Index: head/sys/net/route/nhop_ctl.c =================================================================== --- head/sys/net/route/nhop_ctl.c (revision 365924) +++ head/sys/net/route/nhop_ctl.c (revision 365925) @@ -1,871 +1,877 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2020 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_route.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * This file contains core functionality for the nexthop ("nhop") route subsystem. * The business logic needed to create nexhop objects is implemented here. * * Nexthops in the original sense are the objects containing all the necessary * information to forward the packet to the selected destination. * In particular, nexthop is defined by a combination of * ifp, ifa, aifp, mtu, gw addr(if set), nh_type, nh_family, mask of rt_flags and * NHF_DEFAULT * * Additionally, each nexthop gets assigned its unique index (nexthop index). * It serves two purposes: first one is to ease the ability of userland programs to * reference nexthops by their index. The second one allows lookup algorithms to * to store index instead of pointer (2 bytes vs 8) as a lookup result. * All nexthops are stored in the resizable hash table. * * Basically, this file revolves around supporting 3 functions: * 1) nhop_create_from_info / nhop_create_from_nhop, which contains all * business logic on filling the nexthop fields based on the provided request. * 2) nhop_get(), which gets a usable referenced nexthops. * * Conventions: * 1) non-exported functions start with verb * 2) exported function starts with the subsystem prefix: "nhop" */ static int dump_nhop_entry(struct rib_head *rh, struct nhop_object *nh, struct sysctl_req *w); static struct nhop_priv *alloc_nhop_structure(void); static int get_nhop(struct rib_head *rnh, struct rt_addrinfo *info, struct nhop_priv **pnh_priv); static int finalize_nhop(struct nh_control *ctl, struct rt_addrinfo *info, struct nhop_priv *nh_priv); static struct ifnet *get_aifp(const struct nhop_object *nh, int reference); static void fill_sdl_from_ifp(struct sockaddr_dl_short *sdl, const struct ifnet *ifp); static void destroy_nhop_epoch(epoch_context_t ctx); static void destroy_nhop(struct nhop_priv *nh_priv); static void print_nhop(const char *prefix, const struct nhop_object *nh); _Static_assert(__offsetof(struct nhop_object, nh_ifp) == 32, "nhop_object: wrong nh_ifp offset"); _Static_assert(sizeof(struct nhop_object) <= 128, "nhop_object: size exceeds 128 bytes"); static uma_zone_t nhops_zone; /* Global zone for each and every nexthop */ #define NHOP_OBJECT_ALIGNED_SIZE roundup2(sizeof(struct nhop_object), \ 2 * CACHE_LINE_SIZE) #define NHOP_PRIV_ALIGNED_SIZE roundup2(sizeof(struct nhop_priv), \ 2 * CACHE_LINE_SIZE) void nhops_init(void) { nhops_zone = uma_zcreate("routing nhops", NHOP_OBJECT_ALIGNED_SIZE + NHOP_PRIV_ALIGNED_SIZE, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } /* * Fetches the interface of source address used by the route. * In all cases except interface-address-route it would be the * same as the transmit interfaces. * However, for the interface address this function will return * this interface ifp instead of loopback. This is needed to support * link-local IPv6 loopback communications. * * If @reference is non-zero, found ifp is referenced. * * Returns found ifp. */ static struct ifnet * get_aifp(const struct nhop_object *nh, int reference) { struct ifnet *aifp = NULL; /* * Adjust the "outgoing" interface. If we're going to loop * the packet back to ourselves, the ifp would be the loopback * interface. However, we'd rather know the interface associated * to the destination address (which should probably be one of * our own addresses). */ if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) && nh->gw_sa.sa_family == AF_LINK) { if (reference) aifp = ifnet_byindex_ref(nh->gwl_sa.sdl_index); else aifp = ifnet_byindex(nh->gwl_sa.sdl_index); if (aifp == NULL) { DPRINTF("unable to get aifp for %s index %d", if_name(nh->nh_ifp), nh->gwl_sa.sdl_index); } } if (aifp == NULL) { aifp = nh->nh_ifp; if (reference) if_ref(aifp); } return (aifp); } int cmp_priv(const struct nhop_priv *_one, const struct nhop_priv *_two) { if (memcmp(_one->nh, _two->nh, NHOP_END_CMP) != 0) return (0); if ((_one->nh_type != _two->nh_type) || (_one->nh_family != _two->nh_family)) return (0); return (1); } /* * Conditionally sets @nh mtu data based on the @info data. */ static void set_nhop_mtu_from_info(struct nhop_object *nh, const struct rt_addrinfo *info) { if (info->rti_mflags & RTV_MTU) { if (info->rti_rmx->rmx_mtu != 0) { /* * MTU was explicitly provided by user. * Keep it. */ nh->nh_priv->rt_flags |= RTF_FIXEDMTU; } else { /* * User explicitly sets MTU to 0. * Assume rollback to default. */ nh->nh_priv->rt_flags &= ~RTF_FIXEDMTU; } nh->nh_mtu = info->rti_rmx->rmx_mtu; } } /* * Fills in shorted link-level sockadd version suitable to be stored inside the * nexthop gateway buffer. */ static void fill_sdl_from_ifp(struct sockaddr_dl_short *sdl, const struct ifnet *ifp) { + bzero(sdl, sizeof(struct sockaddr_dl_short)); sdl->sdl_family = AF_LINK; sdl->sdl_len = sizeof(struct sockaddr_dl_short); sdl->sdl_index = ifp->if_index; sdl->sdl_type = ifp->if_type; } static int set_nhop_gw_from_info(struct nhop_object *nh, struct rt_addrinfo *info) { struct sockaddr *gw; gw = info->rti_info[RTAX_GATEWAY]; + KASSERT(gw != NULL, ("gw is NULL")); + if (info->rti_flags & RTF_GATEWAY) { if (gw->sa_len > sizeof(struct sockaddr_in6)) { DPRINTF("nhop SA size too big: AF %d len %u", gw->sa_family, gw->sa_len); return (ENOMEM); } memcpy(&nh->gw_sa, gw, gw->sa_len); } else { /* * Interface route. Currently the route.c code adds * sa of type AF_LINK, which is 56 bytes long. The only * meaningful data there is the interface index. It is used * used is the IPv6 loopback output, where we need to preserve * the original interface to maintain proper scoping. * Despite the fact that nexthop code stores original interface * in the separate field (nh_aifp, see below), write AF_LINK * compatible sa with shorter total length. */ struct sockaddr_dl *sdl; struct ifnet *ifp; /* Fetch and validate interface index */ sdl = (struct sockaddr_dl *)gw; if (sdl->sdl_family != AF_LINK) { DPRINTF("unsupported AF: %d", sdl->sdl_family); return (ENOTSUP); } ifp = ifnet_byindex(sdl->sdl_index); if (ifp == NULL) { DPRINTF("invalid ifindex %d", sdl->sdl_index); return (EINVAL); } fill_sdl_from_ifp(&nh->gwl_sa, ifp); } return (0); } static uint16_t convert_rt_to_nh_flags(int rt_flags) { uint16_t res; res = (rt_flags & RTF_REJECT) ? NHF_REJECT : 0; res |= (rt_flags & RTF_HOST) ? NHF_HOST : 0; res |= (rt_flags & RTF_BLACKHOLE) ? NHF_BLACKHOLE : 0; res |= (rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) ? NHF_REDIRECT : 0; res |= (rt_flags & RTF_BROADCAST) ? NHF_BROADCAST : 0; res |= (rt_flags & RTF_GATEWAY) ? NHF_GATEWAY : 0; return (res); } static int fill_nhop_from_info(struct nhop_priv *nh_priv, struct rt_addrinfo *info) { int error, rt_flags; struct nhop_object *nh; nh = nh_priv->nh; rt_flags = info->rti_flags & NHOP_RT_FLAG_MASK; nh->nh_priv->rt_flags = rt_flags; nh_priv->nh_family = info->rti_info[RTAX_DST]->sa_family; nh_priv->nh_type = 0; // hook responsibility to set nhop type nh->nh_flags = convert_rt_to_nh_flags(rt_flags); set_nhop_mtu_from_info(nh, info); if ((error = set_nhop_gw_from_info(nh, info)) != 0) return (error); nh->nh_ifp = info->rti_ifa->ifa_ifp; nh->nh_ifa = info->rti_ifa; /* depends on the gateway */ nh->nh_aifp = get_aifp(nh, 0); /* * Note some of the remaining data is set by the * per-address-family pre-add hook. */ return (0); } /* * Creates a new nexthop based on the information in @info. * * Returns: * 0 on success, filling @nh_ret with the desired nexthop object ptr * errno otherwise */ int nhop_create_from_info(struct rib_head *rnh, struct rt_addrinfo *info, struct nhop_object **nh_ret) { struct nhop_priv *nh_priv; int error; NET_EPOCH_ASSERT(); + + if (info->rti_info[RTAX_GATEWAY] == NULL) + return (EINVAL); nh_priv = alloc_nhop_structure(); error = fill_nhop_from_info(nh_priv, info); if (error != 0) { uma_zfree(nhops_zone, nh_priv->nh); return (error); } error = get_nhop(rnh, info, &nh_priv); if (error == 0) *nh_ret = nh_priv->nh; return (error); } /* * Gets linked nhop using the provided @pnh_priv nexhop data. * If linked nhop is found, returns it, freeing the provided one. * If there is no such nexthop, attaches the remaining data to the * provided nexthop and links it. * * Returns 0 on success, storing referenced nexthop in @pnh_priv. * Otherwise, errno is returned. */ static int get_nhop(struct rib_head *rnh, struct rt_addrinfo *info, struct nhop_priv **pnh_priv) { const struct sockaddr *dst, *gateway, *netmask; struct nhop_priv *nh_priv, *tmp_priv; int error; nh_priv = *pnh_priv; /* Give the protocols chance to augment the request data */ dst = info->rti_info[RTAX_DST]; netmask = info->rti_info[RTAX_NETMASK]; gateway = info->rti_info[RTAX_GATEWAY]; error = rnh->rnh_preadd(rnh->rib_fibnum, dst, netmask, nh_priv->nh); if (error != 0) { uma_zfree(nhops_zone, nh_priv->nh); return (error); } tmp_priv = find_nhop(rnh->nh_control, nh_priv); if (tmp_priv != NULL) { uma_zfree(nhops_zone, nh_priv->nh); *pnh_priv = tmp_priv; return (0); } /* * Existing nexthop not found, need to create new one. * Note: multiple simultaneous get_nhop() requests * can result in multiple equal nexhops existing in the * nexthop table. This is not a not a problem until the * relative number of such nexthops is significant, which * is extremely unlikely. */ error = finalize_nhop(rnh->nh_control, info, nh_priv); if (error != 0) return (error); return (0); } /* * Update @nh with data supplied in @info. * This is a helper function to support route changes. * * It limits the changes that can be done to the route to the following: * 1) all combination of gateway changes (gw, interface, blackhole/reject) * 2) route flags (FLAG[123],STATIC,BLACKHOLE,REJECT) * 3) route MTU * * Returns: * 0 on success */ static int alter_nhop_from_info(struct nhop_object *nh, struct rt_addrinfo *info) { struct sockaddr *info_gw; int error; /* Update MTU if set in the request*/ set_nhop_mtu_from_info(nh, info); /* XXX: allow only one of BLACKHOLE,REJECT,GATEWAY */ /* Allow some flags (FLAG1,STATIC,BLACKHOLE,REJECT) to be toggled on change. */ nh->nh_priv->rt_flags &= ~RTF_FMASK; nh->nh_priv->rt_flags |= info->rti_flags & RTF_FMASK; /* Consider gateway change */ info_gw = info->rti_info[RTAX_GATEWAY]; if (info_gw != NULL) { error = set_nhop_gw_from_info(nh, info); if (error != 0) return (error); /* Update RTF_GATEWAY flag status */ nh->nh_priv->rt_flags &= ~RTF_GATEWAY; nh->nh_priv->rt_flags |= (RTF_GATEWAY & info->rti_flags); } /* Update datapath flags */ nh->nh_flags = convert_rt_to_nh_flags(nh->nh_priv->rt_flags); if (info->rti_ifa != NULL) nh->nh_ifa = info->rti_ifa; if (info->rti_ifp != NULL) nh->nh_ifp = info->rti_ifp; nh->nh_aifp = get_aifp(nh, 0); return (0); } /* * Creates new nexthop based on @nh_orig and augmentation data from @info. * Helper function used in the route changes, please see * alter_nhop_from_info() comments for more details. * * Returns: * 0 on success, filling @nh_ret with the desired nexthop object * errno otherwise */ int nhop_create_from_nhop(struct rib_head *rnh, const struct nhop_object *nh_orig, struct rt_addrinfo *info, struct nhop_object **pnh) { struct nhop_priv *nh_priv; struct nhop_object *nh; int error; NET_EPOCH_ASSERT(); nh_priv = alloc_nhop_structure(); nh = nh_priv->nh; /* Start with copying data from original nexthop */ nh_priv->nh_family = nh_orig->nh_priv->nh_family; nh_priv->rt_flags = nh_orig->nh_priv->rt_flags; nh_priv->nh_type = nh_orig->nh_priv->nh_type; nh->nh_ifp = nh_orig->nh_ifp; nh->nh_ifa = nh_orig->nh_ifa; nh->nh_aifp = nh_orig->nh_aifp; nh->nh_mtu = nh_orig->nh_mtu; nh->nh_flags = nh_orig->nh_flags; memcpy(&nh->gw_sa, &nh_orig->gw_sa, nh_orig->gw_sa.sa_len); error = alter_nhop_from_info(nh, info); if (error != 0) { uma_zfree(nhops_zone, nh_priv->nh); return (error); } error = get_nhop(rnh, info, &nh_priv); if (error == 0) *pnh = nh_priv->nh; return (error); } /* * Allocates memory for public/private nexthop structures. * * Returns pointer to nhop_priv or NULL. */ static struct nhop_priv * alloc_nhop_structure() { struct nhop_object *nh; struct nhop_priv *nh_priv; nh = (struct nhop_object *)uma_zalloc(nhops_zone, M_NOWAIT | M_ZERO); if (nh == NULL) return (NULL); nh_priv = (struct nhop_priv *)((char *)nh + NHOP_OBJECT_ALIGNED_SIZE); nh->nh_priv = nh_priv; nh_priv->nh = nh; return (nh_priv); } /* * Alocates/references the remaining bits of nexthop data and links * it to the hash table. * Returns 0 if successful, * errno otherwise. @nh_priv is freed in case of error. */ static int finalize_nhop(struct nh_control *ctl, struct rt_addrinfo *info, struct nhop_priv *nh_priv) { struct nhop_object *nh; nh = nh_priv->nh; /* Allocate per-cpu packet counter */ nh->nh_pksent = counter_u64_alloc(M_NOWAIT); if (nh->nh_pksent == NULL) { uma_zfree(nhops_zone, nh); RTSTAT_INC(rts_nh_alloc_failure); DPRINTF("nh_alloc_finalize failed"); return (ENOMEM); } /* Save vnet to ease destruction */ nh_priv->nh_vnet = curvnet; /* Reference external objects and calculate (referenced) ifa */ if_ref(nh->nh_ifp); ifa_ref(nh->nh_ifa); nh->nh_aifp = get_aifp(nh, 1); DPRINTF("AIFP: %p nh_ifp %p", nh->nh_aifp, nh->nh_ifp); refcount_init(&nh_priv->nh_refcnt, 1); /* Please see nhop_free() comments on the initial value */ refcount_init(&nh_priv->nh_linked, 2); print_nhop("FINALIZE", nh); if (link_nhop(ctl, nh_priv) == 0) { /* * Adding nexthop to the datastructures * failed. Call destructor w/o waiting for * the epoch end, as nexthop is not used * and return. */ DPRINTF("link_nhop failed!"); destroy_nhop(nh_priv); return (ENOBUFS); } return (0); } static void print_nhop_sa(char *buf, size_t buflen, const struct sockaddr *sa) { if (sa->sa_family == AF_INET) { const struct sockaddr_in *sin4; sin4 = (const struct sockaddr_in *)sa; inet_ntop(AF_INET, &sin4->sin_addr, buf, buflen); } else if (sa->sa_family == AF_INET6) { const struct sockaddr_in6 *sin6; sin6 = (const struct sockaddr_in6 *)sa; inet_ntop(AF_INET6, &sin6->sin6_addr, buf, buflen); } else if (sa->sa_family == AF_LINK) { const struct sockaddr_dl *sdl; sdl = (const struct sockaddr_dl *)sa; snprintf(buf, buflen, "if#%d", sdl->sdl_index); } else snprintf(buf, buflen, "af:%d", sa->sa_family); } static void print_nhop(const char *prefix, const struct nhop_object *nh) { char src_buf[INET6_ADDRSTRLEN], addr_buf[INET6_ADDRSTRLEN]; print_nhop_sa(src_buf, sizeof(src_buf), nh->nh_ifa->ifa_addr); print_nhop_sa(addr_buf, sizeof(addr_buf), &nh->gw_sa); DPRINTF("%s nhop priv %p: AF %d ifp %p %s addr %s src %p %s aifp %p %s mtu %d nh_flags %X", prefix, nh->nh_priv, nh->nh_priv->nh_family, nh->nh_ifp, if_name(nh->nh_ifp), addr_buf, nh->nh_ifa, src_buf, nh->nh_aifp, if_name(nh->nh_aifp), nh->nh_mtu, nh->nh_flags); } static void destroy_nhop(struct nhop_priv *nh_priv) { struct nhop_object *nh; nh = nh_priv->nh; print_nhop("DEL", nh); if_rele(nh->nh_ifp); if_rele(nh->nh_aifp); ifa_free(nh->nh_ifa); counter_u64_free(nh->nh_pksent); uma_zfree(nhops_zone, nh); } /* * Epoch callback indicating nhop is safe to destroy */ static void destroy_nhop_epoch(epoch_context_t ctx) { struct nhop_priv *nh_priv; nh_priv = __containerof(ctx, struct nhop_priv, nh_epoch_ctx); destroy_nhop(nh_priv); } void nhop_ref_object(struct nhop_object *nh) { u_int old; old = refcount_acquire(&nh->nh_priv->nh_refcnt); KASSERT(old > 0, ("%s: nhop object %p has 0 refs", __func__, nh)); } int nhop_try_ref_object(struct nhop_object *nh) { return (refcount_acquire_if_not_zero(&nh->nh_priv->nh_refcnt)); } void nhop_free(struct nhop_object *nh) { struct nh_control *ctl; struct nhop_priv *nh_priv = nh->nh_priv; struct epoch_tracker et; if (!refcount_release(&nh_priv->nh_refcnt)) return; /* * There are only 2 places, where nh_linked can be decreased: * rib destroy (nhops_destroy_rib) and this function. * nh_link can never be increased. * * Hence, use initial value of 2 to make use of * refcount_release_if_not_last(). * * There can be two scenarious when calling this function: * * 1) nh_linked value is 2. This means that either * nhops_destroy_rib() has not been called OR it is running, * but we are guaranteed that nh_control won't be freed in * this epoch. Hence, nexthop can be safely unlinked. * * 2) nh_linked value is 1. In that case, nhops_destroy_rib() * has been called and nhop unlink can be skipped. */ NET_EPOCH_ENTER(et); if (refcount_release_if_not_last(&nh_priv->nh_linked)) { ctl = nh_priv->nh_control; if (unlink_nhop(ctl, nh_priv) == NULL) { /* Do not try to reclaim */ DPRINTF("Failed to unlink nexhop %p", nh_priv); NET_EPOCH_EXIT(et); return; } } NET_EPOCH_EXIT(et); epoch_call(net_epoch_preempt, destroy_nhop_epoch, &nh_priv->nh_epoch_ctx); } int nhop_ref_any(struct nhop_object *nh) { return (nhop_try_ref_object(nh)); } void nhop_free_any(struct nhop_object *nh) { nhop_free(nh); } /* Helper functions */ uint32_t nhop_get_idx(const struct nhop_object *nh) { return (nh->nh_priv->nh_idx); } enum nhop_type nhop_get_type(const struct nhop_object *nh) { return (nh->nh_priv->nh_type); } void nhop_set_type(struct nhop_object *nh, enum nhop_type nh_type) { nh->nh_priv->nh_type = nh_type; } int nhop_get_rtflags(const struct nhop_object *nh) { return (nh->nh_priv->rt_flags); } void nhop_set_rtflags(struct nhop_object *nh, int rt_flags) { nh->nh_priv->rt_flags = rt_flags; } struct vnet * nhop_get_vnet(const struct nhop_object *nh) { return (nh->nh_priv->nh_vnet); } void nhops_update_ifmtu(struct rib_head *rh, struct ifnet *ifp, uint32_t mtu) { struct nh_control *ctl; struct nhop_priv *nh_priv; struct nhop_object *nh; ctl = rh->nh_control; NHOPS_WLOCK(ctl); CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) { nh = nh_priv->nh; if (nh->nh_ifp == ifp) { if ((nh_priv->rt_flags & RTF_FIXEDMTU) == 0 || nh->nh_mtu > mtu) { /* Update MTU directly */ nh->nh_mtu = mtu; } } } CHT_SLIST_FOREACH_END; NHOPS_WUNLOCK(ctl); } /* * Dumps a single entry to sysctl buffer. * * Layout: * rt_msghdr - generic RTM header to allow users to skip non-understood messages * nhop_external - nexhop description structure (with length) * nhop_addrs - structure encapsulating GW/SRC sockaddrs */ static int dump_nhop_entry(struct rib_head *rh, struct nhop_object *nh, struct sysctl_req *w) { struct { struct rt_msghdr rtm; struct nhop_external nhe; struct nhop_addrs na; } arpc; struct nhop_external *pnhe; struct sockaddr *gw_sa, *src_sa; struct sockaddr_storage ss; size_t addrs_len; int error; //DPRINTF("Dumping: head %p nh %p flags %X req %p\n", rh, nh, nh->nh_flags, w); memset(&arpc, 0, sizeof(arpc)); arpc.rtm.rtm_msglen = sizeof(arpc); arpc.rtm.rtm_version = RTM_VERSION; arpc.rtm.rtm_type = RTM_GET; //arpc.rtm.rtm_flags = RTF_UP; arpc.rtm.rtm_flags = nh->nh_priv->rt_flags; /* nhop_external */ pnhe = &arpc.nhe; pnhe->nh_len = sizeof(struct nhop_external); pnhe->nh_idx = nh->nh_priv->nh_idx; pnhe->nh_fib = rh->rib_fibnum; pnhe->ifindex = nh->nh_ifp->if_index; pnhe->aifindex = nh->nh_aifp->if_index; pnhe->nh_family = nh->nh_priv->nh_family; pnhe->nh_type = nh->nh_priv->nh_type; pnhe->nh_mtu = nh->nh_mtu; pnhe->nh_flags = nh->nh_flags; memcpy(pnhe->nh_prepend, nh->nh_prepend, sizeof(nh->nh_prepend)); pnhe->prepend_len = nh->nh_prepend_len; pnhe->nh_refcount = nh->nh_priv->nh_refcnt; pnhe->nh_pksent = counter_u64_fetch(nh->nh_pksent); /* sockaddr container */ addrs_len = sizeof(struct nhop_addrs); arpc.na.gw_sa_off = addrs_len; gw_sa = (struct sockaddr *)&nh->gw4_sa; addrs_len += gw_sa->sa_len; src_sa = nh->nh_ifa->ifa_addr; if (src_sa->sa_family == AF_LINK) { /* Shorten structure */ memset(&ss, 0, sizeof(struct sockaddr_storage)); fill_sdl_from_ifp((struct sockaddr_dl_short *)&ss, nh->nh_ifa->ifa_ifp); src_sa = (struct sockaddr *)&ss; } arpc.na.src_sa_off = addrs_len; addrs_len += src_sa->sa_len; /* Write total container length */ arpc.na.na_len = addrs_len; arpc.rtm.rtm_msglen += arpc.na.na_len - sizeof(struct nhop_addrs); error = SYSCTL_OUT(w, &arpc, sizeof(arpc)); if (error == 0) error = SYSCTL_OUT(w, gw_sa, gw_sa->sa_len); if (error == 0) error = SYSCTL_OUT(w, src_sa, src_sa->sa_len); return (error); } int nhops_dump_sysctl(struct rib_head *rh, struct sysctl_req *w) { struct nh_control *ctl; struct nhop_priv *nh_priv; int error; ctl = rh->nh_control; NHOPS_RLOCK(ctl); DPRINTF("NHDUMP: count=%u", ctl->nh_head.items_count); CHT_SLIST_FOREACH(&ctl->nh_head, nhops, nh_priv) { error = dump_nhop_entry(rh, nh_priv->nh, w); if (error != 0) { NHOPS_RUNLOCK(ctl); return (error); } } CHT_SLIST_FOREACH_END; NHOPS_RUNLOCK(ctl); return (0); } Index: head/sys/net/route/route_ctl.c =================================================================== --- head/sys/net/route/route_ctl.c (revision 365924) +++ head/sys/net/route/route_ctl.c (revision 365925) @@ -1,1187 +1,1187 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2020 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_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RADIX_MPATH #include #endif #include /* * This file contains control plane routing tables functions. * * All functions assumes they are called in net epoch. */ struct rib_subscription { CK_STAILQ_ENTRY(rib_subscription) next; rib_subscription_cb_t *func; void *arg; enum rib_subscription_type type; struct epoch_context epoch_ctx; }; static int add_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc); static int add_route_nhop(struct rib_head *rnh, struct rtentry *rt, struct rt_addrinfo *info, struct route_nhop_data *rnd, struct rib_cmd_info *rc); static int del_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc); static int change_route(struct rib_head *rnh, struct rt_addrinfo *info, struct route_nhop_data *nhd_orig, struct rib_cmd_info *rc); static int change_route_nhop(struct rib_head *rnh, struct rtentry *rt, struct rt_addrinfo *info, struct route_nhop_data *rnd, struct rib_cmd_info *rc); static int rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc); static void rib_notify(struct rib_head *rnh, enum rib_subscription_type type, struct rib_cmd_info *rc); static void destroy_subscription_epoch(epoch_context_t ctx); /* Routing table UMA zone */ VNET_DEFINE_STATIC(uma_zone_t, rtzone); #define V_rtzone VNET(rtzone) void vnet_rtzone_init() { V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } #ifdef VIMAGE void vnet_rtzone_destroy() { uma_zdestroy(V_rtzone); } #endif static void destroy_rtentry(struct rtentry *rt) { /* * At this moment rnh, nh_control may be already freed. * nhop interface may have been migrated to a different vnet. * Use vnet stored in the nexthop to delete the entry. */ CURVNET_SET(nhop_get_vnet(rt->rt_nhop)); /* Unreference nexthop */ nhop_free(rt->rt_nhop); uma_zfree(V_rtzone, rt); CURVNET_RESTORE(); } /* * Epoch callback indicating rtentry is safe to destroy */ static void destroy_rtentry_epoch(epoch_context_t ctx) { struct rtentry *rt; rt = __containerof(ctx, struct rtentry, rt_epoch_ctx); destroy_rtentry(rt); } /* * Schedule rtentry deletion */ static void rtfree(struct rtentry *rt) { KASSERT(rt != NULL, ("%s: NULL rt", __func__)); epoch_call(net_epoch_preempt, destroy_rtentry_epoch, &rt->rt_epoch_ctx); } static struct rib_head * get_rnh(uint32_t fibnum, const struct rt_addrinfo *info) { struct rib_head *rnh; struct sockaddr *dst; KASSERT((fibnum < rt_numfibs), ("rib_add_route: bad fibnum")); dst = info->rti_info[RTAX_DST]; rnh = rt_tables_get_rnh(fibnum, dst->sa_family); return (rnh); } /* * Check if specified @gw matches gw data in the nexthop @nh. * * Returns true if matches, false otherwise. */ static bool match_nhop_gw(const struct nhop_object *nh, const struct sockaddr *gw) { if (nh->gw_sa.sa_family != gw->sa_family) return (false); switch (gw->sa_family) { case AF_INET: return (nh->gw4_sa.sin_addr.s_addr == ((const struct sockaddr_in *)gw)->sin_addr.s_addr); case AF_INET6: { const struct sockaddr_in6 *gw6; gw6 = (const struct sockaddr_in6 *)gw; /* * Currently (2020-09) IPv6 gws in kernel have their * scope embedded. Once this becomes false, this code * has to be revisited. */ if (IN6_ARE_ADDR_EQUAL(&nh->gw6_sa.sin6_addr, &gw6->sin6_addr)) return (true); return (false); } case AF_LINK: { const struct sockaddr_dl *sdl; sdl = (const struct sockaddr_dl *)gw; return (nh->gwl_sa.sdl_index == sdl->sdl_index); } default: return (memcmp(&nh->gw_sa, gw, nh->gw_sa.sa_len) == 0); } /* NOTREACHED */ return (false); } /* * Checks if data in @info matches nexhop @nh. * * Returns 0 on success, * ESRCH if not matched, * ENOENT if filter function returned false */ int check_info_match_nhop(const struct rt_addrinfo *info, const struct rtentry *rt, const struct nhop_object *nh) { const struct sockaddr *gw = info->rti_info[RTAX_GATEWAY]; if (info->rti_filter != NULL) { if (info->rti_filter(rt, nh, info->rti_filterdata) == 0) return (ENOENT); else return (0); } if ((gw != NULL) && !match_nhop_gw(nh, gw)) return (ESRCH); return (0); } /* * Checks if nexhop @nh can be rewritten by data in @info because * of higher "priority". Currently the only case for such scenario * is kernel installing interface routes, marked by RTF_PINNED flag. * * Returns: * 1 if @info data has higher priority * 0 if priority is the same * -1 if priority is lower */ int can_override_nhop(const struct rt_addrinfo *info, const struct nhop_object *nh) { if (info->rti_flags & RTF_PINNED) { return (NH_IS_PINNED(nh)) ? 0 : 1; } else { return (NH_IS_PINNED(nh)) ? -1 : 0; } } /* * Runs exact prefix match based on @dst and @netmask. * Returns matched @rtentry if found or NULL. * If rtentry was found, saves nexthop / weight value into @rnd. */ static struct rtentry * lookup_prefix_bysa(struct rib_head *rnh, const struct sockaddr *dst, const struct sockaddr *netmask, struct route_nhop_data *rnd) { struct rtentry *rt; RIB_LOCK_ASSERT(rnh); rt = (struct rtentry *)rnh->rnh_lookup(__DECONST(void *, dst), __DECONST(void *, netmask), &rnh->head); if (rt != NULL) { rnd->rnd_nhop = rt->rt_nhop; rnd->rnd_weight = rt->rt_weight; } else { rnd->rnd_nhop = NULL; rnd->rnd_weight = 0; } return (rt); } /* * Runs exact prefix match based on dst/netmask from @info. * Assumes RIB lock is held. * Returns matched @rtentry if found or NULL. * If rtentry was found, saves nexthop / weight value into @rnd. */ struct rtentry * lookup_prefix(struct rib_head *rnh, const struct rt_addrinfo *info, struct route_nhop_data *rnd) { struct rtentry *rt; rt = lookup_prefix_bysa(rnh, info->rti_info[RTAX_DST], info->rti_info[RTAX_NETMASK], rnd); return (rt); } /* * Adds route defined by @info into the kernel table specified by @fibnum and * sa_family in @info->rti_info[RTAX_DST]. * * Returns 0 on success and fills in operation metadata into @rc. */ int rib_add_route(uint32_t fibnum, struct rt_addrinfo *info, struct rib_cmd_info *rc) { struct rib_head *rnh; int error; NET_EPOCH_ASSERT(); rnh = get_rnh(fibnum, info); if (rnh == NULL) return (EAFNOSUPPORT); /* * Check consistency between RTF_HOST flag and netmask * existence. */ if (info->rti_flags & RTF_HOST) info->rti_info[RTAX_NETMASK] = NULL; else if (info->rti_info[RTAX_NETMASK] == NULL) return (EINVAL); bzero(rc, sizeof(struct rib_cmd_info)); rc->rc_cmd = RTM_ADD; error = add_route(rnh, info, rc); if (error == 0) rib_notify(rnh, RIB_NOTIFY_DELAYED, rc); return (error); } /* * Creates rtentry and nexthop based on @info data. * Return 0 and fills in rtentry into @prt on success, * return errno otherwise. */ static int create_rtentry(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **prt) { struct sockaddr *dst, *ndst, *gateway, *netmask; struct rtentry *rt; struct nhop_object *nh; struct ifaddr *ifa; int error, flags; dst = info->rti_info[RTAX_DST]; gateway = info->rti_info[RTAX_GATEWAY]; netmask = info->rti_info[RTAX_NETMASK]; flags = info->rti_flags; if ((flags & RTF_GATEWAY) && !gateway) return (EINVAL); if (dst && gateway && (dst->sa_family != gateway->sa_family) && (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK)) return (EINVAL); if (dst->sa_len > sizeof(((struct rtentry *)NULL)->rt_dstb)) return (EINVAL); if (info->rti_ifa == NULL) { error = rt_getifa_fib(info, rnh->rib_fibnum); if (error) return (error); } else { ifa_ref(info->rti_ifa); } error = nhop_create_from_info(rnh, info, &nh); if (error != 0) { ifa_free(info->rti_ifa); return (error); } rt = uma_zalloc(V_rtzone, M_NOWAIT | M_ZERO); if (rt == NULL) { ifa_free(info->rti_ifa); nhop_free(nh); return (ENOBUFS); } - rt->rte_flags = RTF_UP | flags; + rt->rte_flags = (RTF_UP | flags) & RTE_RT_FLAG_MASK; rt->rt_nhop = nh; /* Fill in dst */ memcpy(&rt->rt_dst, dst, dst->sa_len); rt_key(rt) = &rt->rt_dst; /* * point to the (possibly newly malloc'd) dest address. */ ndst = (struct sockaddr *)rt_key(rt); /* * make sure it contains the value we want (masked if needed). */ if (netmask) { rt_maskedcopy(dst, ndst, netmask); } else bcopy(dst, ndst, dst->sa_len); /* * We use the ifa reference returned by rt_getifa_fib(). * This moved from below so that rnh->rnh_addaddr() can * examine the ifa and ifa->ifa_ifp if it so desires. */ ifa = info->rti_ifa; rt->rt_weight = 1; rt_setmetrics(info, rt); *prt = rt; return (0); } static int add_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc) { struct nhop_object *nh_orig; struct route_nhop_data rnd; struct nhop_object *nh; struct rtentry *rt, *rt_orig; int error; error = create_rtentry(rnh, info, &rt); if (error != 0) return (error); rnd.rnd_nhop = rt->rt_nhop; rnd.rnd_weight = rt->rt_weight; nh = rt->rt_nhop; RIB_WLOCK(rnh); #ifdef RADIX_MPATH struct sockaddr *netmask; netmask = info->rti_info[RTAX_NETMASK]; /* do not permit exactly the same dst/mask/gw pair */ if (rt_mpath_capable(rnh) && rt_mpath_conflict(rnh, rt, netmask)) { RIB_WUNLOCK(rnh); nhop_free(nh); uma_zfree(V_rtzone, rt); return (EEXIST); } #endif error = add_route_nhop(rnh, rt, info, &rnd, rc); if (error == 0) { RIB_WUNLOCK(rnh); return (0); } /* addition failed. Lookup prefix in the rib to determine the cause */ rt_orig = lookup_prefix(rnh, info, &rnd); if (rt_orig == NULL) { /* No prefix -> rnh_addaddr() failed to allocate memory */ RIB_WUNLOCK(rnh); nhop_free(nh); uma_zfree(V_rtzone, rt); return (ENOMEM); } /* We have existing route in the RIB. */ nh_orig = rnd.rnd_nhop; /* Check if new route has higher preference */ if (can_override_nhop(info, nh_orig) > 0) { /* Update nexthop to the new route */ change_route_nhop(rnh, rt_orig, info, &rnd, rc); RIB_WUNLOCK(rnh); uma_zfree(V_rtzone, rt); nhop_free(nh_orig); return (0); } RIB_WUNLOCK(rnh); /* Unable to add - another route with the same preference exists */ error = EEXIST; nhop_free(nh); uma_zfree(V_rtzone, rt); return (error); } /* * Removes route defined by @info from the kernel table specified by @fibnum and * sa_family in @info->rti_info[RTAX_DST]. * * Returns 0 on success and fills in operation metadata into @rc. */ int rib_del_route(uint32_t fibnum, struct rt_addrinfo *info, struct rib_cmd_info *rc) { struct rib_head *rnh; struct sockaddr *dst_orig, *netmask; struct sockaddr_storage mdst; int error; NET_EPOCH_ASSERT(); rnh = get_rnh(fibnum, info); if (rnh == NULL) return (EAFNOSUPPORT); bzero(rc, sizeof(struct rib_cmd_info)); rc->rc_cmd = RTM_DELETE; dst_orig = info->rti_info[RTAX_DST]; netmask = info->rti_info[RTAX_NETMASK]; if (netmask != NULL) { /* Ensure @dst is always properly masked */ if (dst_orig->sa_len > sizeof(mdst)) return (EINVAL); rt_maskedcopy(dst_orig, (struct sockaddr *)&mdst, netmask); info->rti_info[RTAX_DST] = (struct sockaddr *)&mdst; } error = del_route(rnh, info, rc); info->rti_info[RTAX_DST] = dst_orig; return (error); } /* * Conditionally unlinks rtentry matching data inside @info from @rnh. * Returns 0 on success with operation result stored in @rc. * On error, returns: * ESRCH - if prefix was not found, * EADDRINUSE - if trying to delete higher priority route. * ENOENT - if supplied filter function returned 0 (not matched). */ static int rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc) { struct rtentry *rt; struct nhop_object *nh; struct radix_node *rn; struct route_nhop_data rnd; int error; rt = lookup_prefix(rnh, info, &rnd); if (rt == NULL) return (ESRCH); nh = rt->rt_nhop; error = check_info_match_nhop(info, rt, nh); if (error != 0) return (error); if (can_override_nhop(info, nh) < 0) return (EADDRINUSE); /* * Remove the item from the tree and return it. * Complain if it is not there and do no more processing. */ #ifdef RADIX_MPATH info->rti_info[RTAX_GATEWAY] = &nh->gw_sa; if (rt_mpath_capable(rnh)) { rn = rt_mpath_unlink(rnh, info, rt, &error); if (error != 0) return (error); } else #endif rn = rnh->rnh_deladdr(info->rti_info[RTAX_DST], info->rti_info[RTAX_NETMASK], &rnh->head); if (rn == NULL) return (ESRCH); if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) panic ("rtrequest delete"); rt = RNTORT(rn); rt->rte_flags &= ~RTF_UP; /* Finalize notification */ rnh->rnh_gen++; rc->rc_cmd = RTM_DELETE; rc->rc_rt = rt; rc->rc_nh_old = rt->rt_nhop; rc->rc_nh_weight = rt->rt_weight; rib_notify(rnh, RIB_NOTIFY_IMMEDIATE, rc); return (0); } static int del_route(struct rib_head *rnh, struct rt_addrinfo *info, struct rib_cmd_info *rc) { int error; RIB_WLOCK(rnh); error = rt_unlinkrte(rnh, info, rc); RIB_WUNLOCK(rnh); if (error != 0) return (error); rib_notify(rnh, RIB_NOTIFY_DELAYED, rc); /* * If the caller wants it, then it can have it, * the entry will be deleted after the end of the current epoch. */ rtfree(rc->rc_rt); return (0); } int rib_change_route(uint32_t fibnum, struct rt_addrinfo *info, struct rib_cmd_info *rc) { RIB_RLOCK_TRACKER; struct route_nhop_data rnd_orig; struct rib_head *rnh; struct rtentry *rt; int error; NET_EPOCH_ASSERT(); rnh = get_rnh(fibnum, info); if (rnh == NULL) return (EAFNOSUPPORT); bzero(rc, sizeof(struct rib_cmd_info)); rc->rc_cmd = RTM_CHANGE; /* Check if updated gateway exists */ if ((info->rti_flags & RTF_GATEWAY) && (info->rti_info[RTAX_GATEWAY] == NULL)) return (EINVAL); /* * route change is done in multiple steps, with dropping and * reacquiring lock. In the situations with multiple processes * changes the same route in can lead to the case when route * is changed between the steps. Address it by retrying the operation * multiple times before failing. */ RIB_RLOCK(rnh); rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST], info->rti_info[RTAX_NETMASK], &rnh->head); if (rt == NULL) { RIB_RUNLOCK(rnh); return (ESRCH); } #ifdef RADIX_MPATH /* * If we got multipath routes, * we require users to specify a matching RTAX_GATEWAY. */ if (rt_mpath_capable(rnh)) { rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]); if (rt == NULL) { RIB_RUNLOCK(rnh); return (ESRCH); } } #endif rnd_orig.rnd_nhop = rt->rt_nhop; rnd_orig.rnd_weight = rt->rt_weight; RIB_RUNLOCK(rnh); for (int i = 0; i < RIB_MAX_RETRIES; i++) { error = change_route(rnh, info, &rnd_orig, rc); if (error != EAGAIN) break; } return (error); } static int change_route(struct rib_head *rnh, struct rt_addrinfo *info, struct route_nhop_data *rnd_orig, struct rib_cmd_info *rc) { int error = 0; int free_ifa = 0; struct nhop_object *nh, *nh_orig; struct route_nhop_data rnd_new; nh = NULL; nh_orig = rnd_orig->rnd_nhop; if (nh_orig == NULL) return (ESRCH); /* * New gateway could require new ifaddr, ifp; * flags may also be different; ifp may be specified * by ll sockaddr when protocol address is ambiguous */ if (((nh_orig->nh_flags & NHF_GATEWAY) && info->rti_info[RTAX_GATEWAY] != NULL) || info->rti_info[RTAX_IFP] != NULL || (info->rti_info[RTAX_IFA] != NULL && !sa_equal(info->rti_info[RTAX_IFA], nh_orig->nh_ifa->ifa_addr))) { error = rt_getifa_fib(info, rnh->rib_fibnum); if (info->rti_ifa != NULL) free_ifa = 1; if (error != 0) { if (free_ifa) { ifa_free(info->rti_ifa); info->rti_ifa = NULL; } return (error); } } error = nhop_create_from_nhop(rnh, nh_orig, info, &nh); if (free_ifa) { ifa_free(info->rti_ifa); info->rti_ifa = NULL; } if (error != 0) return (error); rnd_new.rnd_nhop = nh; if (info->rti_mflags & RTV_WEIGHT) rnd_new.rnd_weight = info->rti_rmx->rmx_weight; else rnd_new.rnd_weight = rnd_orig->rnd_weight; error = change_route_conditional(rnh, NULL, info, rnd_orig, &rnd_new, rc); return (error); } /* * Insert @rt with nhop data from @rnd_new to @rnh. * Returns 0 on success and stores operation results in @rc. */ static int add_route_nhop(struct rib_head *rnh, struct rtentry *rt, struct rt_addrinfo *info, struct route_nhop_data *rnd, struct rib_cmd_info *rc) { struct sockaddr *ndst, *netmask; struct radix_node *rn; int error = 0; RIB_WLOCK_ASSERT(rnh); ndst = (struct sockaddr *)rt_key(rt); netmask = info->rti_info[RTAX_NETMASK]; rt->rt_nhop = rnd->rnd_nhop; rt->rt_weight = rnd->rnd_weight; rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes); if (rn != NULL) { if (rt->rt_expire > 0) tmproutes_update(rnh, rt); /* Finalize notification */ rnh->rnh_gen++; rc->rc_cmd = RTM_ADD; rc->rc_rt = rt; rc->rc_nh_old = NULL; rc->rc_nh_new = rnd->rnd_nhop; rc->rc_nh_weight = rnd->rnd_weight; rib_notify(rnh, RIB_NOTIFY_IMMEDIATE, rc); } else { /* Existing route or memory allocation failure */ error = EEXIST; } return (error); } /* * Switch @rt nhop/weigh to the ones specified in @rnd. * Conditionally set rt_expire if set in @info. * Returns 0 on success. */ static int change_route_nhop(struct rib_head *rnh, struct rtentry *rt, struct rt_addrinfo *info, struct route_nhop_data *rnd, struct rib_cmd_info *rc) { struct nhop_object *nh_orig; RIB_WLOCK_ASSERT(rnh); nh_orig = rt->rt_nhop; if (rnd->rnd_nhop != NULL) { /* Changing expiration & nexthop & weight to a new one */ rt_setmetrics(info, rt); rt->rt_nhop = rnd->rnd_nhop; rt->rt_weight = rnd->rnd_weight; if (rt->rt_expire > 0) tmproutes_update(rnh, rt); } else { /* Route deletion requested. */ struct sockaddr *ndst, *netmask; struct radix_node *rn; ndst = (struct sockaddr *)rt_key(rt); netmask = info->rti_info[RTAX_NETMASK]; rn = rnh->rnh_deladdr(ndst, netmask, &rnh->head); if (rn == NULL) return (ESRCH); } /* Finalize notification */ rnh->rnh_gen++; rc->rc_cmd = (rnd->rnd_nhop != NULL) ? RTM_CHANGE : RTM_DELETE; rc->rc_rt = rt; rc->rc_nh_old = nh_orig; rc->rc_nh_new = rnd->rnd_nhop; rc->rc_nh_weight = rnd->rnd_weight; rib_notify(rnh, RIB_NOTIFY_IMMEDIATE, rc); return (0); } /* * Conditionally update route nhop/weight IFF data in @nhd_orig is * consistent with the current route data. * Nexthop in @nhd_new is consumed. */ int change_route_conditional(struct rib_head *rnh, struct rtentry *rt, struct rt_addrinfo *info, struct route_nhop_data *rnd_orig, struct route_nhop_data *rnd_new, struct rib_cmd_info *rc) { struct rtentry *rt_new; int error = 0; RIB_WLOCK(rnh); rt_new = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST], info->rti_info[RTAX_NETMASK], &rnh->head); if (rt_new == NULL) { if (rnd_orig->rnd_nhop == NULL) error = add_route_nhop(rnh, rt, info, rnd_new, rc); else { /* * Prefix does not exist, which was not our assumption. * Update @rnd_orig with the new data and return */ rnd_orig->rnd_nhop = NULL; rnd_orig->rnd_weight = 0; error = EAGAIN; } } else { /* Prefix exists, try to update */ if (rnd_orig->rnd_nhop == rt_new->rt_nhop) { /* * Nhop/mpath group hasn't changed. Flip * to the new precalculated one and return */ error = change_route_nhop(rnh, rt_new, info, rnd_new, rc); } else { /* Update and retry */ rnd_orig->rnd_nhop = rt_new->rt_nhop; rnd_orig->rnd_weight = rt_new->rt_weight; error = EAGAIN; } } RIB_WUNLOCK(rnh); if (error == 0) { rib_notify(rnh, RIB_NOTIFY_DELAYED, rc); if (rnd_orig->rnd_nhop != NULL) nhop_free_any(rnd_orig->rnd_nhop); } else { if (rnd_new->rnd_nhop != NULL) nhop_free_any(rnd_new->rnd_nhop); } return (error); } /* * Performs modification of routing table specificed by @action. * Table is specified by @fibnum and sa_family in @info->rti_info[RTAX_DST]. * Needs to be run in network epoch. * * Returns 0 on success and fills in @rc with action result. */ int rib_action(uint32_t fibnum, int action, struct rt_addrinfo *info, struct rib_cmd_info *rc) { int error; switch (action) { case RTM_ADD: error = rib_add_route(fibnum, info, rc); break; case RTM_DELETE: error = rib_del_route(fibnum, info, rc); break; case RTM_CHANGE: error = rib_change_route(fibnum, info, rc); break; default: error = ENOTSUP; } return (error); } struct rt_delinfo { struct rt_addrinfo info; struct rib_head *rnh; struct rtentry *head; struct rib_cmd_info rc; }; /* * Conditionally unlinks @rn from radix tree based * on info data passed in @arg. */ static int rt_checkdelroute(struct radix_node *rn, void *arg) { struct rt_delinfo *di; struct rt_addrinfo *info; struct rtentry *rt; int error; di = (struct rt_delinfo *)arg; rt = (struct rtentry *)rn; info = &di->info; info->rti_info[RTAX_DST] = rt_key(rt); info->rti_info[RTAX_NETMASK] = rt_mask(rt); info->rti_info[RTAX_GATEWAY] = &rt->rt_nhop->gw_sa; error = rt_unlinkrte(di->rnh, info, &di->rc); /* * Add deleted rtentries to the list to GC them * after dropping the lock. * * XXX: Delayed notifications not implemented * for nexthop updates. */ if (error == 0) { /* Add to the list and return */ rt->rt_chain = di->head; di->head = rt; } return (0); } /* * Iterates over a routing table specified by @fibnum and @family and * deletes elements marked by @filter_f. * @fibnum: rtable id * @family: AF_ address family * @filter_f: function returning non-zero value for items to delete * @arg: data to pass to the @filter_f function * @report: true if rtsock notification is needed. */ void rib_walk_del(u_int fibnum, int family, rt_filter_f_t *filter_f, void *arg, bool report) { struct rib_head *rnh; struct rt_delinfo di; struct rtentry *rt; struct epoch_tracker et; rnh = rt_tables_get_rnh(fibnum, family); if (rnh == NULL) return; bzero(&di, sizeof(di)); di.info.rti_filter = filter_f; di.info.rti_filterdata = arg; di.rnh = rnh; di.rc.rc_cmd = RTM_DELETE; NET_EPOCH_ENTER(et); RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di); RIB_WUNLOCK(rnh); /* We might have something to reclaim. */ bzero(&di.rc, sizeof(di.rc)); di.rc.rc_cmd = RTM_DELETE; while (di.head != NULL) { rt = di.head; di.head = rt->rt_chain; rt->rt_chain = NULL; di.rc.rc_rt = rt; di.rc.rc_nh_old = rt->rt_nhop; rib_notify(rnh, RIB_NOTIFY_DELAYED, &di.rc); /* TODO std rt -> rt_addrinfo export */ di.info.rti_info[RTAX_DST] = rt_key(rt); di.info.rti_info[RTAX_NETMASK] = rt_mask(rt); if (report) rt_routemsg(RTM_DELETE, rt, rt->rt_nhop->nh_ifp, 0, fibnum); rtfree(rt); } NET_EPOCH_EXIT(et); } static void rib_notify(struct rib_head *rnh, enum rib_subscription_type type, struct rib_cmd_info *rc) { struct rib_subscription *rs; CK_STAILQ_FOREACH(rs, &rnh->rnh_subscribers, next) { if (rs->type == type) rs->func(rnh, rc, rs->arg); } } static struct rib_subscription * allocate_subscription(rib_subscription_cb_t *f, void *arg, enum rib_subscription_type type, bool waitok) { struct rib_subscription *rs; int flags = M_ZERO | (waitok ? M_WAITOK : 0); rs = malloc(sizeof(struct rib_subscription), M_RTABLE, flags); if (rs == NULL) return (NULL); rs->func = f; rs->arg = arg; rs->type = type; return (rs); } /* * Subscribe for the changes in the routing table specified by @fibnum and * @family. * * Returns pointer to the subscription structure on success. */ struct rib_subscription * rib_subscribe(uint32_t fibnum, int family, rib_subscription_cb_t *f, void *arg, enum rib_subscription_type type, bool waitok) { struct rib_head *rnh; struct rib_subscription *rs; struct epoch_tracker et; if ((rs = allocate_subscription(f, arg, type, waitok)) == NULL) return (NULL); NET_EPOCH_ENTER(et); KASSERT((fibnum < rt_numfibs), ("%s: bad fibnum", __func__)); rnh = rt_tables_get_rnh(fibnum, family); RIB_WLOCK(rnh); CK_STAILQ_INSERT_TAIL(&rnh->rnh_subscribers, rs, next); RIB_WUNLOCK(rnh); NET_EPOCH_EXIT(et); return (rs); } struct rib_subscription * rib_subscribe_internal(struct rib_head *rnh, rib_subscription_cb_t *f, void *arg, enum rib_subscription_type type, bool waitok) { struct rib_subscription *rs; struct epoch_tracker et; if ((rs = allocate_subscription(f, arg, type, waitok)) == NULL) return (NULL); NET_EPOCH_ENTER(et); RIB_WLOCK(rnh); CK_STAILQ_INSERT_TAIL(&rnh->rnh_subscribers, rs, next); RIB_WUNLOCK(rnh); NET_EPOCH_EXIT(et); return (rs); } /* * Remove rtable subscription @rs from the table specified by @fibnum * and @family. * Needs to be run in network epoch. * * Returns 0 on success. */ int rib_unsibscribe(uint32_t fibnum, int family, struct rib_subscription *rs) { struct rib_head *rnh; NET_EPOCH_ASSERT(); KASSERT((fibnum < rt_numfibs), ("%s: bad fibnum", __func__)); rnh = rt_tables_get_rnh(fibnum, family); if (rnh == NULL) return (ENOENT); RIB_WLOCK(rnh); CK_STAILQ_REMOVE(&rnh->rnh_subscribers, rs, rib_subscription, next); RIB_WUNLOCK(rnh); epoch_call(net_epoch_preempt, destroy_subscription_epoch, &rs->epoch_ctx); return (0); } /* * Epoch callback indicating subscription is safe to destroy */ static void destroy_subscription_epoch(epoch_context_t ctx) { struct rib_subscription *rs; rs = __containerof(ctx, struct rib_subscription, epoch_ctx); free(rs, M_RTABLE); } void rib_init_subscriptions(struct rib_head *rnh) { CK_STAILQ_INIT(&rnh->rnh_subscribers); } void rib_destroy_subscriptions(struct rib_head *rnh) { struct rib_subscription *rs; struct epoch_tracker et; NET_EPOCH_ENTER(et); RIB_WLOCK(rnh); while ((rs = CK_STAILQ_FIRST(&rnh->rnh_subscribers)) != NULL) { CK_STAILQ_REMOVE_HEAD(&rnh->rnh_subscribers, next); epoch_call(net_epoch_preempt, destroy_subscription_epoch, &rs->epoch_ctx); } RIB_WUNLOCK(rnh); NET_EPOCH_EXIT(et); }