmsg_error_to_linux(hdr, nlp, nw)); case NLMSG_NOOP: case NLMSG_DONE: case NLMSG_OVERRUN: return (handle_default_out(hdr, nw)); default: RT_LOG(LOG_DEBUG, "[WARN] Passing message type %d untranslated", hdr->nlmsg_type); return (handle_default_out(hdr, nw)); } } switch (netlink_family) { case NETLINK_ROUTE: return (rtnl_to_linux(hdr, nlp, nw)); default: return (handle_default_out(hdr, nw)); } } static struct mbuf * nlmsgs_to_linux(int netlink_family, char *buf, int data_length, struct nlpcb *nlp) { RT_LOG(LOG_DEBUG3, "LINUX: get %p size %d", buf, data_length); struct nl_writer nw = {}; struct mbuf *m = NULL; if (!nlmsg_get_chain_writer(&nw, data_length, &m)) { RT_LOG(LOG_DEBUG, "unable to setup chain writer for size %d", data_length); return (NULL); } /* Assume correct headers. Buffer IS mutable */ int count = 0; for (int offset = 0; offset + sizeof(struct nlmsghdr) <= data_length;) { struct nlmsghdr *hdr = (struct nlmsghdr *)&buf[offset]; int msglen = NLMSG_ALIGN(hdr->nlmsg_len); count++; if (!nlmsg_to_linux(netlink_family, hdr, nlp, &nw)) { RT_LOG(LOG_DEBUG, "failed to process msg type %d", hdr->nlmsg_type); m_freem(m); return (NULL); } offset += msglen; } nlmsg_flush(&nw); RT_LOG(LOG_DEBUG3, "Processed %d messages, chain size %d", count, m ? m_length(m, NULL) : 0); return (m); } static struct mbuf * mbufs_to_linux(int netlink_family, struct mbuf *m, struct nlpcb *nlp) { /* XXX: easiest solution, not optimized for performance */ int data_length = m_length(m, NULL); char *buf = malloc(data_length, M_LINUX, M_NOWAIT); if (buf == NULL) { RT_LOG(LOG_DEBUG, "unable to allocate %d bytes, dropping message", data_length); m_freem(m); return (NULL); } m_copydata(m, 0, data_length, buf); m_freem(m); m = nlmsgs_to_linux(netlink_family, buf, data_length, nlp); free(buf, M_LINUX); return (m); } static struct linux_netlink_provider linux_netlink_v1 = { .mbufs_to_linux = mbufs_to_linux, .msgs_to_linux = nlmsgs_to_linux, .msg_from_linux = nlmsg_from_linux, }; void -linux_netlink_register() +linux_netlink_register(void) { linux_netlink_p = &linux_netlink_v1; } void -linux_netlink_deregister() +linux_netlink_deregister(void) { linux_netlink_p = NULL; } diff --git a/sys/netlink/netlink_domain.c b/sys/netlink/netlink_domain.c index bca30bedce11..44b5fb732896 100644 --- a/sys/netlink/netlink_domain.c +++ b/sys/netlink/netlink_domain.c @@ -1,692 +1,693 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2021 Ng Peng Nam Sean * Copyright (c) 2022 Alexander V. Chernikov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This file contains socket and protocol bindings for netlink. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* priv_check */ #include #include #include #define DEBUG_MOD_NAME nl_domain #define DEBUG_MAX_LEVEL LOG_DEBUG3 #include _DECLARE_DEBUG(LOG_DEBUG); #define NLCTL_TRACKER struct rm_priotracker nl_tracker #define NLCTL_RLOCK(_ctl) rm_rlock(&((_ctl)->ctl_lock), &nl_tracker) #define NLCTL_RUNLOCK(_ctl) rm_runlock(&((_ctl)->ctl_lock), &nl_tracker) #define NLCTL_WLOCK(_ctl) rm_wlock(&((_ctl)->ctl_lock)) #define NLCTL_WUNLOCK(_ctl) rm_wunlock(&((_ctl)->ctl_lock)) static u_long nl_sendspace = NLSNDQ; SYSCTL_ULONG(_net_netlink, OID_AUTO, sendspace, CTLFLAG_RW, &nl_sendspace, 0, "Default netlink socket send space"); static u_long nl_recvspace = NLSNDQ; SYSCTL_ULONG(_net_netlink, OID_AUTO, recvspace, CTLFLAG_RW, &nl_recvspace, 0, "Default netlink socket receive space"); extern u_long sb_max_adj; static u_long nl_maxsockbuf = 512 * 1024 * 1024; /* 512M, XXX: init based on physmem */ uint32_t nlp_get_pid(const struct nlpcb *nlp) { return (nlp->nl_process_id); } /* * Looks up a nlpcb struct based on the @portid. Need to claim nlsock_mtx. * Returns nlpcb pointer if present else NULL */ static struct nlpcb * nl_port_lookup(uint32_t port_id) { struct nlpcb *nlp; CK_LIST_FOREACH(nlp, &V_nl_ctl->ctl_port_head, nl_port_next) { if (nlp->nl_port == port_id) return (nlp); } return (NULL); } static void nl_update_groups_locked(struct nlpcb *nlp, uint64_t nl_groups) { /* Update group mask */ NL_LOG(LOG_DEBUG2, "socket %p, groups 0x%X -> 0x%X", nlp->nl_socket, (uint32_t)nlp->nl_groups, (uint32_t)nl_groups); nlp->nl_groups = nl_groups; } /* * Broadcasts message @m to the protocol @proto group specified by @group_id */ void nl_send_group(struct mbuf *m, int num_messages, int proto, int group_id) { struct nlpcb *nlp_last = NULL; struct nlpcb *nlp; NLCTL_TRACKER; IF_DEBUG_LEVEL(LOG_DEBUG2) { struct nlmsghdr *hdr = mtod(m, struct nlmsghdr *); NL_LOG(LOG_DEBUG2, "MCAST mbuf len %u msg type %d len %u to group %d/%d", m->m_len, hdr->nlmsg_type, hdr->nlmsg_len, proto, group_id); } struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); if (__predict_false(ctl == NULL)) { /* * Can be the case when notification is sent within VNET * which doesn't have any netlink sockets. */ m_freem(m); return; } NLCTL_RLOCK(ctl); int io_flags = NL_IOF_UNTRANSLATED; uint64_t groups_mask = 1 << ((uint64_t)group_id - 1); CK_LIST_FOREACH(nlp, &ctl->ctl_pcb_head, nl_next) { if (nlp->nl_groups & groups_mask && nlp->nl_proto == proto) { if (nlp_last != NULL) { struct mbuf *m_copy; m_copy = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (m_copy != NULL) nl_send_one(m_copy, nlp_last, num_messages, io_flags); else { NLP_LOCK(nlp_last); if (nlp_last->nl_socket != NULL) sorwakeup(nlp_last->nl_socket); NLP_UNLOCK(nlp_last); } } nlp_last = nlp; } } if (nlp_last != NULL) nl_send_one(m, nlp_last, num_messages, io_flags); else m_freem(m); NLCTL_RUNLOCK(ctl); } bool nl_has_listeners(int netlink_family, uint32_t groups_mask) { return (V_nl_ctl != NULL); } bool nlp_has_priv(struct nlpcb *nlp, int priv) { return (priv_check_cred(nlp->nl_cred, priv) == 0); } static uint32_t -nl_find_port() { +nl_find_port(void) +{ /* * app can open multiple netlink sockets. * Start with current pid, if already taken, * try random numbers in 65k..256k+65k space, * avoiding clash with pids. */ if (nl_port_lookup(curproc->p_pid) == NULL) return (curproc->p_pid); for (int i = 0; i < 16; i++) { uint32_t nl_port = (arc4random() % 65536) + 65536 * 4; if (nl_port_lookup(nl_port) == 0) return (nl_port); NL_LOG(LOG_DEBUG3, "tried %u\n", nl_port); } return (curproc->p_pid); } static int nl_bind_locked(struct nlpcb *nlp, struct sockaddr_nl *snl) { if (nlp->nl_bound) { if (nlp->nl_port != snl->nl_pid) { NL_LOG(LOG_DEBUG, "bind() failed: program pid %d " "is different from provided pid %d", nlp->nl_port, snl->nl_pid); return (EINVAL); // XXX: better error } } else { if (snl->nl_pid == 0) snl->nl_pid = nl_find_port(); if (nl_port_lookup(snl->nl_pid) != NULL) return (EADDRINUSE); nlp->nl_port = snl->nl_pid; nlp->nl_bound = true; CK_LIST_INSERT_HEAD(&V_nl_ctl->ctl_port_head, nlp, nl_port_next); } nl_update_groups_locked(nlp, snl->nl_groups); return (0); } static int nl_pru_attach(struct socket *so, int proto, struct thread *td) { struct nlpcb *nlp; int error; if (__predict_false(netlink_unloading != 0)) return (EAFNOSUPPORT); error = nl_verify_proto(proto); if (error != 0) return (error); bool is_linux = SV_PROC_ABI(td->td_proc) == SV_ABI_LINUX; NL_LOG(LOG_DEBUG2, "socket %p, %sPID %d: attaching socket to %s", so, is_linux ? "(linux) " : "", curproc->p_pid, nl_get_proto_name(proto)); /* Create per-VNET state on first socket init */ struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); if (ctl == NULL) ctl = vnet_nl_ctl_init(); KASSERT(V_nl_ctl != NULL, ("nl_attach: vnet_sock_init() failed")); MPASS(sotonlpcb(so) == NULL); nlp = malloc(sizeof(struct nlpcb), M_PCB, M_WAITOK | M_ZERO); error = soreserve(so, nl_sendspace, nl_recvspace); if (error != 0) { free(nlp, M_PCB); return (error); } so->so_pcb = nlp; nlp->nl_socket = so; /* Copy so_cred to avoid having socket_var.h in every header */ nlp->nl_cred = so->so_cred; nlp->nl_proto = proto; nlp->nl_process_id = curproc->p_pid; nlp->nl_linux = is_linux; nlp->nl_active = true; NLP_LOCK_INIT(nlp); refcount_init(&nlp->nl_refcount, 1); nl_init_io(nlp); nlp->nl_taskqueue = taskqueue_create("netlink_socket", M_WAITOK, taskqueue_thread_enqueue, &nlp->nl_taskqueue); TASK_INIT(&nlp->nl_task, 0, nl_taskqueue_handler, nlp); taskqueue_start_threads(&nlp->nl_taskqueue, 1, PWAIT, "netlink_socket (PID %u)", nlp->nl_process_id); NLCTL_WLOCK(ctl); /* XXX: check ctl is still alive */ CK_LIST_INSERT_HEAD(&ctl->ctl_pcb_head, nlp, nl_next); NLCTL_WUNLOCK(ctl); soisconnected(so); return (0); } static void nl_pru_abort(struct socket *so) { NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); MPASS(sotonlpcb(so) != NULL); soisdisconnected(so); } static int nl_pru_bind(struct socket *so, struct sockaddr *sa, struct thread *td) { struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); struct nlpcb *nlp = sotonlpcb(so); struct sockaddr_nl *snl = (struct sockaddr_nl *)sa; int error; NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); if (snl->nl_len != sizeof(*snl)) { NL_LOG(LOG_DEBUG, "socket %p, wrong sizeof(), ignoring bind()", so); return (EINVAL); } NLCTL_WLOCK(ctl); NLP_LOCK(nlp); error = nl_bind_locked(nlp, snl); NLP_UNLOCK(nlp); NLCTL_WUNLOCK(ctl); NL_LOG(LOG_DEBUG2, "socket %p, bind() to %u, groups %u, error %d", so, snl->nl_pid, snl->nl_groups, error); return (error); } static int nl_assign_port(struct nlpcb *nlp, uint32_t port_id) { struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); struct sockaddr_nl snl = { .nl_pid = port_id, }; int error; NLCTL_WLOCK(ctl); NLP_LOCK(nlp); snl.nl_groups = nlp->nl_groups; error = nl_bind_locked(nlp, &snl); NLP_UNLOCK(nlp); NLCTL_WUNLOCK(ctl); NL_LOG(LOG_DEBUG3, "socket %p, port assign: %d, error: %d", nlp->nl_socket, port_id, error); return (error); } /* * nl_autobind_port binds a unused portid to @nlp * @nlp: pcb data for the netlink socket * @candidate_id: first id to consider */ static int nl_autobind_port(struct nlpcb *nlp, uint32_t candidate_id) { struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); uint32_t port_id = candidate_id; NLCTL_TRACKER; bool exist; int error; for (int i = 0; i < 10; i++) { NL_LOG(LOG_DEBUG3, "socket %p, trying to assign port %d", nlp->nl_socket, port_id); NLCTL_RLOCK(ctl); exist = nl_port_lookup(port_id) != 0; NLCTL_RUNLOCK(ctl); if (!exist) { error = nl_assign_port(nlp, port_id); if (error != EADDRINUSE) break; } port_id++; } NL_LOG(LOG_DEBUG3, "socket %p, autobind to %d, error: %d", nlp->nl_socket, port_id, error); return (error); } static int nl_pru_connect(struct socket *so, struct sockaddr *sa, struct thread *td) { struct sockaddr_nl *snl = (struct sockaddr_nl *)sa; struct nlpcb *nlp; NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); if (snl->nl_len != sizeof(*snl)) { NL_LOG(LOG_DEBUG, "socket %p, wrong sizeof(), ignoring bind()", so); return (EINVAL); } nlp = sotonlpcb(so); if (!nlp->nl_bound) { int error = nl_autobind_port(nlp, td->td_proc->p_pid); if (error != 0) { NL_LOG(LOG_DEBUG, "socket %p, nl_autobind() failed: %d", so, error); return (error); } } /* XXX: Handle socket flags & multicast */ soisconnected(so); NL_LOG(LOG_DEBUG2, "socket %p, connect to %u", so, snl->nl_pid); return (0); } static void destroy_nlpcb(struct nlpcb *nlp) { NLP_LOCK(nlp); nl_free_io(nlp); NLP_LOCK_DESTROY(nlp); free(nlp, M_PCB); } static void destroy_nlpcb_epoch(epoch_context_t ctx) { struct nlpcb *nlp; nlp = __containerof(ctx, struct nlpcb, nl_epoch_ctx); destroy_nlpcb(nlp); } static void nl_pru_detach(struct socket *so) { struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); MPASS(sotonlpcb(so) != NULL); struct nlpcb *nlp; NL_LOG(LOG_DEBUG2, "detaching socket %p, PID %d", so, curproc->p_pid); nlp = sotonlpcb(so); /* Mark as inactive so no new work can be enqueued */ NLP_LOCK(nlp); bool was_bound = nlp->nl_bound; nlp->nl_active = false; NLP_UNLOCK(nlp); /* Wait till all scheduled work has been completed */ taskqueue_drain_all(nlp->nl_taskqueue); taskqueue_free(nlp->nl_taskqueue); NLCTL_WLOCK(ctl); NLP_LOCK(nlp); if (was_bound) { CK_LIST_REMOVE(nlp, nl_port_next); NL_LOG(LOG_DEBUG3, "socket %p, unlinking bound pid %u", so, nlp->nl_port); } CK_LIST_REMOVE(nlp, nl_next); nlp->nl_socket = NULL; NLP_UNLOCK(nlp); NLCTL_WUNLOCK(ctl); so->so_pcb = NULL; NL_LOG(LOG_DEBUG3, "socket %p, detached", so); /* XXX: is delayed free needed? */ epoch_call(net_epoch_preempt, destroy_nlpcb_epoch, &nlp->nl_epoch_ctx); } static int nl_pru_disconnect(struct socket *so) { NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); MPASS(sotonlpcb(so) != NULL); return (ENOTCONN); } static int nl_pru_peeraddr(struct socket *so, struct sockaddr **sa) { NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); MPASS(sotonlpcb(so) != NULL); return (ENOTCONN); } static int nl_pru_shutdown(struct socket *so) { NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); MPASS(sotonlpcb(so) != NULL); socantsendmore(so); return (0); } static int nl_pru_sockaddr(struct socket *so, struct sockaddr **sa) { struct sockaddr_nl *snl; snl = malloc(sizeof(struct sockaddr_nl), M_SONAME, M_WAITOK | M_ZERO); /* TODO: set other fields */ snl->nl_len = sizeof(struct sockaddr_nl); snl->nl_family = AF_NETLINK; snl->nl_pid = sotonlpcb(so)->nl_port; *sa = (struct sockaddr *)snl; return (0); } static void nl_pru_close(struct socket *so) { NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); MPASS(sotonlpcb(so) != NULL); soisdisconnected(so); } static int nl_pru_output(struct mbuf *m, struct socket *so, ...) { if (__predict_false(m == NULL || ((m->m_len < sizeof(struct nlmsghdr)) && (m = m_pullup(m, sizeof(struct nlmsghdr))) == NULL))) return (ENOBUFS); MPASS((m->m_flags & M_PKTHDR) != 0); NL_LOG(LOG_DEBUG3, "sending message to kernel async processing"); nl_receive_async(m, so); return (0); } static int nl_pru_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *sa, struct mbuf *control, struct thread *td) { NL_LOG(LOG_DEBUG2, "sending message to kernel"); if (__predict_false(control != NULL)) { if (control->m_len) { m_freem(control); return (EINVAL); } m_freem(control); } return (nl_pru_output(m, so)); } static int nl_pru_rcvd(struct socket *so, int flags) { NL_LOG(LOG_DEBUG3, "socket %p, PID %d", so, curproc->p_pid); MPASS(sotonlpcb(so) != NULL); nl_on_transmit(sotonlpcb(so)); return (0); } static int nl_getoptflag(int sopt_name) { switch (sopt_name) { case NETLINK_CAP_ACK: return (NLF_CAP_ACK); case NETLINK_EXT_ACK: return (NLF_EXT_ACK); case NETLINK_GET_STRICT_CHK: return (NLF_STRICT); } return (0); } static int nl_ctloutput(struct socket *so, struct sockopt *sopt) { struct nl_control *ctl = atomic_load_ptr(&V_nl_ctl); struct nlpcb *nlp = sotonlpcb(so); uint32_t flag; uint64_t groups, group_mask; int optval, error = 0; NLCTL_TRACKER; NL_LOG(LOG_DEBUG2, "%ssockopt(%p, %d)", (sopt->sopt_dir) ? "set" : "get", so, sopt->sopt_name); switch (sopt->sopt_dir) { case SOPT_SET: switch (sopt->sopt_name) { case NETLINK_ADD_MEMBERSHIP: case NETLINK_DROP_MEMBERSHIP: sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (optval <= 0 || optval >= 64) { error = ERANGE; break; } group_mask = (uint64_t)1 << (optval - 1); NL_LOG(LOG_DEBUG2, "ADD/DEL group %d mask (%X)", (uint32_t)optval, (uint32_t)group_mask); NLCTL_WLOCK(ctl); if (sopt->sopt_name == NETLINK_ADD_MEMBERSHIP) groups = nlp->nl_groups | group_mask; else groups = nlp->nl_groups & ~group_mask; nl_update_groups_locked(nlp, groups); NLCTL_WUNLOCK(ctl); break; case NETLINK_CAP_ACK: case NETLINK_EXT_ACK: case NETLINK_GET_STRICT_CHK: sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); flag = nl_getoptflag(sopt->sopt_name); NLCTL_WLOCK(ctl); if (optval != 0) nlp->nl_flags |= flag; else nlp->nl_flags &= ~flag; NLCTL_WUNLOCK(ctl); break; default: error = ENOPROTOOPT; } break; case SOPT_GET: switch (sopt->sopt_name) { case NETLINK_LIST_MEMBERSHIPS: NLCTL_RLOCK(ctl); optval = nlp->nl_groups; NLCTL_RUNLOCK(ctl); error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case NETLINK_CAP_ACK: case NETLINK_EXT_ACK: case NETLINK_GET_STRICT_CHK: NLCTL_RLOCK(ctl); optval = (nlp->nl_flags & nl_getoptflag(sopt->sopt_name)) != 0; NLCTL_RUNLOCK(ctl); error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = ENOPROTOOPT; } break; default: error = ENOPROTOOPT; } return (error); } static int nl_setsbopt(struct socket *so, struct sockopt *sopt) { int error, optval; bool result; if (sopt->sopt_name != SO_RCVBUF) return (sbsetopt(so, sopt)); /* Allow to override max buffer size in certain conditions */ error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error != 0) return (error); NL_LOG(LOG_DEBUG2, "socket %p, PID %d, SO_RCVBUF=%d", so, curproc->p_pid, optval); if (optval > sb_max_adj) { if (priv_check(curthread, PRIV_NET_ROUTE) != 0) return (EPERM); } SOCK_RECVBUF_LOCK(so); result = sbreserve_locked_limit(so, SO_RCV, optval, nl_maxsockbuf, curthread); SOCK_RECVBUF_UNLOCK(so); return (result ? 0 : ENOBUFS); } static struct protosw netlinksw = { .pr_type = SOCK_RAW, .pr_flags = PR_ATOMIC | PR_ADDR | PR_WANTRCVD, .pr_ctloutput = nl_ctloutput, .pr_setsbopt = nl_setsbopt, .pr_abort = nl_pru_abort, .pr_attach = nl_pru_attach, .pr_bind = nl_pru_bind, .pr_connect = nl_pru_connect, .pr_detach = nl_pru_detach, .pr_disconnect = nl_pru_disconnect, .pr_peeraddr = nl_pru_peeraddr, .pr_send = nl_pru_send, .pr_rcvd = nl_pru_rcvd, .pr_shutdown = nl_pru_shutdown, .pr_sockaddr = nl_pru_sockaddr, .pr_close = nl_pru_close }; static struct domain netlinkdomain = { .dom_family = PF_NETLINK, .dom_name = "netlink", .dom_flags = DOMF_UNLOADABLE, .dom_nprotosw = 1, .dom_protosw = { &netlinksw }, }; DOMAIN_SET(netlink); diff --git a/sys/netlink/netlink_generic.c b/sys/netlink/netlink_generic.c index 64985f656a63..7d226f77e44b 100644 --- a/sys/netlink/netlink_generic.c +++ b/sys/netlink/netlink_generic.c @@ -1,475 +1,475 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2022 Alexander V. Chernikov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #define DEBUG_MOD_NAME nl_generic #define DEBUG_MAX_LEVEL LOG_DEBUG3 #include _DECLARE_DEBUG(LOG_DEBUG3); #define MAX_FAMILIES 20 #define MAX_GROUPS 20 #define MIN_GROUP_NUM 48 static struct sx sx_lock; #define GENL_LOCK_INIT() sx_init(&sx_lock, "genetlink lock") #define GENL_LOCK_DESTROY() sx_destroy(&sx_lock) #define GENL_LOCK() sx_xlock(&sx_lock) #define GENL_UNLOCK() sx_xunlock(&sx_lock) struct genl_family { const char *family_name; uint16_t family_hdrsize; uint16_t family_id; uint16_t family_version; uint16_t family_attr_max; uint16_t family_cmd_size; uint16_t family_num_groups; struct genl_cmd *family_cmds; }; static struct genl_family families[MAX_FAMILIES]; struct genl_group { struct genl_family *group_family; const char *group_name; }; static struct genl_group groups[MAX_GROUPS]; static int dump_family(struct nlmsghdr *hdr, struct genlmsghdr *ghdr, const struct genl_family *gf, struct nl_writer *nw); static void nlctrl_notify(const struct genl_family *gf, int action); static struct genl_family * find_family(const char *family_name) { for (int i = 0; i < MAX_FAMILIES; i++) { struct genl_family *gf = &families[i]; if (gf->family_name != NULL && !strcmp(gf->family_name, family_name)) return (gf); } return (NULL); } uint32_t genl_register_family(const char *family_name, size_t hdrsize, int family_version, int max_attr_idx) { uint32_t family_id = 0; MPASS(family_name != NULL); if (find_family(family_name) != NULL) return (0); GENL_LOCK(); for (int i = 0; i < MAX_FAMILIES; i++) { struct genl_family *gf = &families[i]; if (gf->family_name == NULL) { gf->family_name = family_name; gf->family_version = family_version; gf->family_hdrsize = hdrsize; gf->family_attr_max = max_attr_idx; gf->family_id = i + GENL_MIN_ID; NL_LOG(LOG_DEBUG2, "Registered family %s id %d", gf->family_name, gf->family_id); family_id = gf->family_id; nlctrl_notify(gf, CTRL_CMD_NEWFAMILY); break; } } GENL_UNLOCK(); return (family_id); } static void free_family(struct genl_family *gf) { if (gf->family_cmds != NULL) free(gf->family_cmds, M_NETLINK); } /* * Can sleep, I guess */ bool genl_unregister_family(const char *family_name) { bool found = false; GENL_LOCK(); struct genl_family *gf = find_family(family_name); nlctrl_notify(gf, CTRL_CMD_DELFAMILY); if (gf != NULL) { found = true; /* TODO: zero pointer first */ free_family(gf); bzero(gf, sizeof(*gf)); } GENL_UNLOCK(); return (found); } bool genl_register_cmds(const char *family_name, const struct genl_cmd *cmds, int count) { GENL_LOCK(); struct genl_family *gf = find_family(family_name); if (gf == NULL) { GENL_UNLOCK(); return (false); } int cmd_size = gf->family_cmd_size; for (int i = 0; i < count; i++) { MPASS(cmds[i].cmd_cb != NULL); if (cmds[i].cmd_num >= cmd_size) cmd_size = cmds[i].cmd_num + 1; } if (cmd_size > gf->family_cmd_size) { /* need to realloc */ size_t sz = cmd_size * sizeof(struct genl_cmd); void *data = malloc(sz, M_NETLINK, M_WAITOK | M_ZERO); memcpy(data, gf->family_cmds, gf->family_cmd_size * sizeof(struct genl_cmd)); void *old_data = gf->family_cmds; gf->family_cmds = data; gf->family_cmd_size = cmd_size; free(old_data, M_NETLINK); } for (int i = 0; i < count; i++) { const struct genl_cmd *cmd = &cmds[i]; MPASS(gf->family_cmds[cmd->cmd_num].cmd_cb == NULL); gf->family_cmds[cmd->cmd_num] = cmds[i]; NL_LOG(LOG_DEBUG2, "Adding cmd %s(%d) to family %s", cmd->cmd_name, cmd->cmd_num, gf->family_name); } GENL_UNLOCK(); return (true); } static struct genl_group * find_group(const struct genl_family *gf, const char *group_name) { for (int i = 0; i < MAX_GROUPS; i++) { struct genl_group *gg = &groups[i]; if (gg->group_family == gf && !strcmp(gg->group_name, group_name)) return (gg); } return (NULL); } uint32_t genl_register_group(const char *family_name, const char *group_name) { uint32_t group_id = 0; MPASS(family_name != NULL); MPASS(group_name != NULL); GENL_LOCK(); struct genl_family *gf = find_family(family_name); if (gf == NULL || find_group(gf, group_name) != NULL) { GENL_UNLOCK(); return (0); } for (int i = 0; i < MAX_GROUPS; i++) { struct genl_group *gg = &groups[i]; if (gg->group_family == NULL) { gf->family_num_groups++; gg->group_family = gf; gg->group_name = group_name; group_id = i + MIN_GROUP_NUM; break; } } GENL_UNLOCK(); return (group_id); } /* * Handler called by netlink subsystem when matching netlink message is received */ static int genl_handle_message(struct nlmsghdr *hdr, struct nl_pstate *npt) { struct nlpcb *nlp = npt->nlp; int error = 0; int family_id = (int)hdr->nlmsg_type - GENL_MIN_ID; if (__predict_false(family_id < 0 || family_id > MAX_FAMILIES)) { NLP_LOG(LOG_DEBUG, nlp, "invalid message type: %d", hdr->nlmsg_type); return (ENOTSUP); } if (__predict_false(hdr->nlmsg_len < sizeof(hdr) + GENL_HDRLEN)) { NLP_LOG(LOG_DEBUG, nlp, "invalid message size: %d", hdr->nlmsg_len); return (EINVAL); } struct genl_family *gf = &families[family_id]; struct genlmsghdr *ghdr = (struct genlmsghdr *)(hdr + 1); if (ghdr->cmd >= gf->family_cmd_size || gf->family_cmds[ghdr->cmd].cmd_cb == NULL) { NLP_LOG(LOG_DEBUG, nlp, "family %s: invalid cmd %d", gf->family_name, ghdr->cmd); return (ENOTSUP); } struct genl_cmd *cmd = &gf->family_cmds[ghdr->cmd]; if (cmd->cmd_priv != 0 && !nlp_has_priv(nlp, cmd->cmd_priv)) { NLP_LOG(LOG_DEBUG, nlp, "family %s: cmd %d priv_check() failed", gf->family_name, ghdr->cmd); return (EPERM); } NLP_LOG(LOG_DEBUG2, nlp, "received family %s cmd %s(%d) len %d", gf->family_name, cmd->cmd_name, ghdr->cmd, hdr->nlmsg_len); error = cmd->cmd_cb(hdr, npt); return (error); } static uint32_t get_cmd_flags(const struct genl_cmd *cmd) { uint32_t flags = cmd->cmd_flags; if (cmd->cmd_priv != 0) flags |= GENL_ADMIN_PERM; return (flags); } static int dump_family(struct nlmsghdr *hdr, struct genlmsghdr *ghdr, const struct genl_family *gf, struct nl_writer *nw) { if (!nlmsg_reply(nw, hdr, sizeof(struct genlmsghdr))) goto enomem; struct genlmsghdr *ghdr_new = nlmsg_reserve_object(nw, struct genlmsghdr); ghdr_new->cmd = ghdr->cmd; ghdr_new->version = gf->family_version; ghdr_new->reserved = 0; nlattr_add_string(nw, CTRL_ATTR_FAMILY_NAME, gf->family_name); nlattr_add_u16(nw, CTRL_ATTR_FAMILY_ID, gf->family_id); nlattr_add_u32(nw, CTRL_ATTR_VERSION, gf->family_version); nlattr_add_u32(nw, CTRL_ATTR_HDRSIZE, gf->family_hdrsize); nlattr_add_u32(nw, CTRL_ATTR_MAXATTR, gf->family_attr_max); if (gf->family_cmd_size > 0) { int off = nlattr_add_nested(nw, CTRL_ATTR_OPS); if (off == 0) goto enomem; for (int i = 0, cnt=0; i < gf->family_cmd_size; i++) { struct genl_cmd *cmd = &gf->family_cmds[i]; if (cmd->cmd_cb == NULL) continue; int cmd_off = nlattr_add_nested(nw, ++cnt); if (cmd_off == 0) goto enomem; nlattr_add_u32(nw, CTRL_ATTR_OP_ID, cmd->cmd_num); nlattr_add_u32(nw, CTRL_ATTR_OP_FLAGS, get_cmd_flags(cmd)); nlattr_set_len(nw, cmd_off); } nlattr_set_len(nw, off); } if (gf->family_num_groups > 0) { int off = nlattr_add_nested(nw, CTRL_ATTR_MCAST_GROUPS); if (off == 0) goto enomem; for (int i = 0, cnt = 0; i < MAX_GROUPS; i++) { struct genl_group *gg = &groups[i]; if (gg->group_family != gf) continue; int cmd_off = nlattr_add_nested(nw, ++cnt); if (cmd_off == 0) goto enomem; nlattr_add_u32(nw, CTRL_ATTR_MCAST_GRP_ID, i + MIN_GROUP_NUM); nlattr_add_string(nw, CTRL_ATTR_MCAST_GRP_NAME, gg->group_name); nlattr_set_len(nw, cmd_off); } nlattr_set_len(nw, off); } if (nlmsg_end(nw)) return (0); enomem: NL_LOG(LOG_DEBUG, "unable to dump family %s state (ENOMEM)", gf->family_name); nlmsg_abort(nw); return (ENOMEM); } /* Declare ourself as a user */ #define CTRL_FAMILY_NAME "nlctrl" static uint32_t ctrl_family_id; static uint32_t ctrl_group_id; struct nl_parsed_family { uint32_t family_id; char *family_name; uint8_t version; }; #define _IN(_field) offsetof(struct genlmsghdr, _field) #define _OUT(_field) offsetof(struct nl_parsed_family, _field) static const struct nlfield_parser nlf_p_generic[] = { { .off_in = _IN(version), .off_out = _OUT(version), .cb = nlf_get_u8 }, }; static struct nlattr_parser nla_p_generic[] = { { .type = CTRL_ATTR_FAMILY_ID , .off = _OUT(family_id), .cb = nlattr_get_uint32 }, { .type = CTRL_ATTR_FAMILY_NAME , .off = _OUT(family_id), .cb = nlattr_get_string }, }; #undef _IN #undef _OUT NL_DECLARE_PARSER(genl_parser, struct genlmsghdr, nlf_p_generic, nla_p_generic); static int nlctrl_handle_getfamily(struct nlmsghdr *hdr, struct nl_pstate *npt) { int error = 0; struct nl_parsed_family attrs = {}; error = nl_parse_nlmsg(hdr, &genl_parser, npt, &attrs); if (error != 0) return (error); struct genlmsghdr ghdr = { .cmd = CTRL_CMD_NEWFAMILY, }; for (int i = 0; i < MAX_FAMILIES; i++) { struct genl_family *gf = &families[i]; if (gf->family_name == NULL) continue; if (attrs.family_id != 0 && attrs.family_id != gf->family_id) continue; if (attrs.family_name != NULL && strcmp(attrs.family_name, gf->family_name)) continue; error = dump_family(hdr, &ghdr, &families[i], npt->nw); if (error != 0) break; } return (error); } static void nlctrl_notify(const struct genl_family *gf, int cmd) { struct nlmsghdr hdr = {.nlmsg_type = NETLINK_GENERIC }; struct genlmsghdr ghdr = { .cmd = cmd }; struct nl_writer nw = {}; if (nlmsg_get_group_writer(&nw, NLMSG_SMALL, NETLINK_GENERIC, ctrl_group_id)) { dump_family(&hdr, &ghdr, gf, &nw); nlmsg_flush(&nw); return; } NL_LOG(LOG_DEBUG, "error allocating group writer"); } static const struct genl_cmd nlctrl_cmds[] = { { .cmd_num = CTRL_CMD_GETFAMILY, .cmd_name = "GETFAMILY", .cmd_cb = nlctrl_handle_getfamily, .cmd_flags = GENL_CMD_CAP_DO | GENL_CMD_CAP_DUMP, GENL_CMD_CAP_HASPOL, }, }; static void -genl_nlctrl_init() +genl_nlctrl_init(void) { ctrl_family_id = genl_register_family(CTRL_FAMILY_NAME, 0, 2, CTRL_ATTR_MAX); genl_register_cmds(CTRL_FAMILY_NAME, nlctrl_cmds, NL_ARRAY_LEN(nlctrl_cmds)); ctrl_group_id = genl_register_group(CTRL_FAMILY_NAME, "notify"); } static void -genl_nlctrl_destroy() +genl_nlctrl_destroy(void) { genl_unregister_family(CTRL_FAMILY_NAME); } static const struct nlhdr_parser *all_parsers[] = { &genl_parser }; static void genl_load(void *u __unused) { GENL_LOCK_INIT(); NL_VERIFY_PARSERS(all_parsers); netlink_register_proto(NETLINK_GENERIC, "NETLINK_GENERIC", genl_handle_message); genl_nlctrl_init(); } SYSINIT(genl_load, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, genl_load, NULL); static void genl_unload(void *u __unused) { genl_nlctrl_destroy(); GENL_LOCK_DESTROY(); epoch_wait_preempt(net_epoch_preempt); } SYSUNINIT(genl_unload, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, genl_unload, NULL); diff --git a/sys/netlink/route/neigh.c b/sys/netlink/route/neigh.c index 076f4bfafdf4..9bcbb68dd375 100644 --- a/sys/netlink/route/neigh.c +++ b/sys/netlink/route/neigh.c @@ -1,574 +1,574 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2022 Alexander V. Chernikov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include /* nd6.h requires this */ #include /* nd6 state machine */ #include /* scope deembedding */ #define DEBUG_MOD_NAME nl_neigh #define DEBUG_MAX_LEVEL LOG_DEBUG3 #include _DECLARE_DEBUG(LOG_DEBUG); static int lle_families[] = { AF_INET, AF_INET6 }; static eventhandler_tag lle_event_p; struct netlink_walkargs { struct nl_writer *nw; struct nlmsghdr hdr; struct nlpcb *so; struct ifnet *ifp; int family; int error; int count; int dumped; }; static int lle_state_to_nl_state(int family, struct llentry *lle) { int state = lle->ln_state; switch (family) { case AF_INET: if (lle->la_flags & (LLE_STATIC | LLE_IFADDR)) state = 1; switch (state) { case 0: /* ARP_LLINFO_INCOMPLETE */ return (NUD_INCOMPLETE); case 1: /* ARP_LLINFO_REACHABLE */ return (NUD_REACHABLE); case 2: /* ARP_LLINFO_VERIFY */ return (NUD_PROBE); } break; case AF_INET6: switch (state) { case ND6_LLINFO_INCOMPLETE: return (NUD_INCOMPLETE); case ND6_LLINFO_REACHABLE: return (NUD_REACHABLE); case ND6_LLINFO_STALE: return (NUD_STALE); case ND6_LLINFO_DELAY: return (NUD_DELAY); case ND6_LLINFO_PROBE: return (NUD_PROBE); } break; } return (NUD_NONE); } static uint32_t lle_flags_to_nl_flags(const struct llentry *lle) { uint32_t nl_flags = 0; if (lle->la_flags & LLE_IFADDR) nl_flags |= NTF_SELF; if (lle->la_flags & LLE_PUB) nl_flags |= NTF_PROXY; if (lle->la_flags & LLE_STATIC) nl_flags |= NTF_STICKY; if (lle->ln_router != 0) nl_flags |= NTF_ROUTER; return (nl_flags); } static int dump_lle_locked(struct llentry *lle, void *arg) { struct netlink_walkargs *wa = (struct netlink_walkargs *)arg; struct nlmsghdr *hdr = &wa->hdr; struct nl_writer *nw = wa->nw; struct ndmsg *ndm; #if defined(INET) || defined(INET6) union { struct in_addr in; struct in6_addr in6; } addr; #endif IF_DEBUG_LEVEL(LOG_DEBUG2) { char llebuf[NHOP_PRINT_BUFSIZE]; llentry_print_buf_lltable(lle, llebuf, sizeof(llebuf)); NL_LOG(LOG_DEBUG2, "dumping %s", llebuf); } if (!nlmsg_reply(nw, hdr, sizeof(struct ndmsg))) goto enomem; ndm = nlmsg_reserve_object(nw, struct ndmsg); ndm->ndm_family = wa->family; ndm->ndm_ifindex = wa->ifp->if_index; ndm->ndm_state = lle_state_to_nl_state(wa->family, lle); ndm->ndm_flags = lle_flags_to_nl_flags(lle); switch (wa->family) { #ifdef INET case AF_INET: addr.in = lle->r_l3addr.addr4; nlattr_add(nw, NDA_DST, 4, &addr); break; #endif #ifdef INET6 case AF_INET6: addr.in6 = lle->r_l3addr.addr6; in6_clearscope(&addr.in6); nlattr_add(nw, NDA_DST, 16, &addr); break; #endif } if (lle->r_flags & RLLE_VALID) { /* Has L2 */ int addrlen = wa->ifp->if_addrlen; nlattr_add(nw, NDA_LLADDR, addrlen, lle->ll_addr); } nlattr_add_u32(nw, NDA_PROBES, lle->la_asked); struct nda_cacheinfo *cache; cache = nlmsg_reserve_attr(nw, NDA_CACHEINFO, struct nda_cacheinfo); if (cache == NULL) goto enomem; /* TODO: provide confirmed/updated */ cache->ndm_refcnt = lle->lle_refcnt; if (nlmsg_end(nw)) return (0); enomem: NL_LOG(LOG_DEBUG, "unable to dump lle state (ENOMEM)"); nlmsg_abort(nw); return (ENOMEM); } static int dump_lle(struct lltable *llt, struct llentry *lle, void *arg) { int error; LLE_RLOCK(lle); error = dump_lle_locked(lle, arg); LLE_RUNLOCK(lle); return (error); } static bool dump_llt(struct lltable *llt, struct netlink_walkargs *wa) { lltable_foreach_lle(llt, dump_lle, wa); return (true); } static int dump_llts_iface(struct netlink_walkargs *wa, struct ifnet *ifp, int family) { int error = 0; wa->ifp = ifp; for (int i = 0; i < sizeof(lle_families) / sizeof(int); i++) { int fam = lle_families[i]; struct lltable *llt = lltable_get(ifp, fam); if (llt != NULL && (family == 0 || family == fam)) { wa->count++; wa->family = fam; if (!dump_llt(llt, wa)) { error = ENOMEM; break; } wa->dumped++; } } return (error); } static int dump_llts(struct netlink_walkargs *wa, struct ifnet *ifp, int family) { NL_LOG(LOG_DEBUG, "Start dump ifp=%s family=%d", ifp ? if_name(ifp) : "NULL", family); wa->hdr.nlmsg_flags |= NLM_F_MULTI; if (ifp != NULL) { dump_llts_iface(wa, ifp, family); } else { CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { dump_llts_iface(wa, ifp, family); } } NL_LOG(LOG_DEBUG, "End dump, iterated %d dumped %d", wa->count, wa->dumped); if (!nlmsg_end_dump(wa->nw, wa->error, &wa->hdr)) { NL_LOG(LOG_DEBUG, "Unable to add new message"); return (ENOMEM); } return (0); } static int get_lle(struct netlink_walkargs *wa, struct ifnet *ifp, int family, struct sockaddr *dst) { struct lltable *llt = lltable_get(ifp, family); if (llt == NULL) return (ESRCH); #ifdef INET6 if (dst->sa_family == AF_INET6) { struct sockaddr_in6 *dst6 = (struct sockaddr_in6 *)dst; if (IN6_IS_SCOPE_LINKLOCAL(&dst6->sin6_addr)) in6_set_unicast_scopeid(&dst6->sin6_addr, ifp->if_index); } #endif struct llentry *lle = lla_lookup(llt, LLE_UNLOCKED, dst); if (lle == NULL) return (ESRCH); wa->ifp = ifp; wa->family = family; return (dump_lle(llt, lle, wa)); } struct nl_parsed_neigh { struct sockaddr *nda_dst; struct ifnet *nda_ifp; struct nlattr *nda_lladdr; uint32_t ndm_flags; uint16_t ndm_state; uint8_t ndm_family; }; #define _IN(_field) offsetof(struct ndmsg, _field) #define _OUT(_field) offsetof(struct nl_parsed_neigh, _field) static struct nlfield_parser nlf_p_neigh[] = { { .off_in = _IN(ndm_family), .off_out = _OUT(ndm_family), .cb = nlf_get_u8 }, { .off_in = _IN(ndm_flags), .off_out = _OUT(ndm_flags), .cb = nlf_get_u8_u32 }, { .off_in = _IN(ndm_state), .off_out = _OUT(ndm_state), .cb = nlf_get_u16 }, { .off_in = _IN(ndm_ifindex), .off_out = _OUT(nda_ifp), .cb = nlf_get_ifpz }, }; static struct nlattr_parser nla_p_neigh[] = { { .type = NDA_DST, .off = _OUT(nda_dst), .cb = nlattr_get_ip }, { .type = NDA_LLADDR, .off = _OUT(nda_lladdr), .cb = nlattr_get_nla }, { .type = NDA_IFINDEX, .off = _OUT(nda_ifp), .cb = nlattr_get_ifp }, { .type = NDA_FLAGS_EXT, .off = _OUT(ndm_flags), .cb = nlattr_get_uint32 }, }; #undef _IN #undef _OUT NL_DECLARE_PARSER(ndmsg_parser, struct ndmsg, nlf_p_neigh, nla_p_neigh); /* * type=RTM_NEWNEIGH, flags=NLM_F_REQUEST|NLM_F_ACK|NLM_F_EXCL|NLM_F_CREATE, seq=1661941473, pid=0}, * {ndm_family=AF_INET6, ndm_ifindex=if_nametoindex("enp0s31f6"), ndm_state=NUD_PERMANENT, ndm_flags=0, ndm_type=RTN_UNSPEC}, * [ * {{nla_len=20, nla_type=NDA_DST}, inet_pton(AF_INET6, "2a01:4f8:13a:70c::3")}, * {{nla_len=10, nla_type=NDA_LLADDR}, 20:4e:71:62:ae:f2}]}, iov_len=60} */ static int rtnl_handle_newneigh(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { int error; struct nl_parsed_neigh attrs = {}; error = nl_parse_nlmsg(hdr, &ndmsg_parser, npt, &attrs); if (error != 0) return (error); if (attrs.nda_ifp == NULL || attrs.nda_dst == NULL || attrs.nda_lladdr == NULL) { if (attrs.nda_ifp == NULL) NLMSG_REPORT_ERR_MSG(npt, "NDA_IFINDEX / ndm_ifindex not set"); if (attrs.nda_dst == NULL) NLMSG_REPORT_ERR_MSG(npt, "NDA_DST not set"); if (attrs.nda_lladdr == NULL) NLMSG_REPORT_ERR_MSG(npt, "NDA_LLADDR not set"); return (EINVAL); } if (attrs.nda_dst->sa_family != attrs.ndm_family) { NLMSG_REPORT_ERR_MSG(npt, "NDA_DST family (%d) is different from ndm_family (%d)", attrs.nda_dst->sa_family, attrs.ndm_family); return (EINVAL); } int addrlen = attrs.nda_ifp->if_addrlen; if (attrs.nda_lladdr->nla_len != sizeof(struct nlattr) + addrlen) { NLMSG_REPORT_ERR_MSG(npt, "NDA_LLADDR address length (%d) is different from expected (%d)", (int)attrs.nda_lladdr->nla_len - (int)sizeof(struct nlattr), addrlen); return (EINVAL); } if (attrs.ndm_state != NUD_PERMANENT) { NLMSG_REPORT_ERR_MSG(npt, "ndm_state %d not supported", attrs.ndm_state); return (ENOTSUP); } const uint16_t supported_flags = NTF_PROXY | NTF_STICKY; if ((attrs.ndm_flags & supported_flags) != attrs.ndm_flags) { NLMSG_REPORT_ERR_MSG(npt, "ndm_flags %X not supported", attrs.ndm_flags &~ supported_flags); return (ENOTSUP); } /* Replacement requires new entry creation anyway */ if ((hdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_REPLACE)) == 0) return (ENOTSUP); struct lltable *llt = lltable_get(attrs.nda_ifp, attrs.ndm_family); if (llt == NULL) return (EAFNOSUPPORT); uint8_t linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize = sizeof(linkhdr); int lladdr_off = 0; if (lltable_calc_llheader(attrs.nda_ifp, attrs.ndm_family, (char *)(attrs.nda_lladdr + 1), linkhdr, &linkhdrsize, &lladdr_off) != 0) { NLMSG_REPORT_ERR_MSG(npt, "unable to calculate lle prepend data"); return (EINVAL); } int lle_flags = LLE_STATIC | ((attrs.ndm_flags & NTF_PROXY) ? LLE_PUB : 0); struct llentry *lle = lltable_alloc_entry(llt, lle_flags, attrs.nda_dst); if (lle == NULL) return (ENOMEM); lltable_set_entry_addr(attrs.nda_ifp, lle, linkhdr, linkhdrsize, lladdr_off); /* llentry created, try to insert or update :*/ IF_AFDATA_WLOCK(attrs.nda_ifp); LLE_WLOCK(lle); struct llentry *lle_tmp = lla_lookup(llt, LLE_EXCLUSIVE, attrs.nda_dst); if (lle_tmp != NULL) { if (hdr->nlmsg_flags & NLM_F_EXCL) { LLE_WUNLOCK(lle_tmp); lle_tmp = NULL; error = EEXIST; } else if (hdr->nlmsg_flags & NLM_F_REPLACE) { lltable_unlink_entry(llt, lle_tmp); lltable_link_entry(llt, lle); } else error = EEXIST; } else { if (hdr->nlmsg_flags & NLM_F_CREATE) lltable_link_entry(llt, lle); else error = ENOENT; } IF_AFDATA_WUNLOCK(attrs.nda_ifp); if (error != 0) { if (lle != NULL) llentry_free(lle); return (error); } if (lle_tmp != NULL) llentry_free(lle_tmp); /* XXX: We're inside epoch */ EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED); LLE_WUNLOCK(lle); return (0); } static int rtnl_handle_delneigh(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { int error; struct nl_parsed_neigh attrs = {}; error = nl_parse_nlmsg(hdr, &ndmsg_parser, npt, &attrs); if (error != 0) return (error); if (attrs.nda_dst == NULL) { NLMSG_REPORT_ERR_MSG(npt, "NDA_DST not set"); return (EINVAL); } if (attrs.nda_ifp == NULL) { NLMSG_REPORT_ERR_MSG(npt, "no ifindex provided"); return (EINVAL); } struct lltable *llt = lltable_get(attrs.nda_ifp, attrs.ndm_family); if (llt == NULL) return (EAFNOSUPPORT); IF_AFDATA_WLOCK(attrs.nda_ifp); struct llentry *lle = lla_lookup(llt, LLE_EXCLUSIVE, attrs.nda_dst); if (lle != NULL) { if ((lle->la_flags & LLE_IFADDR) != 0) { LLE_WUNLOCK(lle); lle = NULL; error = EPERM; } else lltable_unlink_entry(llt, lle); } else error = ENOENT; IF_AFDATA_WUNLOCK(attrs.nda_ifp); if (error == 0 && lle != NULL) EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_DELETED); if (lle != NULL) llentry_free(lle); return (error); } static int rtnl_handle_getneigh(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { int error; struct nl_parsed_neigh attrs = {}; error = nl_parse_nlmsg(hdr, &ndmsg_parser, npt, &attrs); if (error != 0) return (error); if (attrs.nda_dst != NULL && attrs.nda_ifp == NULL) { NLMSG_REPORT_ERR_MSG(npt, "has NDA_DST but no ifindex provided"); return (EINVAL); } struct netlink_walkargs wa = { .so = nlp, .nw = npt->nw, .hdr.nlmsg_pid = hdr->nlmsg_pid, .hdr.nlmsg_seq = hdr->nlmsg_seq, .hdr.nlmsg_flags = hdr->nlmsg_flags, .hdr.nlmsg_type = NL_RTM_NEWNEIGH, }; if (attrs.nda_dst == NULL) error = dump_llts(&wa, attrs.nda_ifp, attrs.ndm_family); else error = get_lle(&wa, attrs.nda_ifp, attrs.ndm_family, attrs.nda_dst); return (error); } static const struct rtnl_cmd_handler cmd_handlers[] = { { .cmd = NL_RTM_NEWNEIGH, .name = "RTM_NEWNEIGH", .cb = &rtnl_handle_newneigh, }, { .cmd = NL_RTM_DELNEIGH, .name = "RTM_DELNEIGH", .cb = &rtnl_handle_delneigh, .priv = PRIV_NET_ROUTE, }, { .cmd = NL_RTM_GETNEIGH, .name = "RTM_GETNEIGH", .cb = &rtnl_handle_getneigh, .priv = PRIV_NET_ROUTE, } }; static void rtnl_lle_event(void *arg __unused, struct llentry *lle, int evt) { struct ifnet *ifp; int family; LLE_WLOCK_ASSERT(lle); ifp = lltable_get_ifp(lle->lle_tbl); family = lltable_get_af(lle->lle_tbl); if (family != AF_INET && family != AF_INET6) return; int nlmsgs_type = evt == LLENTRY_RESOLVED ? NL_RTM_NEWNEIGH : NL_RTM_DELNEIGH; struct nl_writer nw = {}; if (!nlmsg_get_group_writer(&nw, NLMSG_SMALL, NETLINK_ROUTE, RTNLGRP_NEIGH)) { NL_LOG(LOG_DEBUG, "error allocating group writer"); return; } struct netlink_walkargs wa = { .hdr.nlmsg_type = nlmsgs_type, .nw = &nw, .ifp = ifp, .family = family, }; dump_lle_locked(lle, &wa); nlmsg_flush(&nw); } static const struct nlhdr_parser *all_parsers[] = { &ndmsg_parser }; void -rtnl_neighs_init() +rtnl_neighs_init(void) { NL_VERIFY_PARSERS(all_parsers); rtnl_register_messages(cmd_handlers, NL_ARRAY_LEN(cmd_handlers)); lle_event_p = EVENTHANDLER_REGISTER(lle_event, rtnl_lle_event, NULL, EVENTHANDLER_PRI_ANY); } void -rtnl_neighs_destroy() +rtnl_neighs_destroy(void) { EVENTHANDLER_DEREGISTER(lle_event, lle_event_p); } diff --git a/sys/netlink/route/nexthop.c b/sys/netlink/route/nexthop.c index 31816d84f189..64f46ad035fc 100644 --- a/sys/netlink/route/nexthop.c +++ b/sys/netlink/route/nexthop.c @@ -1,1005 +1,1005 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2022 Alexander V. Chernikov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_route.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define DEBUG_MOD_NAME nl_nhop #define DEBUG_MAX_LEVEL LOG_DEBUG3 #include _DECLARE_DEBUG(LOG_DEBUG3); /* * This file contains the logic to maintain kernel nexthops and * nexhop groups based om the data provided by the user. * * Kernel stores (nearly) all of the routing data in the nexthops, * including the prefix-specific flags (NHF_HOST and NHF_DEFAULT). * * Netlink API provides higher-level abstraction for the user. Each * user-created nexthop may map to multiple kernel nexthops. * * The following variations require separate kernel nexthop to be * created: * * prefix flags (NHF_HOST, NHF_DEFAULT) * * using IPv6 gateway for IPv4 routes * * different fibnum * * These kernel nexthops have the lifetime bound to the lifetime of * the user_nhop object. They are not collected until user requests * to delete the created user_nhop. * */ struct user_nhop { uint32_t un_idx; /* Userland-provided index */ uint32_t un_fibfam; /* fibnum+af(as highest byte) */ uint8_t un_protocol; /* protocol that install the record */ struct nhop_object *un_nhop; /* "production" nexthop */ struct nhop_object *un_nhop_src; /* nexthop to copy from */ struct weightened_nhop *un_nhgrp_src; /* nexthops for nhg */ uint32_t un_nhgrp_count; /* number of nexthops */ struct user_nhop *un_next; /* next item in hash chain */ struct user_nhop *un_nextchild; /* master -> children */ struct epoch_context un_epoch_ctx; /* epoch ctl helper */ }; /* produce hash value for an object */ #define unhop_hash_obj(_obj) (hash_unhop(_obj)) /* compare two objects */ #define unhop_cmp(_one, _two) (cmp_unhop(_one, _two)) /* next object accessor */ #define unhop_next(_obj) (_obj)->un_next CHT_SLIST_DEFINE(unhop, struct user_nhop); struct unhop_ctl { struct unhop_head un_head; struct rmlock un_lock; }; #define UN_LOCK_INIT(_ctl) rm_init(&(_ctl)->un_lock, "unhop_ctl") #define UN_TRACKER struct rm_priotracker un_tracker #define UN_RLOCK(_ctl) rm_rlock(&((_ctl)->un_lock), &un_tracker) #define UN_RUNLOCK(_ctl) rm_runlock(&((_ctl)->un_lock), &un_tracker) #define UN_WLOCK(_ctl) rm_wlock(&(_ctl)->un_lock); #define UN_WUNLOCK(_ctl) rm_wunlock(&(_ctl)->un_lock); VNET_DEFINE_STATIC(struct unhop_ctl *, un_ctl) = NULL; #define V_un_ctl VNET(un_ctl) static void consider_resize(struct unhop_ctl *ctl, uint32_t new_size); static int cmp_unhop(const struct user_nhop *a, const struct user_nhop *b); static unsigned int hash_unhop(const struct user_nhop *obj); static void destroy_unhop(struct user_nhop *unhop); static struct nhop_object *clone_unhop(const struct user_nhop *unhop, uint32_t fibnum, int family, int nh_flags); static int cmp_unhop(const struct user_nhop *a, const struct user_nhop *b) { return (a->un_idx == b->un_idx && a->un_fibfam == b->un_fibfam); } /* * Hash callback: calculate hash of an object */ static unsigned int hash_unhop(const struct user_nhop *obj) { return (obj->un_idx ^ obj->un_fibfam); } #define UNHOP_IS_MASTER(_unhop) ((_unhop)->un_fibfam == 0) /* * Factory interface for creating matching kernel nexthops/nexthop groups * * @uidx: userland nexhop index used to create the nexthop * @fibnum: fibnum nexthop will be used in * @family: upper family nexthop will be used in * @nh_flags: desired nexthop prefix flags * @perror: pointer to store error to * * Returns referenced nexthop linked to @fibnum/@family rib on success. */ struct nhop_object * nl_find_nhop(uint32_t fibnum, int family, uint32_t uidx, int nh_flags, int *perror) { struct unhop_ctl *ctl = atomic_load_ptr(&V_un_ctl); UN_TRACKER; if (__predict_false(ctl == NULL)) return (NULL); struct user_nhop key= { .un_idx = uidx, .un_fibfam = fibnum | ((uint32_t)family) << 24, }; struct user_nhop *unhop; nh_flags = nh_flags & (NHF_HOST | NHF_DEFAULT); if (__predict_false(family == 0)) return (NULL); UN_RLOCK(ctl); CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop); if (unhop != NULL) { struct nhop_object *nh = unhop->un_nhop; UN_RLOCK(ctl); *perror = 0; nhop_ref_any(nh); return (nh); } /* * Exact nexthop not found. Search for template nexthop to clone from. */ key.un_fibfam = 0; CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop); if (unhop == NULL) { UN_RUNLOCK(ctl); *perror = ESRCH; return (NULL); } UN_RUNLOCK(ctl); /* Create entry to insert first */ struct user_nhop *un_new, *un_tmp; un_new = malloc(sizeof(struct user_nhop), M_NETLINK, M_NOWAIT | M_ZERO); if (un_new == NULL) { *perror = ENOMEM; return (NULL); } un_new->un_idx = uidx; un_new->un_fibfam = fibnum | ((uint32_t)family) << 24; /* Relying on epoch to protect unhop here */ un_new->un_nhop = clone_unhop(unhop, fibnum, family, nh_flags); if (un_new->un_nhop == NULL) { free(un_new, M_NETLINK); *perror = ENOMEM; return (NULL); } /* Insert back and report */ UN_WLOCK(ctl); /* First, find template record once again */ CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop); if (unhop == NULL) { /* Someone deleted the nexthop during the call */ UN_WUNLOCK(ctl); *perror = ESRCH; destroy_unhop(un_new); return (NULL); } /* Second, check the direct match */ CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, un_new, un_tmp); struct nhop_object *nh; if (un_tmp != NULL) { /* Another thread already created the desired nextop, use it */ nh = un_tmp->un_nhop; } else { /* Finally, insert the new nexthop and link it to the primary */ nh = un_new->un_nhop; CHT_SLIST_INSERT_HEAD(&ctl->un_head, unhop, un_new); un_new->un_nextchild = unhop->un_nextchild; unhop->un_nextchild = un_new; un_new = NULL; NL_LOG(LOG_DEBUG2, "linked cloned nexthop %p", nh); } UN_WUNLOCK(ctl); if (un_new != NULL) destroy_unhop(un_new); *perror = 0; nhop_ref_any(nh); return (nh); } static struct user_nhop * nl_find_base_unhop(struct unhop_ctl *ctl, uint32_t uidx) { struct user_nhop key= { .un_idx = uidx }; struct user_nhop *unhop = NULL; UN_TRACKER; UN_RLOCK(ctl); CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop); UN_RUNLOCK(ctl); return (unhop); } #define MAX_STACK_NHOPS 4 static struct nhop_object * clone_unhop(const struct user_nhop *unhop, uint32_t fibnum, int family, int nh_flags) { #ifdef ROUTE_MPATH const struct weightened_nhop *wn; struct weightened_nhop *wn_new, wn_base[MAX_STACK_NHOPS]; uint32_t num_nhops; #endif struct nhop_object *nh = NULL; int error; if (unhop->un_nhop_src != NULL) { IF_DEBUG_LEVEL(LOG_DEBUG2) { char nhbuf[NHOP_PRINT_BUFSIZE]; nhop_print_buf_any(unhop->un_nhop_src, nhbuf, sizeof(nhbuf)); FIB_NH_LOG(LOG_DEBUG2, unhop->un_nhop_src, "cloning nhop %s -> %u.%u flags 0x%X", nhbuf, fibnum, family, nh_flags); } struct nhop_object *nh; nh = nhop_alloc(fibnum, AF_UNSPEC); if (nh == NULL) return (NULL); nhop_copy(nh, unhop->un_nhop_src); /* Check that nexthop gateway is compatible with the new family */ if (!nhop_set_upper_family(nh, family)) { nhop_free(nh); return (NULL); } nhop_set_uidx(nh, unhop->un_idx); nhop_set_pxtype_flag(nh, nh_flags); return (nhop_get_nhop(nh, &error)); } #ifdef ROUTE_MPATH wn = unhop->un_nhgrp_src; num_nhops = unhop->un_nhgrp_count; if (num_nhops > MAX_STACK_NHOPS) { wn_new = malloc(num_nhops * sizeof(struct weightened_nhop), M_TEMP, M_NOWAIT); if (wn_new == NULL) return (NULL); } else wn_new = wn_base; for (int i = 0; i < num_nhops; i++) { uint32_t uidx = nhop_get_uidx(wn[i].nh); MPASS(uidx != 0); wn_new[i].nh = nl_find_nhop(fibnum, family, uidx, nh_flags, &error); if (error != 0) break; wn_new[i].weight = wn[i].weight; } if (error == 0) { struct rib_head *rh = nhop_get_rh(wn_new[0].nh); struct nhgrp_object *nhg; error = nhgrp_get_group(rh, wn_new, num_nhops, unhop->un_idx, &nhg); nh = (struct nhop_object *)nhg; } if (wn_new != wn_base) free(wn_new, M_TEMP); #endif return (nh); } static void destroy_unhop(struct user_nhop *unhop) { if (unhop->un_nhop != NULL) nhop_free_any(unhop->un_nhop); if (unhop->un_nhop_src != NULL) nhop_free_any(unhop->un_nhop_src); free(unhop, M_NETLINK); } static void destroy_unhop_epoch(epoch_context_t ctx) { struct user_nhop *unhop; unhop = __containerof(ctx, struct user_nhop, un_epoch_ctx); destroy_unhop(unhop); } static uint32_t find_spare_uidx(struct unhop_ctl *ctl) { struct user_nhop *unhop, key = {}; uint32_t uidx = 0; UN_TRACKER; UN_RLOCK(ctl); /* This should return spare uid with 75% of 65k used in ~99/100 cases */ for (int i = 0; i < 16; i++) { key.un_idx = (arc4random() % 65536) + 65536 * 4; CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop); if (unhop == NULL) { uidx = key.un_idx; break; } } UN_RUNLOCK(ctl); return (uidx); } /* * Actual netlink code */ struct netlink_walkargs { struct nl_writer *nw; struct nlmsghdr hdr; struct nlpcb *so; int family; int error; int count; int dumped; }; #define ENOMEM_IF_NULL(_v) if ((_v) == NULL) goto enomem static bool dump_nhgrp(const struct user_nhop *unhop, struct nlmsghdr *hdr, struct nl_writer *nw) { if (!nlmsg_reply(nw, hdr, sizeof(struct nhmsg))) goto enomem; struct nhmsg *nhm = nlmsg_reserve_object(nw, struct nhmsg); nhm->nh_family = AF_UNSPEC; nhm->nh_scope = 0; nhm->nh_protocol = unhop->un_protocol; nhm->nh_flags = 0; nlattr_add_u32(nw, NHA_ID, unhop->un_idx); nlattr_add_u16(nw, NHA_GROUP_TYPE, NEXTHOP_GRP_TYPE_MPATH); struct weightened_nhop *wn = unhop->un_nhgrp_src; uint32_t num_nhops = unhop->un_nhgrp_count; /* TODO: a better API? */ int nla_len = sizeof(struct nlattr); nla_len += NETLINK_ALIGN(num_nhops * sizeof(struct nexthop_grp)); struct nlattr *nla = nlmsg_reserve_data(nw, nla_len, struct nlattr); if (nla == NULL) goto enomem; nla->nla_type = NHA_GROUP; nla->nla_len = nla_len; for (int i = 0; i < num_nhops; i++) { struct nexthop_grp *grp = &((struct nexthop_grp *)(nla + 1))[i]; grp->id = nhop_get_uidx(wn[i].nh); grp->weight = wn[i].weight; grp->resvd1 = 0; grp->resvd2 = 0; } if (nlmsg_end(nw)) return (true); enomem: NL_LOG(LOG_DEBUG, "error: unable to allocate attribute memory"); nlmsg_abort(nw); return (false); } static bool dump_nhop(const struct user_nhop *unhop, struct nlmsghdr *hdr, struct nl_writer *nw) { struct nhop_object *nh = unhop->un_nhop_src; if (!nlmsg_reply(nw, hdr, sizeof(struct nhmsg))) goto enomem; struct nhmsg *nhm = nlmsg_reserve_object(nw, struct nhmsg); ENOMEM_IF_NULL(nhm); nhm->nh_family = nhop_get_neigh_family(nh); nhm->nh_scope = 0; // XXX: what's that? nhm->nh_protocol = unhop->un_protocol; nhm->nh_flags = 0; nlattr_add_u32(nw, NHA_ID, unhop->un_idx); if (nh->nh_flags & NHF_BLACKHOLE) { nlattr_add_flag(nw, NHA_BLACKHOLE); goto done; } nlattr_add_u32(nw, NHA_OIF, nh->nh_ifp->if_index); switch (nh->gw_sa.sa_family) { #ifdef INET case AF_INET: nlattr_add(nw, NHA_GATEWAY, 4, &nh->gw4_sa.sin_addr); break; #endif #ifdef INET6 case AF_INET6: { struct in6_addr addr = nh->gw6_sa.sin6_addr; in6_clearscope(&addr); nlattr_add(nw, NHA_GATEWAY, 16, &addr); break; } #endif } done: if (nlmsg_end(nw)) return (true); enomem: nlmsg_abort(nw); return (false); } static void dump_unhop(const struct user_nhop *unhop, struct nlmsghdr *hdr, struct nl_writer *nw) { if (unhop->un_nhop_src != NULL) dump_nhop(unhop, hdr, nw); else dump_nhgrp(unhop, hdr, nw); } static int delete_unhop(struct unhop_ctl *ctl, struct nlmsghdr *hdr, uint32_t uidx) { struct user_nhop *unhop_ret, *unhop_base, *unhop_chain; struct user_nhop key = { .un_idx = uidx }; UN_WLOCK(ctl); CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop_base); if (unhop_base != NULL) { CHT_SLIST_REMOVE(&ctl->un_head, unhop, unhop_base, unhop_ret); IF_DEBUG_LEVEL(LOG_DEBUG2) { char nhbuf[NHOP_PRINT_BUFSIZE]; nhop_print_buf_any(unhop_base->un_nhop, nhbuf, sizeof(nhbuf)); FIB_NH_LOG(LOG_DEBUG3, unhop_base->un_nhop, "removed base nhop %u: %s", uidx, nhbuf); } /* Unlink all child nexhops as well, keeping the chain intact */ unhop_chain = unhop_base->un_nextchild; while (unhop_chain != NULL) { CHT_SLIST_REMOVE(&ctl->un_head, unhop, unhop_chain, unhop_ret); MPASS(unhop_chain == unhop_ret); IF_DEBUG_LEVEL(LOG_DEBUG3) { char nhbuf[NHOP_PRINT_BUFSIZE]; nhop_print_buf_any(unhop_chain->un_nhop, nhbuf, sizeof(nhbuf)); FIB_NH_LOG(LOG_DEBUG3, unhop_chain->un_nhop, "removed child nhop %u: %s", uidx, nhbuf); } unhop_chain = unhop_chain->un_nextchild; } } UN_WUNLOCK(ctl); if (unhop_base == NULL) { NL_LOG(LOG_DEBUG, "unable to find unhop %u", uidx); return (ENOENT); } /* Report nexthop deletion */ struct netlink_walkargs wa = { .hdr.nlmsg_pid = hdr->nlmsg_pid, .hdr.nlmsg_seq = hdr->nlmsg_seq, .hdr.nlmsg_flags = hdr->nlmsg_flags, .hdr.nlmsg_type = NL_RTM_DELNEXTHOP, }; struct nl_writer nw = {}; if (!nlmsg_get_group_writer(&nw, NLMSG_SMALL, NETLINK_ROUTE, RTNLGRP_NEXTHOP)) { NL_LOG(LOG_DEBUG, "error allocating message writer"); return (ENOMEM); } dump_unhop(unhop_base, &wa.hdr, &nw); nlmsg_flush(&nw); while (unhop_base != NULL) { unhop_chain = unhop_base->un_nextchild; epoch_call(net_epoch_preempt, destroy_unhop_epoch, &unhop_base->un_epoch_ctx); unhop_base = unhop_chain; } return (0); } static void consider_resize(struct unhop_ctl *ctl, uint32_t new_size) { void *new_ptr = NULL; size_t alloc_size; if (new_size == 0) return; if (new_size != 0) { alloc_size = CHT_SLIST_GET_RESIZE_SIZE(new_size); new_ptr = malloc(alloc_size, M_NETLINK, M_NOWAIT | M_ZERO); if (new_ptr == NULL) return; } NL_LOG(LOG_DEBUG, "resizing hash: %u -> %u", ctl->un_head.hash_size, new_size); UN_WLOCK(ctl); if (new_ptr != NULL) { CHT_SLIST_RESIZE(&ctl->un_head, unhop, new_ptr, new_size); } UN_WUNLOCK(ctl); if (new_ptr != NULL) free(new_ptr, M_NETLINK); } static bool __noinline -vnet_init_unhops() +vnet_init_unhops(void) { uint32_t num_buckets = 16; size_t alloc_size = CHT_SLIST_GET_RESIZE_SIZE(num_buckets); struct unhop_ctl *ctl = malloc(sizeof(struct unhop_ctl), M_NETLINK, M_NOWAIT | M_ZERO); if (ctl == NULL) return (false); void *ptr = malloc(alloc_size, M_NETLINK, M_NOWAIT | M_ZERO); if (ptr == NULL) { free(ctl, M_NETLINK); return (false); } CHT_SLIST_INIT(&ctl->un_head, ptr, num_buckets); UN_LOCK_INIT(ctl); if (!atomic_cmpset_ptr((uintptr_t *)&V_un_ctl, (uintptr_t)NULL, (uintptr_t)ctl)) { free(ptr, M_NETLINK); free(ctl, M_NETLINK); } if (atomic_load_ptr(&V_un_ctl) == NULL) return (false); NL_LOG(LOG_NOTICE, "UNHOPS init done"); return (true); } static void vnet_destroy_unhops(const void *unused __unused) { struct unhop_ctl *ctl = atomic_load_ptr(&V_un_ctl); struct user_nhop *unhop, *tmp; if (ctl == NULL) return; V_un_ctl = NULL; /* Wait till all unhop users finish their reads */ epoch_wait_preempt(net_epoch_preempt); UN_WLOCK(ctl); CHT_SLIST_FOREACH_SAFE(&ctl->un_head, unhop, unhop, tmp) { destroy_unhop(unhop); } CHT_SLIST_FOREACH_SAFE_END; UN_WUNLOCK(ctl); free(ctl->un_head.ptr, M_NETLINK); free(ctl, M_NETLINK); } VNET_SYSUNINIT(vnet_destroy_unhops, SI_SUB_PROTO_IF, SI_ORDER_ANY, vnet_destroy_unhops, NULL); static int nlattr_get_nhg(struct nlattr *nla, struct nl_pstate *npt, const void *arg, void *target) { int error = 0; /* Verify attribute correctness */ struct nexthop_grp *grp = NLA_DATA(nla); int data_len = NLA_DATA_LEN(nla); int count = data_len / sizeof(*grp); if (count == 0 || (count * sizeof(*grp) != data_len)) { NL_LOG(LOG_DEBUG, "Invalid length for RTA_GROUP: %d", data_len); return (EINVAL); } *((struct nlattr **)target) = nla; return (error); } struct nl_parsed_nhop { uint32_t nha_id; uint8_t nha_blackhole; uint8_t nha_groups; struct ifnet *nha_oif; struct sockaddr *nha_gw; struct nlattr *nha_group; uint8_t nh_family; uint8_t nh_protocol; }; #define _IN(_field) offsetof(struct nhmsg, _field) #define _OUT(_field) offsetof(struct nl_parsed_nhop, _field) static const struct nlfield_parser nlf_p_nh[] = { { .off_in = _IN(nh_family), .off_out = _OUT(nh_family), .cb = nlf_get_u8 }, { .off_in = _IN(nh_protocol), .off_out = _OUT(nh_protocol), .cb = nlf_get_u8 }, }; static const struct nlattr_parser nla_p_nh[] = { { .type = NHA_ID, .off = _OUT(nha_id), .cb = nlattr_get_uint32 }, { .type = NHA_GROUP, .off = _OUT(nha_group), .cb = nlattr_get_nhg }, { .type = NHA_BLACKHOLE, .off = _OUT(nha_blackhole), .cb = nlattr_get_flag }, { .type = NHA_OIF, .off = _OUT(nha_oif), .cb = nlattr_get_ifp }, { .type = NHA_GATEWAY, .off = _OUT(nha_gw), .cb = nlattr_get_ip }, { .type = NHA_GROUPS, .off = _OUT(nha_groups), .cb = nlattr_get_flag }, }; #undef _IN #undef _OUT NL_DECLARE_PARSER(nhmsg_parser, struct nhmsg, nlf_p_nh, nla_p_nh); static bool eligible_nhg(const struct nhop_object *nh) { return (nh->nh_flags & NHF_GATEWAY); } static int newnhg(struct unhop_ctl *ctl, struct nl_parsed_nhop *attrs, struct user_nhop *unhop) { struct nexthop_grp *grp = NLA_DATA(attrs->nha_group); int count = NLA_DATA_LEN(attrs->nha_group) / sizeof(*grp); struct weightened_nhop *wn; wn = malloc(sizeof(*wn) * count, M_NETLINK, M_NOWAIT | M_ZERO); if (wn == NULL) return (ENOMEM); for (int i = 0; i < count; i++) { struct user_nhop *unhop; unhop = nl_find_base_unhop(ctl, grp[i].id); if (unhop == NULL) { NL_LOG(LOG_DEBUG, "unable to find uidx %u", grp[i].id); free(wn, M_NETLINK); return (ESRCH); } else if (unhop->un_nhop_src == NULL) { NL_LOG(LOG_DEBUG, "uidx %u is a group, nested group unsupported", grp[i].id); free(wn, M_NETLINK); return (ENOTSUP); } else if (!eligible_nhg(unhop->un_nhop_src)) { NL_LOG(LOG_DEBUG, "uidx %u nhop is not mpath-eligible", grp[i].id); free(wn, M_NETLINK); return (ENOTSUP); } /* * TODO: consider more rigid eligibility checks: * restrict nexthops with the same gateway */ wn[i].nh = unhop->un_nhop_src; wn[i].weight = grp[i].weight; } unhop->un_nhgrp_src = wn; unhop->un_nhgrp_count = count; return (0); } static int newnhop(struct nl_parsed_nhop *attrs, struct user_nhop *unhop) { struct ifaddr *ifa = NULL; struct nhop_object *nh; int error; if (!attrs->nha_blackhole) { if (attrs->nha_gw == NULL) { NL_LOG(LOG_DEBUG, "missing NHA_GATEWAY"); return (EINVAL); } if (attrs->nha_oif == NULL) { NL_LOG(LOG_DEBUG, "missing NHA_OIF"); return (EINVAL); } if (ifa == NULL) ifa = ifaof_ifpforaddr(attrs->nha_gw, attrs->nha_oif); if (ifa == NULL) { NL_LOG(LOG_DEBUG, "Unable to determine default source IP"); return (EINVAL); } } int family = attrs->nha_gw != NULL ? attrs->nha_gw->sa_family : attrs->nh_family; nh = nhop_alloc(RT_DEFAULT_FIB, family); if (nh == NULL) { NL_LOG(LOG_DEBUG, "Unable to allocate nexthop"); return (ENOMEM); } nhop_set_uidx(nh, attrs->nha_id); if (attrs->nha_blackhole) nhop_set_blackhole(nh, NHF_BLACKHOLE); else { nhop_set_gw(nh, attrs->nha_gw, true); nhop_set_transmit_ifp(nh, attrs->nha_oif); nhop_set_src(nh, ifa); } error = nhop_get_unlinked(nh); if (error != 0) { NL_LOG(LOG_DEBUG, "unable to finalize nexthop"); return (error); } IF_DEBUG_LEVEL(LOG_DEBUG2) { char nhbuf[NHOP_PRINT_BUFSIZE]; nhop_print_buf(nh, nhbuf, sizeof(nhbuf)); NL_LOG(LOG_DEBUG2, "Adding unhop %u: %s", attrs->nha_id, nhbuf); } unhop->un_nhop_src = nh; return (0); } static int rtnl_handle_newnhop(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { struct user_nhop *unhop; int error; if ((__predict_false(V_un_ctl == NULL)) && (!vnet_init_unhops())) return (ENOMEM); struct unhop_ctl *ctl = V_un_ctl; struct nl_parsed_nhop attrs = {}; error = nl_parse_nlmsg(hdr, &nhmsg_parser, npt, &attrs); if (error != 0) return (error); /* * Get valid nha_id. Treat nha_id == 0 (auto-assignment) as a second-class * citizen. */ if (attrs.nha_id == 0) { attrs.nha_id = find_spare_uidx(ctl); if (attrs.nha_id == 0) { NL_LOG(LOG_DEBUG, "Unable to get spare uidx"); return (ENOSPC); } } NL_LOG(LOG_DEBUG, "IFINDEX %d", attrs.nha_oif ? attrs.nha_oif->if_index : 0); unhop = malloc(sizeof(struct user_nhop), M_NETLINK, M_NOWAIT | M_ZERO); if (unhop == NULL) { NL_LOG(LOG_DEBUG, "Unable to allocate user_nhop"); return (ENOMEM); } unhop->un_idx = attrs.nha_id; unhop->un_protocol = attrs.nh_protocol; if (attrs.nha_group) error = newnhg(ctl, &attrs, unhop); else error = newnhop(&attrs, unhop); if (error != 0) { free(unhop, M_NETLINK); return (error); } UN_WLOCK(ctl); /* Check if uidx already exists */ struct user_nhop *tmp = NULL; CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, unhop, tmp); if (tmp != NULL) { UN_WUNLOCK(ctl); NL_LOG(LOG_DEBUG, "nhop idx %u already exists", attrs.nha_id); destroy_unhop(unhop); return (EEXIST); } CHT_SLIST_INSERT_HEAD(&ctl->un_head, unhop, unhop); uint32_t num_buckets_new = CHT_SLIST_GET_RESIZE_BUCKETS(&ctl->un_head); UN_WUNLOCK(ctl); /* Report addition of the next nexhop */ struct netlink_walkargs wa = { .hdr.nlmsg_pid = hdr->nlmsg_pid, .hdr.nlmsg_seq = hdr->nlmsg_seq, .hdr.nlmsg_flags = hdr->nlmsg_flags, .hdr.nlmsg_type = NL_RTM_NEWNEXTHOP, }; struct nl_writer nw = {}; if (!nlmsg_get_group_writer(&nw, NLMSG_SMALL, NETLINK_ROUTE, RTNLGRP_NEXTHOP)) { NL_LOG(LOG_DEBUG, "error allocating message writer"); return (ENOMEM); } dump_unhop(unhop, &wa.hdr, &nw); nlmsg_flush(&nw); consider_resize(ctl, num_buckets_new); return (0); } static int rtnl_handle_delnhop(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { struct unhop_ctl *ctl = atomic_load_ptr(&V_un_ctl); int error; if (__predict_false(ctl == NULL)) return (ESRCH); struct nl_parsed_nhop attrs = {}; error = nl_parse_nlmsg(hdr, &nhmsg_parser, npt, &attrs); if (error != 0) return (error); if (attrs.nha_id == 0) { NL_LOG(LOG_DEBUG, "NHA_ID not set"); return (EINVAL); } error = delete_unhop(ctl, hdr, attrs.nha_id); return (error); } static bool match_unhop(const struct nl_parsed_nhop *attrs, struct user_nhop *unhop) { if (attrs->nha_id != 0 && unhop->un_idx != attrs->nha_id) return (false); if (attrs->nha_groups != 0 && unhop->un_nhgrp_src == NULL) return (false); if (attrs->nha_oif != NULL && (unhop->un_nhop_src == NULL || unhop->un_nhop_src->nh_ifp != attrs->nha_oif)) return (false); return (true); } static int rtnl_handle_getnhop(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { struct unhop_ctl *ctl = atomic_load_ptr(&V_un_ctl); struct user_nhop *unhop; UN_TRACKER; int error; if (__predict_false(ctl == NULL)) return (ESRCH); struct nl_parsed_nhop attrs = {}; error = nl_parse_nlmsg(hdr, &nhmsg_parser, npt, &attrs); if (error != 0) return (error); struct netlink_walkargs wa = { .nw = npt->nw, .hdr.nlmsg_pid = hdr->nlmsg_pid, .hdr.nlmsg_seq = hdr->nlmsg_seq, .hdr.nlmsg_flags = hdr->nlmsg_flags, .hdr.nlmsg_type = NL_RTM_NEWNEXTHOP, }; if (attrs.nha_id != 0) { NL_LOG(LOG_DEBUG2, "searching for uidx %u", attrs.nha_id); struct user_nhop key= { .un_idx = attrs.nha_id }; UN_RLOCK(ctl); CHT_SLIST_FIND_BYOBJ(&ctl->un_head, unhop, &key, unhop); UN_RUNLOCK(ctl); if (unhop == NULL) return (ESRCH); dump_unhop(unhop, &wa.hdr, wa.nw); return (0); } UN_RLOCK(ctl); wa.hdr.nlmsg_flags |= NLM_F_MULTI; CHT_SLIST_FOREACH(&ctl->un_head, unhop, unhop) { if (UNHOP_IS_MASTER(unhop) && match_unhop(&attrs, unhop)) dump_unhop(unhop, &wa.hdr, wa.nw); } CHT_SLIST_FOREACH_END; UN_RUNLOCK(ctl); if (wa.error == 0) { if (!nlmsg_end_dump(wa.nw, wa.error, &wa.hdr)) return (ENOMEM); } return (0); } static const struct rtnl_cmd_handler cmd_handlers[] = { { .cmd = NL_RTM_NEWNEXTHOP, .name = "RTM_NEWNEXTHOP", .cb = &rtnl_handle_newnhop, .priv = PRIV_NET_ROUTE, }, { .cmd = NL_RTM_DELNEXTHOP, .name = "RTM_DELNEXTHOP", .cb = &rtnl_handle_delnhop, .priv = PRIV_NET_ROUTE, }, { .cmd = NL_RTM_GETNEXTHOP, .name = "RTM_GETNEXTHOP", .cb = &rtnl_handle_getnhop, } }; static const struct nlhdr_parser *all_parsers[] = { &nhmsg_parser }; void -rtnl_nexthops_init() +rtnl_nexthops_init(void) { NL_VERIFY_PARSERS(all_parsers); rtnl_register_messages(cmd_handlers, NL_ARRAY_LEN(cmd_handlers)); } diff --git a/sys/netlink/route/route.c b/sys/netlink/route/route.c index 7b4fc64ec4bd..852843af1b7d 100644 --- a/sys/netlink/route/route.c +++ b/sys/netlink/route/route.c @@ -1,988 +1,988 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2021 Ng Peng Nam Sean * Copyright (c) 2022 Alexander V. Chernikov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_route.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #define DEBUG_MOD_NAME nl_route #define DEBUG_MAX_LEVEL LOG_DEBUG3 #include _DECLARE_DEBUG(LOG_DEBUG); static unsigned char get_rtm_type(const struct nhop_object *nh) { int nh_flags = nh->nh_flags; /* Use the fact that nhg runtime flags are only NHF_MULTIPATH */ if (nh_flags & NHF_BLACKHOLE) return (RTN_BLACKHOLE); else if (nh_flags & NHF_REJECT) return (RTN_PROHIBIT); return (RTN_UNICAST); } static uint8_t nl_get_rtm_protocol(const struct nhop_object *nh) { #ifdef ROUTE_MPATH if (NH_IS_NHGRP(nh)) { const struct nhgrp_object *nhg = (const struct nhgrp_object *)nh; uint8_t origin = nhgrp_get_origin(nhg); if (origin != RTPROT_UNSPEC) return (origin); nh = nhg->nhops[0]; } #endif uint8_t origin = nhop_get_origin(nh); if (origin != RTPROT_UNSPEC) return (origin); /* TODO: remove guesswork once all kernel users fill in origin */ int rt_flags = nhop_get_rtflags(nh); if (rt_flags & RTF_PROTO1) return (RTPROT_ZEBRA); if (rt_flags & RTF_STATIC) return (RTPROT_STATIC); return (RTPROT_KERNEL); } static int get_rtmsg_type_from_rtsock(int cmd) { switch (cmd) { case RTM_ADD: case RTM_CHANGE: case RTM_GET: return NL_RTM_NEWROUTE; case RTM_DELETE: return NL_RTM_DELROUTE; } return (0); } /* * fibnum heuristics * * if (dump && rtm_table == 0 && !rta_table) RT_ALL_FIBS * msg rtm_table RTA_TABLE result * RTM_GETROUTE/dump 0 - RT_ALL_FIBS * RTM_GETROUTE/dump 1 - 1 * RTM_GETROUTE/get 0 - 0 * */ static struct nhop_object * rc_get_nhop(const struct rib_cmd_info *rc) { return ((rc->rc_cmd == RTM_DELETE) ? rc->rc_nh_old : rc->rc_nh_new); } static void dump_rc_nhop_gw(struct nl_writer *nw, const struct nhop_object *nh) { int upper_family; switch (nhop_get_neigh_family(nh)) { case AF_LINK: /* onlink prefix, skip */ break; case AF_INET: nlattr_add(nw, NL_RTA_GATEWAY, 4, &nh->gw4_sa.sin_addr); break; case AF_INET6: upper_family = nhop_get_upper_family(nh); if (upper_family == AF_INET6) { nlattr_add(nw, NL_RTA_GATEWAY, 16, &nh->gw6_sa.sin6_addr); } else if (upper_family == AF_INET) { /* IPv4 over IPv6 */ char buf[20]; struct rtvia *via = (struct rtvia *)&buf[0]; via->rtvia_family = AF_INET6; memcpy(via->rtvia_addr, &nh->gw6_sa.sin6_addr, 16); nlattr_add(nw, NL_RTA_VIA, 17, via); } break; } } static void dump_rc_nhop_mtu(struct nl_writer *nw, const struct nhop_object *nh) { int nla_len = sizeof(struct nlattr) * 2 + sizeof(uint32_t); struct nlattr *nla = nlmsg_reserve_data(nw, nla_len, struct nlattr); if (nla == NULL) return; nla->nla_type = NL_RTA_METRICS; nla->nla_len = nla_len; nla++; nla->nla_type = NL_RTAX_MTU; nla->nla_len = sizeof(struct nlattr) + sizeof(uint32_t); *((uint32_t *)(nla + 1)) = nh->nh_mtu; } #ifdef ROUTE_MPATH static void dump_rc_nhg(struct nl_writer *nw, const struct nhgrp_object *nhg, struct rtmsg *rtm) { uint32_t uidx = nhgrp_get_uidx(nhg); uint32_t num_nhops; const struct weightened_nhop *wn = nhgrp_get_nhops(nhg, &num_nhops); uint32_t base_rtflags = nhop_get_rtflags(wn[0].nh); if (uidx != 0) nlattr_add_u32(nw, NL_RTA_NH_ID, uidx); nlattr_add_u32(nw, NL_RTA_RTFLAGS, base_rtflags); int off = nlattr_add_nested(nw, NL_RTA_MULTIPATH); if (off == 0) return; for (int i = 0; i < num_nhops; i++) { int nh_off = nlattr_save_offset(nw); struct rtnexthop *rtnh = nlmsg_reserve_object(nw, struct rtnexthop); if (rtnh == NULL) return; rtnh->rtnh_flags = 0; rtnh->rtnh_ifindex = wn[i].nh->nh_ifp->if_index; rtnh->rtnh_hops = wn[i].weight; dump_rc_nhop_gw(nw, wn[i].nh); uint32_t rtflags = nhop_get_rtflags(wn[i].nh); if (rtflags != base_rtflags) nlattr_add_u32(nw, NL_RTA_RTFLAGS, rtflags); if (rtflags & RTF_FIXEDMTU) dump_rc_nhop_mtu(nw, wn[i].nh); rtnh = nlattr_restore_offset(nw, nh_off, struct rtnexthop); /* * nlattr_add() allocates 4-byte aligned storage, no need to aligh * length here * */ rtnh->rtnh_len = nlattr_save_offset(nw) - nh_off; } nlattr_set_len(nw, off); } #endif static void dump_rc_nhop(struct nl_writer *nw, const struct nhop_object *nh, struct rtmsg *rtm) { #ifdef ROUTE_MPATH if (NH_IS_NHGRP(nh)) { dump_rc_nhg(nw, (const struct nhgrp_object *)nh, rtm); return; } #endif uint32_t rtflags = nhop_get_rtflags(nh); /* * IPv4 over IPv6 * ('RTA_VIA', {'family': 10, 'addr': 'fe80::20c:29ff:fe67:2dd'}), ('RTA_OIF', 2), * IPv4 w/ gw * ('RTA_GATEWAY', '172.16.107.131'), ('RTA_OIF', 2)], * Direct route: * ('RTA_OIF', 2) */ if (nh->nh_flags & NHF_GATEWAY) dump_rc_nhop_gw(nw, nh); uint32_t uidx = nhop_get_uidx(nh); if (uidx != 0) nlattr_add_u32(nw, NL_RTA_NH_ID, uidx); nlattr_add_u32(nw, NL_RTA_KNH_ID, nhop_get_idx(nh)); nlattr_add_u32(nw, NL_RTA_RTFLAGS, rtflags); if (rtflags & RTF_FIXEDMTU) dump_rc_nhop_mtu(nw, nh); uint32_t nh_expire = nhop_get_expire(nh); if (nh_expire > 0) nlattr_add_u32(nw, NL_RTA_EXPIRES, nh_expire - time_uptime); /* In any case, fill outgoing interface */ nlattr_add_u32(nw, NL_RTA_OIF, nh->nh_ifp->if_index); } /* * Dumps output from a rib command into an rtmsg */ static int dump_px(uint32_t fibnum, const struct nlmsghdr *hdr, const struct rtentry *rt, struct route_nhop_data *rnd, struct nl_writer *nw) { struct rtmsg *rtm; int error = 0; NET_EPOCH_ASSERT(); if (!nlmsg_reply(nw, hdr, sizeof(struct rtmsg))) goto enomem; int family = rt_get_family(rt); int rtm_off = nlattr_save_offset(nw); rtm = nlmsg_reserve_object(nw, struct rtmsg); rtm->rtm_family = family; rtm->rtm_dst_len = 0; rtm->rtm_src_len = 0; rtm->rtm_tos = 0; if (fibnum < 255) rtm->rtm_table = (unsigned char)fibnum; rtm->rtm_scope = RT_SCOPE_UNIVERSE; if (!NH_IS_NHGRP(rnd->rnd_nhop)) { rtm->rtm_protocol = nl_get_rtm_protocol(rnd->rnd_nhop); rtm->rtm_type = get_rtm_type(rnd->rnd_nhop); } else { rtm->rtm_protocol = RTPROT_UNSPEC; /* TODO: protocol from nhg? */ rtm->rtm_type = RTN_UNICAST; } nlattr_add_u32(nw, NL_RTA_TABLE, fibnum); int plen = 0; #if defined(INET) || defined(INET6) uint32_t scopeid; #endif switch (family) { #ifdef INET case AF_INET: { struct in_addr addr; rt_get_inet_prefix_plen(rt, &addr, &plen, &scopeid); nlattr_add(nw, NL_RTA_DST, 4, &addr); break; } #endif #ifdef INET6 case AF_INET6: { struct in6_addr addr; rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid); nlattr_add(nw, NL_RTA_DST, 16, &addr); break; } #endif default: FIB_LOG(LOG_NOTICE, fibnum, family, "unsupported rt family: %d", family); error = EAFNOSUPPORT; goto flush; } rtm = nlattr_restore_offset(nw, rtm_off, struct rtmsg); if (plen > 0) rtm->rtm_dst_len = plen; dump_rc_nhop(nw, rnd->rnd_nhop, rtm); if (nlmsg_end(nw)) return (0); enomem: error = ENOMEM; flush: nlmsg_abort(nw); return (error); } static int family_to_group(int family) { switch (family) { case AF_INET: return (RTNLGRP_IPV4_ROUTE); case AF_INET6: return (RTNLGRP_IPV6_ROUTE); } return (0); } static void report_operation(uint32_t fibnum, struct rib_cmd_info *rc, struct nlpcb *nlp, struct nlmsghdr *hdr) { struct nl_writer nw; uint32_t group_id = family_to_group(rt_get_family(rc->rc_rt)); if (nlmsg_get_group_writer(&nw, NLMSG_SMALL, NETLINK_ROUTE, group_id)) { struct route_nhop_data rnd = { .rnd_nhop = rc_get_nhop(rc), .rnd_weight = rc->rc_nh_weight, }; hdr->nlmsg_flags &= ~(NLM_F_REPLACE | NLM_F_CREATE); hdr->nlmsg_flags &= ~(NLM_F_EXCL | NLM_F_APPEND); switch (rc->rc_cmd) { case RTM_ADD: hdr->nlmsg_type = NL_RTM_NEWROUTE; hdr->nlmsg_flags |= NLM_F_CREATE | NLM_F_EXCL; break; case RTM_CHANGE: hdr->nlmsg_type = NL_RTM_NEWROUTE; hdr->nlmsg_flags |= NLM_F_REPLACE; break; case RTM_DELETE: hdr->nlmsg_type = NL_RTM_DELROUTE; break; } dump_px(fibnum, hdr, rc->rc_rt, &rnd, &nw); nlmsg_flush(&nw); } rtsock_callback_p->route_f(fibnum, rc); } struct rta_mpath_nh { struct sockaddr *gw; struct ifnet *ifp; uint8_t rtnh_flags; uint8_t rtnh_weight; }; #define _IN(_field) offsetof(struct rtnexthop, _field) #define _OUT(_field) offsetof(struct rta_mpath_nh, _field) const static struct nlattr_parser nla_p_rtnh[] = { { .type = NL_RTA_GATEWAY, .off = _OUT(gw), .cb = nlattr_get_ip }, { .type = NL_RTA_VIA, .off = _OUT(gw), .cb = nlattr_get_ipvia }, }; const static struct nlfield_parser nlf_p_rtnh[] = { { .off_in = _IN(rtnh_flags), .off_out = _OUT(rtnh_flags), .cb = nlf_get_u8 }, { .off_in = _IN(rtnh_hops), .off_out = _OUT(rtnh_weight), .cb = nlf_get_u8 }, { .off_in = _IN(rtnh_ifindex), .off_out = _OUT(ifp), .cb = nlf_get_ifpz }, }; #undef _IN #undef _OUT NL_DECLARE_PARSER(mpath_parser, struct rtnexthop, nlf_p_rtnh, nla_p_rtnh); struct rta_mpath { int num_nhops; struct rta_mpath_nh nhops[0]; }; static int nlattr_get_multipath(struct nlattr *nla, struct nl_pstate *npt, const void *arg, void *target) { int data_len = nla->nla_len - sizeof(struct nlattr); struct rtnexthop *rtnh; int max_nhops = data_len / sizeof(struct rtnexthop); struct rta_mpath *mp = npt_alloc(npt, (max_nhops + 2) * sizeof(struct rta_mpath_nh)); mp->num_nhops = 0; for (rtnh = (struct rtnexthop *)(nla + 1); data_len > 0; ) { struct rta_mpath_nh *mpnh = &mp->nhops[mp->num_nhops++]; int error = nl_parse_header(rtnh, rtnh->rtnh_len, &mpath_parser, npt, mpnh); if (error != 0) { NLMSG_REPORT_ERR_MSG(npt, "RTA_MULTIPATH: nexhop %d: parse failed", mp->num_nhops - 1); return (error); } int len = NL_ITEM_ALIGN(rtnh->rtnh_len); data_len -= len; rtnh = (struct rtnexthop *)((char *)rtnh + len); } if (data_len != 0 || mp->num_nhops == 0) { NLMSG_REPORT_ERR_MSG(npt, "invalid RTA_MULTIPATH attr"); return (EINVAL); } *((struct rta_mpath **)target) = mp; return (0); } struct nl_parsed_route { struct sockaddr *rta_dst; struct sockaddr *rta_gw; struct ifnet *rta_oif; struct rta_mpath *rta_multipath; uint32_t rta_table; uint32_t rta_rtflags; uint32_t rta_nh_id; uint32_t rtax_mtu; uint8_t rtm_family; uint8_t rtm_dst_len; }; #define _IN(_field) offsetof(struct rtmsg, _field) #define _OUT(_field) offsetof(struct nl_parsed_route, _field) static struct nlattr_parser nla_p_rtmetrics[] = { { .type = NL_RTAX_MTU, .off = _OUT(rtax_mtu), .cb = nlattr_get_uint32 }, }; NL_DECLARE_ATTR_PARSER(metrics_parser, nla_p_rtmetrics); static const struct nlattr_parser nla_p_rtmsg[] = { { .type = NL_RTA_DST, .off = _OUT(rta_dst), .cb = nlattr_get_ip }, { .type = NL_RTA_OIF, .off = _OUT(rta_oif), .cb = nlattr_get_ifp }, { .type = NL_RTA_GATEWAY, .off = _OUT(rta_gw), .cb = nlattr_get_ip }, { .type = NL_RTA_METRICS, .arg = &metrics_parser, .cb = nlattr_get_nested }, { .type = NL_RTA_MULTIPATH, .off = _OUT(rta_multipath), .cb = nlattr_get_multipath }, { .type = NL_RTA_RTFLAGS, .off = _OUT(rta_rtflags), .cb = nlattr_get_uint32 }, { .type = NL_RTA_TABLE, .off = _OUT(rta_table), .cb = nlattr_get_uint32 }, { .type = NL_RTA_VIA, .off = _OUT(rta_gw), .cb = nlattr_get_ipvia }, { .type = NL_RTA_NH_ID, .off = _OUT(rta_nh_id), .cb = nlattr_get_uint32 }, }; static const struct nlfield_parser nlf_p_rtmsg[] = { {.off_in = _IN(rtm_family), .off_out = _OUT(rtm_family), .cb = nlf_get_u8 }, {.off_in = _IN(rtm_dst_len), .off_out = _OUT(rtm_dst_len), .cb = nlf_get_u8 }, }; #undef _IN #undef _OUT NL_DECLARE_PARSER(rtm_parser, struct rtmsg, nlf_p_rtmsg, nla_p_rtmsg); struct netlink_walkargs { struct nl_writer *nw; struct route_nhop_data rnd; struct nlmsghdr hdr; struct nlpcb *nlp; uint32_t fibnum; int family; int error; int count; int dumped; int dumped_tables; }; static int dump_rtentry(struct rtentry *rt, void *_arg) { struct netlink_walkargs *wa = (struct netlink_walkargs *)_arg; int error; wa->count++; if (wa->error != 0) return (0); wa->dumped++; rt_get_rnd(rt, &wa->rnd); error = dump_px(wa->fibnum, &wa->hdr, rt, &wa->rnd, wa->nw); IF_DEBUG_LEVEL(LOG_DEBUG3) { char rtbuf[INET6_ADDRSTRLEN + 5]; FIB_LOG(LOG_DEBUG3, wa->fibnum, wa->family, "Dump %s, offset %u, error %d", rt_print_buf(rt, rtbuf, sizeof(rtbuf)), wa->nw->offset, error); } wa->error = error; return (0); } static void dump_rtable_one(struct netlink_walkargs *wa, uint32_t fibnum, int family) { FIB_LOG(LOG_DEBUG2, fibnum, family, "Start dump"); wa->count = 0; wa->dumped = 0; rib_walk(fibnum, family, false, dump_rtentry, wa); wa->dumped_tables++; FIB_LOG(LOG_DEBUG2, fibnum, family, "End dump, iterated %d dumped %d", wa->count, wa->dumped); NL_LOG(LOG_DEBUG2, "Current offset: %d", wa->nw->offset); } static int dump_rtable_fib(struct netlink_walkargs *wa, uint32_t fibnum, int family) { wa->fibnum = fibnum; if (family == AF_UNSPEC) { for (int i = 0; i < AF_MAX; i++) { if (rt_tables_get_rnh(fibnum, i) != 0) { wa->family = i; dump_rtable_one(wa, fibnum, i); if (wa->error != 0) break; } } } else { if (rt_tables_get_rnh(fibnum, family) != 0) { wa->family = family; dump_rtable_one(wa, fibnum, family); } } return (wa->error); } static int handle_rtm_getroute(struct nlpcb *nlp, struct nl_parsed_route *attrs, struct nlmsghdr *hdr, struct nl_pstate *npt) { RIB_RLOCK_TRACKER; struct rib_head *rnh; struct rtentry *rt; uint32_t fibnum = attrs->rta_table; sa_family_t family = attrs->rtm_family; if (attrs->rta_dst == NULL) { NLMSG_REPORT_ERR_MSG(npt, "No RTA_DST supplied"); return (EINVAL); } FIB_LOG(LOG_DEBUG, fibnum, family, "getroute called"); rnh = rt_tables_get_rnh(fibnum, family); if (rnh == NULL) return (EAFNOSUPPORT); RIB_RLOCK(rnh); rt = (struct rtentry *)rnh->rnh_matchaddr(attrs->rta_dst, &rnh->head); if (rt == NULL) { RIB_RUNLOCK(rnh); return (ESRCH); } struct route_nhop_data rnd; rt_get_rnd(rt, &rnd); rnd.rnd_nhop = nhop_select_func(rnd.rnd_nhop, 0); RIB_RUNLOCK(rnh); IF_DEBUG_LEVEL(LOG_DEBUG2) { char rtbuf[NHOP_PRINT_BUFSIZE] __unused, nhbuf[NHOP_PRINT_BUFSIZE] __unused; FIB_LOG(LOG_DEBUG2, fibnum, family, "getroute completed: got %s for %s", nhop_print_buf_any(rnd.rnd_nhop, nhbuf, sizeof(nhbuf)), rt_print_buf(rt, rtbuf, sizeof(rtbuf))); } hdr->nlmsg_type = NL_RTM_NEWROUTE; dump_px(fibnum, hdr, rt, &rnd, npt->nw); return (0); } static int handle_rtm_dump(struct nlpcb *nlp, uint32_t fibnum, int family, struct nlmsghdr *hdr, struct nl_writer *nw) { struct netlink_walkargs wa = { .nlp = nlp, .nw = nw, .hdr.nlmsg_pid = hdr->nlmsg_pid, .hdr.nlmsg_seq = hdr->nlmsg_seq, .hdr.nlmsg_type = NL_RTM_NEWROUTE, .hdr.nlmsg_flags = hdr->nlmsg_flags | NLM_F_MULTI, }; if (fibnum == RT_TABLE_UNSPEC) { for (int i = 0; i < V_rt_numfibs; i++) { dump_rtable_fib(&wa, fibnum, family); if (wa.error != 0) break; } } else dump_rtable_fib(&wa, fibnum, family); if (wa.error == 0 && wa.dumped_tables == 0) { FIB_LOG(LOG_DEBUG, fibnum, family, "incorrect fibnum/family"); wa.error = ESRCH; // How do we propagate it? } if (!nlmsg_end_dump(wa.nw, wa.error, &wa.hdr)) { NL_LOG(LOG_DEBUG, "Unable to finalize the dump"); return (ENOMEM); } return (wa.error); } static struct nhop_object * finalize_nhop(struct nhop_object *nh, int *perror) { /* * The following MUST be filled: * nh_ifp, nh_ifa, nh_gw */ if (nh->gw_sa.sa_family == 0) { /* * Empty gateway. Can be direct route with RTA_OIF set. */ if (nh->nh_ifp != NULL) nhop_set_direct_gw(nh, nh->nh_ifp); else { NL_LOG(LOG_DEBUG, "empty gateway and interface, skipping"); *perror = EINVAL; return (NULL); } /* Both nh_ifp and gateway are set */ } else { /* Gateway is set up, we can derive ifp if not set */ if (nh->nh_ifp == NULL) { struct ifaddr *ifa = ifa_ifwithnet(&nh->gw_sa, 1, nhop_get_fibnum(nh)); if (ifa == NULL) { NL_LOG(LOG_DEBUG, "Unable to determine ifp, skipping"); *perror = EINVAL; return (NULL); } nhop_set_transmit_ifp(nh, ifa->ifa_ifp); } } /* Both nh_ifp and gateway are set */ if (nh->nh_ifa == NULL) { struct ifaddr *ifa = ifaof_ifpforaddr(&nh->gw_sa, nh->nh_ifp); if (ifa == NULL) { NL_LOG(LOG_DEBUG, "Unable to determine ifa, skipping"); *perror = EINVAL; return (NULL); } nhop_set_src(nh, ifa); } return (nhop_get_nhop(nh, perror)); } static int get_pxflag(const struct nl_parsed_route *attrs) { int pxflag = 0; switch (attrs->rtm_family) { case AF_INET: if (attrs->rtm_dst_len == 32) pxflag = NHF_HOST; else if (attrs->rtm_dst_len == 0) pxflag = NHF_DEFAULT; break; case AF_INET6: if (attrs->rtm_dst_len == 32) pxflag = NHF_HOST; else if (attrs->rtm_dst_len == 0) pxflag = NHF_DEFAULT; break; } return (pxflag); } static int get_op_flags(int nlm_flags) { int op_flags = 0; op_flags |= (nlm_flags & NLM_F_REPLACE) ? RTM_F_REPLACE : 0; op_flags |= (nlm_flags & NLM_F_EXCL) ? RTM_F_EXCL : 0; op_flags |= (nlm_flags & NLM_F_CREATE) ? RTM_F_CREATE : 0; op_flags |= (nlm_flags & NLM_F_APPEND) ? RTM_F_APPEND : 0; return (op_flags); } #ifdef ROUTE_MPATH static int create_nexthop_one(struct nl_parsed_route *attrs, struct rta_mpath_nh *mpnh, struct nl_pstate *npt, struct nhop_object **pnh) { int error; if (mpnh->gw == NULL) return (EINVAL); struct nhop_object *nh = nhop_alloc(attrs->rta_table, attrs->rtm_family); if (nh == NULL) return (ENOMEM); nhop_set_gw(nh, mpnh->gw, true); if (mpnh->ifp != NULL) nhop_set_transmit_ifp(nh, mpnh->ifp); nhop_set_rtflags(nh, attrs->rta_rtflags); *pnh = finalize_nhop(nh, &error); return (error); } #endif static struct nhop_object * create_nexthop_from_attrs(struct nl_parsed_route *attrs, struct nl_pstate *npt, int *perror) { struct nhop_object *nh = NULL; int error = 0; if (attrs->rta_multipath != NULL) { #ifdef ROUTE_MPATH /* Multipath w/o explicit nexthops */ int num_nhops = attrs->rta_multipath->num_nhops; struct weightened_nhop *wn = npt_alloc(npt, sizeof(*wn) * num_nhops); for (int i = 0; i < num_nhops; i++) { struct rta_mpath_nh *mpnh = &attrs->rta_multipath->nhops[i]; error = create_nexthop_one(attrs, mpnh, npt, &wn[i].nh); if (error != 0) { for (int j = 0; j < i; j++) nhop_free(wn[j].nh); break; } wn[i].weight = mpnh->rtnh_weight > 0 ? mpnh->rtnh_weight : 1; } if (error == 0) { struct rib_head *rh = nhop_get_rh(wn[0].nh); error = nhgrp_get_group(rh, wn, num_nhops, 0, (struct nhgrp_object **)&nh); for (int i = 0; i < num_nhops; i++) nhop_free(wn[i].nh); } #else error = ENOTSUP; #endif *perror = error; } else { nh = nhop_alloc(attrs->rta_table, attrs->rtm_family); if (nh == NULL) { *perror = ENOMEM; return (NULL); } if (attrs->rta_gw != NULL) nhop_set_gw(nh, attrs->rta_gw, true); if (attrs->rta_oif != NULL) nhop_set_transmit_ifp(nh, attrs->rta_oif); if (attrs->rtax_mtu != 0) nhop_set_mtu(nh, attrs->rtax_mtu, true); if (attrs->rta_rtflags & RTF_BROADCAST) nhop_set_broadcast(nh, true); if (attrs->rta_rtflags & RTF_BLACKHOLE) nhop_set_blackhole(nh, NHF_BLACKHOLE); if (attrs->rta_rtflags & RTF_REJECT) nhop_set_blackhole(nh, NHF_REJECT); nhop_set_rtflags(nh, attrs->rta_rtflags); nh = finalize_nhop(nh, perror); } return (nh); } static int rtnl_handle_newroute(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { struct rib_cmd_info rc = {}; struct nhop_object *nh = NULL; int error; struct nl_parsed_route attrs = {}; error = nl_parse_nlmsg(hdr, &rtm_parser, npt, &attrs); if (error != 0) return (error); /* Check if we have enough data */ if (attrs.rta_dst == NULL) { NL_LOG(LOG_DEBUG, "missing RTA_DST"); return (EINVAL); } if (attrs.rta_nh_id != 0) { /* Referenced uindex */ int pxflag = get_pxflag(&attrs); nh = nl_find_nhop(attrs.rta_table, attrs.rtm_family, attrs.rta_nh_id, pxflag, &error); if (error != 0) return (error); } else { nh = create_nexthop_from_attrs(&attrs, npt, &error); if (error != 0) { NL_LOG(LOG_DEBUG, "Error creating nexthop"); return (error); } } int weight = NH_IS_NHGRP(nh) ? 0 : RT_DEFAULT_WEIGHT; struct route_nhop_data rnd = { .rnd_nhop = nh, .rnd_weight = weight }; int op_flags = get_op_flags(hdr->nlmsg_flags); error = rib_add_route_px(attrs.rta_table, attrs.rta_dst, attrs.rtm_dst_len, &rnd, op_flags, &rc); if (error == 0) report_operation(attrs.rta_table, &rc, nlp, hdr); return (error); } static int path_match_func(const struct rtentry *rt, const struct nhop_object *nh, void *_data) { struct nl_parsed_route *attrs = (struct nl_parsed_route *)_data; if ((attrs->rta_gw != NULL) && !rib_match_gw(rt, nh, attrs->rta_gw)) return (0); if ((attrs->rta_oif != NULL) && (attrs->rta_oif != nh->nh_ifp)) return (0); return (1); } static int rtnl_handle_delroute(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { struct rib_cmd_info rc; int error; struct nl_parsed_route attrs = {}; error = nl_parse_nlmsg(hdr, &rtm_parser, npt, &attrs); if (error != 0) return (error); if (attrs.rta_dst == NULL) { NLMSG_REPORT_ERR_MSG(npt, "RTA_DST is not set"); return (ESRCH); } error = rib_del_route_px(attrs.rta_table, attrs.rta_dst, attrs.rtm_dst_len, path_match_func, &attrs, 0, &rc); if (error == 0) report_operation(attrs.rta_table, &rc, nlp, hdr); return (error); } static int rtnl_handle_getroute(struct nlmsghdr *hdr, struct nlpcb *nlp, struct nl_pstate *npt) { int error; struct nl_parsed_route attrs = {}; error = nl_parse_nlmsg(hdr, &rtm_parser, npt, &attrs); if (error != 0) return (error); if (hdr->nlmsg_flags & NLM_F_DUMP) error = handle_rtm_dump(nlp, attrs.rta_table, attrs.rtm_family, hdr, npt->nw); else error = handle_rtm_getroute(nlp, &attrs, hdr, npt); return (error); } void rtnl_handle_route_event(uint32_t fibnum, const struct rib_cmd_info *rc) { int family, nlm_flags = 0; struct nl_writer nw; family = rt_get_family(rc->rc_rt); /* XXX: check if there are active listeners first */ /* TODO: consider passing PID/type/seq */ switch (rc->rc_cmd) { case RTM_ADD: nlm_flags = NLM_F_EXCL | NLM_F_CREATE; break; case RTM_CHANGE: nlm_flags = NLM_F_REPLACE; break; case RTM_DELETE: nlm_flags = 0; break; } IF_DEBUG_LEVEL(LOG_DEBUG2) { char rtbuf[NHOP_PRINT_BUFSIZE] __unused; FIB_LOG(LOG_DEBUG2, fibnum, family, "received event %s for %s / nlm_flags=%X", rib_print_cmd(rc->rc_cmd), rt_print_buf(rc->rc_rt, rtbuf, sizeof(rtbuf)), nlm_flags); } struct nlmsghdr hdr = { .nlmsg_flags = nlm_flags, .nlmsg_type = get_rtmsg_type_from_rtsock(rc->rc_cmd), }; struct route_nhop_data rnd = { .rnd_nhop = rc_get_nhop(rc), .rnd_weight = rc->rc_nh_weight, }; uint32_t group_id = family_to_group(family); if (!nlmsg_get_group_writer(&nw, NLMSG_SMALL, NETLINK_ROUTE, group_id)) { NL_LOG(LOG_DEBUG, "error allocating event buffer"); return; } dump_px(fibnum, &hdr, rc->rc_rt, &rnd, &nw); nlmsg_flush(&nw); } static const struct rtnl_cmd_handler cmd_handlers[] = { { .cmd = NL_RTM_GETROUTE, .name = "RTM_GETROUTE", .cb = &rtnl_handle_getroute, }, { .cmd = NL_RTM_DELROUTE, .name = "RTM_DELROUTE", .cb = &rtnl_handle_delroute, .priv = PRIV_NET_ROUTE, }, { .cmd = NL_RTM_NEWROUTE, .name = "RTM_NEWROUTE", .cb = &rtnl_handle_newroute, .priv = PRIV_NET_ROUTE, } }; static const struct nlhdr_parser *all_parsers[] = {&mpath_parser, &metrics_parser, &rtm_parser}; void -rtnl_routes_init() +rtnl_routes_init(void) { NL_VERIFY_PARSERS(all_parsers); rtnl_register_messages(cmd_handlers, NL_ARRAY_LEN(cmd_handlers)); }