#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_inet6.h"
#include <sys/types.h>
#include <sys/malloc.h>
#include <sys/rmlock.h>
#include <sys/socket.h>
#include <net/if.h>
#include <net/route.h>
#include <net/route/nhop.h>
#include <net/route/route_ctl.h>
#include <net/route/route_var.h>
#include <netinet6/scope6_var.h>
#include <netlink/netlink.h>
#include <netlink/netlink_ctl.h>
#include <netlink/netlink_route.h>
#include <netlink/route/route_var.h>
#define DEBUG_MOD_NAME nl_route
#define DEBUG_MAX_LEVEL LOG_DEBUG3
#include <netlink/netlink_debug.h>
_DECLARE_DEBUG(LOG_INFO);
static unsigned char
get_rtm_type(const struct nhop_object *nh)
{
int nh_flags = nh->nh_flags;
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)
{
const struct nhgrp_object *nhg = (const struct nhgrp_object *)nh;
int rt_flags;
uint8_t origin;
if (NH_IS_NHGRP(nh)) {
origin = nhgrp_get_origin(nhg);
if (origin != RTPROT_UNSPEC)
return (origin);
nh = nhg->nhops[0];
}
origin = nhop_get_origin(nh);
if (origin != RTPROT_UNSPEC)
return (origin);
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);
}
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)
{
#ifdef INET6
int upper_family;
#endif
switch (nhop_get_neigh_family(nh)) {
case AF_LINK:
break;
case AF_INET:
nlattr_add(nw, NL_RTA_GATEWAY, 4, &nh->gw4_sa.sin_addr);
break;
#ifdef INET6
case AF_INET6:
upper_family = nhop_get_upper_family(nh);
if (upper_family == AF_INET6) {
struct in6_addr gw6 = nh->gw6_sa.sin6_addr;
in6_clearscope(&gw6);
nlattr_add(nw, NL_RTA_GATEWAY, 16, &gw6);
} else if (upper_family == AF_INET) {
struct in6_addr gw6 = nh->gw6_sa.sin6_addr;
in6_clearscope(&gw6);
char buf[20];
struct rtvia *via = (struct rtvia *)&buf[0];
via->rtvia_family = AF_INET6;
memcpy(via->rtvia_addr, &gw6, 16);
nlattr_add(nw, NL_RTA_VIA, 17, via);
}
break;
#endif
}
}
static void
dump_rc_nhop_prefsrc(struct nl_writer *nw, const struct nhop_object *nh)
{
const struct ifaddr *ifa = nh->nh_ifa;
#ifdef INET6
struct in6_addr in6;
#endif
if ((nh->nh_flags & NHF_PREFSRC) == 0)
return;
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
nlattr_add(nw, NL_RTA_PREFSRC, 4,
&((struct sockaddr_in *)((ifa)->ifa_addr))->sin_addr);
break;
#endif
#ifdef INET6
case AF_INET6:
in6 = ((struct sockaddr_in6 *)((ifa)->ifa_addr))->sin6_addr;
in6_clearscope(&in6);
nlattr_add(nw, NL_RTA_PREFSRC, 16, &in6);
break;
#endif
}
}
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;
}
static void
dump_rc_nhg(struct nl_writer *nw, const struct route_nhop_data *rnd, struct rtmsg *rtm)
{
const struct nhgrp_object *nhg = rnd->rnd_nhgrp;
const struct weightened_nhop *wn;
struct nhop_object *nh;
uint32_t uidx, num_nhops, nh_expire;
uint32_t base_rtflags, rtflags, nhop_weight, nhop_metric;
MPASS((NH_IS_NHGRP(rnd->rnd_nhop)));
nhop_weight = RT_DEFAULT_WEIGHT;
nhop_metric = RT_DEFAULT_METRIC;
nh = nhop_select_func(rnd->rnd_nhop, 0);
rtflags = nhop_get_rtflags(nh);
if (nh->nh_flags & NHF_GATEWAY)
dump_rc_nhop_gw(nw, nh);
dump_rc_nhop_prefsrc(nw, nh);
wn = nhgrp_get_nhops(nhg, &num_nhops);
base_rtflags = nhop_get_rtflags(wn[0].nh);
uidx = nhgrp_get_uidx(nhg);
if (uidx != 0)
nlattr_add_u32(nw, NL_RTA_NH_ID, uidx);
nlattr_add_u32(nw, NL_RTA_KNH_ID, nhgrp_get_idx(nhg));
nlattr_add_u32(nw, NL_RTA_RTFLAGS, base_rtflags);
if (rtflags & RTF_FIXEDMTU)
dump_rc_nhop_mtu(nw, nh);
nlattr_add_u32(nw, NL_RTA_OIF, if_getindex(nh->nh_ifp));
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 = if_getindex(wn[i].nh->nh_ifp);
rtnh->rtnh_hops = MIN(wn[i].weight, UINT8_MAX);
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);
nlattr_add_u32(nw, NL_RTA_PRIORITY, nhop_get_metric(wn[i].nh));
nh_expire = nhop_get_expire(wn[i].nh);
if (nh_expire > 0)
nlattr_add_u32(nw, NL_RTA_EXPIRES, nh_expire - time_uptime);
rtnh = nlattr_restore_offset(nw, nh_off, struct rtnexthop);
if (nh == wn[i].nh) {
nhop_weight = wn[i].weight;
nhop_metric = nhop_get_metric(wn[i].nh);
}
rtnh->rtnh_len = nlattr_save_offset(nw) - nh_off;
}
nlattr_set_len(nw, off);
nlattr_add_u32(nw, NL_RTA_PRIORITY, nhop_metric);
if (nhop_weight != RT_DEFAULT_WEIGHT)
nlattr_add_u32(nw, NL_RTA_WEIGHT, nhop_weight);
}
static void
dump_rc_nhop(struct nl_writer *nw, const struct route_nhop_data *rnd, struct rtmsg *rtm)
{
const struct nhop_object *nh = rnd->rnd_nhop;
uint32_t rtflags, uidx, nh_expire;
if (NH_IS_NHGRP(rnd->rnd_nhop)) {
dump_rc_nhg(nw, rnd, rtm);
return;
}
rtflags = nhop_get_rtflags(nh);
if (nh->nh_flags & NHF_GATEWAY)
dump_rc_nhop_gw(nw, nh);
dump_rc_nhop_prefsrc(nw, nh);
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);
nh_expire = nhop_get_expire(nh);
if (nh_expire > 0)
nlattr_add_u32(nw, NL_RTA_EXPIRES, nh_expire - time_uptime);
nlattr_add_u32(nw, NL_RTA_OIF, if_getindex(nh->nh_ifp));
nlattr_add_u32(nw, NL_RTA_PRIORITY, nhop_get_metric(nh));
if (rnd->rnd_weight != RT_DEFAULT_WEIGHT)
nlattr_add_u32(nw, NL_RTA_WEIGHT, rnd->rnd_weight);
}
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;
rtm->rtm_protocol = nl_get_rtm_protocol(rnd->rnd_nhop);
rtm->rtm_type = get_rtm_type(rnd->rnd_nhop);
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;
if (NH_IS_NHGRP(rnd->rnd_nhop))
dump_rc_nhg(nw, rnd, rtm);
else
dump_rc_nhop(nw, rnd, 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 (nl_writer_group(&nw, NLMSG_SMALL, NETLINK_ROUTE, group_id, 0,
false)) {
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);
}
static void
set_scope6(struct sockaddr *sa, struct ifnet *ifp)
{
#ifdef INET6
if (sa != NULL && sa->sa_family == AF_INET6 && ifp != NULL) {
struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
if (IN6_IS_ADDR_LINKLOCAL(&sa6->sin6_addr))
in6_set_unicast_scopeid(&sa6->sin6_addr, if_getindex(ifp));
}
#endif
}
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
static bool
post_p_rtnh(void *_attrs, struct nl_pstate *npt __unused)
{
struct rta_mpath_nh *attrs = (struct rta_mpath_nh *)_attrs;
set_scope6(attrs->gw, attrs->ifp);
return (true);
}
NL_DECLARE_PARSER_EXT(mpath_parser, struct rtnexthop, NULL, nlf_p_rtnh, nla_p_rtnh, post_p_rtnh);
struct rta_mpath {
u_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)
{
struct rta_mpath *mp;
struct rtnexthop *rtnh;
uint16_t data_len, len;
u_int max_nhops;
int error;
data_len = nla->nla_len - sizeof(struct nlattr);
max_nhops = data_len / sizeof(struct rtnexthop);
mp = npt_alloc(npt, (max_nhops + 2) * sizeof(struct rta_mpath_nh));
if (mp == NULL) {
NLMSG_REPORT_ERR_MSG(npt, "%s: too many RTA_MULTIPATH", __func__);
return (ENOMEM);
}
mp->num_nhops = 0;
for (rtnh = (struct rtnexthop *)(nla + 1); data_len > 0; ) {
struct rta_mpath_nh *mpnh;
len = NL_ITEM_ALIGN(rtnh->rtnh_len);
if (__predict_false(rtnh->rtnh_len <= sizeof(*rtnh) ||
len < rtnh->rtnh_len || len > data_len)) {
NLMSG_REPORT_ERR_MSG(npt, "%s: bad length %u",
__func__, rtnh->rtnh_len);
return (EINVAL);
}
mpnh = &mp->nhops[mp->num_nhops++];
error = nl_parse_header(rtnh, rtnh->rtnh_len, &mpath_parser,
npt, mpnh);
if (error != 0) {
NLMSG_REPORT_ERR_MSG(npt,
"RTA_MULTIPATH: nexthop %u: parse failed",
mp->num_nhops - 1);
return (error);
}
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 sockaddr *rta_pref_src;
struct ifnet *rta_oif;
struct rta_mpath *rta_multipath;
uint32_t rta_table;
uint32_t rta_rtflags;
uint32_t rta_nh_id;
uint32_t rta_metric;
uint32_t rta_weight;
uint32_t rta_expire;
uint32_t rtax_mtu;
uint8_t rtm_table;
uint8_t rtm_family;
uint8_t rtm_dst_len;
uint8_t rtm_protocol;
uint8_t rtm_type;
uint32_t rtm_flags;
};
#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_PRIORITY, .off = _OUT(rta_metric), .cb = nlattr_get_uint32 },
{ .type = NL_RTA_PREFSRC, .off = _OUT(rta_pref_src), .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_WEIGHT, .off = _OUT(rta_weight), .cb = nlattr_get_uint32 },
{ .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_EXPIRES, .off = _OUT(rta_expire), .cb = nlattr_get_uint32 },
{ .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 },
{ .off_in = _IN(rtm_protocol), .off_out = _OUT(rtm_protocol), .cb = nlf_get_u8 },
{ .off_in = _IN(rtm_type), .off_out = _OUT(rtm_type), .cb = nlf_get_u8 },
{ .off_in = _IN(rtm_table), .off_out = _OUT(rtm_table), .cb = nlf_get_u8 },
{ .off_in = _IN(rtm_flags), .off_out = _OUT(rtm_flags), .cb = nlf_get_u32 },
};
#undef _IN
#undef _OUT
static bool
post_p_rtmsg(void *_attrs, struct nl_pstate *npt __unused)
{
struct nl_parsed_route *attrs = (struct nl_parsed_route *)_attrs;
set_scope6(attrs->rta_dst, attrs->rta_oif);
set_scope6(attrs->rta_gw, attrs->rta_oif);
set_scope6(attrs->rta_pref_src, attrs->rta_oif);
return (true);
}
NL_DECLARE_PARSER_EXT(rtm_parser, struct rtmsg, NULL, nlf_p_rtmsg, nla_p_rtmsg, post_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);
if (!rt_is_exportable(rt, nlp_get_cred(wa->nlp)))
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, error %d",
rt_print_buf(rt, rtbuf, sizeof(rtbuf)), 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);
}
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;
const struct rtentry *rt;
struct route_nhop_data rnd;
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);
}
rnh = rt_tables_get_rnh(fibnum, family);
if (rnh == NULL)
return (EAFNOSUPPORT);
RIB_RLOCK(rnh);
struct sockaddr *dst = attrs->rta_dst;
if (attrs->rtm_flags & RTM_F_PREFIX)
rt = rib_lookup_prefix_plen(rnh, dst, attrs->rtm_dst_len, &rnd);
else
rt = (const struct rtentry *)rnh->rnh_matchaddr(dst, &rnh->head);
if (rt == NULL) {
RIB_RUNLOCK(rnh);
return (ESRCH);
}
rt_get_rnd(rt, &rnd);
RIB_RUNLOCK(rnh);
if (!rt_is_exportable(rt, nlp_get_cred(nlp)))
return (ESRCH);
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, i, 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;
}
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, const struct sockaddr *dst, int *perror)
{
if (nh->gw_sa.sa_family == 0) {
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);
}
} else {
if (nh->nh_ifp == NULL) {
uint32_t fibnum = nhop_get_fibnum(nh);
uint32_t flags = 0;
if (nh->nh_flags & NHF_GATEWAY)
flags = RTF_GATEWAY;
else if (nh->nh_flags & NHF_HOST)
flags = RTF_HOST;
struct ifaddr *ifa = ifa_ifwithroute(flags, dst, &nh->gw_sa, fibnum);
if (ifa == NULL) {
NL_LOG(LOG_DEBUG, "Unable to determine ifp, skipping");
*perror = EINVAL;
return (NULL);
}
nhop_set_transmit_ifp(nh, ifa->ifa_ifp);
}
}
if (nh->nh_ifa == NULL) {
const struct sockaddr *gw_sa = &nh->gw_sa;
if (gw_sa->sa_family != dst->sa_family) {
gw_sa = dst;
}
struct ifaddr *ifa = ifaof_ifpforaddr(gw_sa, nh->nh_ifp);
if (ifa == NULL) {
gw_sa = &nh->gw_sa;
ifa = ifaof_ifpforaddr(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 == 128)
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);
}
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);
error = nl_set_nexthop_gw(nh, mpnh->gw, mpnh->ifp, npt);
if (error != 0) {
nhop_free(nh);
return (error);
}
if (mpnh->ifp != NULL)
nhop_set_transmit_ifp(nh, mpnh->ifp);
if (attrs->rta_pref_src != NULL) {
error = nl_set_nexthop_prefsrc(nh, attrs->rta_pref_src);
if (error != 0) {
nhop_free(nh);
return (error);
}
}
nhop_set_pxtype_flag(nh, get_pxflag(attrs));
nhop_set_rtflags(nh, attrs->rta_rtflags);
nhop_set_metric(nh, attrs->rta_metric);
if (attrs->rtm_protocol > RTPROT_STATIC)
nhop_set_origin(nh, attrs->rtm_protocol);
*pnh = finalize_nhop(nh, attrs->rta_dst, &error);
return (error);
}
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;
uint32_t nh_expire = 0;
if (attrs->rta_multipath != NULL) {
int num_nhops = attrs->rta_multipath->num_nhops;
struct weightened_nhop *wn = npt_alloc(npt, sizeof(*wn) * num_nhops);
if (wn == NULL) {
*perror = ENOMEM;
return (NULL);
}
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);
struct nhgrp_object *nhg;
nhg = nhgrp_alloc(rh->rib_fibnum, rh->rib_family,
wn, num_nhops, perror);
if (nhg != NULL) {
if (attrs->rtm_protocol > RTPROT_STATIC)
nhgrp_set_origin(nhg, attrs->rtm_protocol);
nhg = nhgrp_get_nhgrp(nhg, perror);
}
for (int i = 0; i < num_nhops; i++)
nhop_free(wn[i].nh);
if (nhg != NULL)
return ((struct nhop_object *)nhg);
error = *perror;
}
*perror = error;
} else {
nh = nhop_alloc(attrs->rta_table, attrs->rtm_family);
if (nh == NULL) {
*perror = ENOMEM;
return (NULL);
}
if (attrs->rta_gw != NULL) {
*perror = nl_set_nexthop_gw(nh, attrs->rta_gw, attrs->rta_oif, npt);
if (*perror != 0) {
nhop_free(nh);
return (NULL);
}
}
if (attrs->rta_oif != NULL)
nhop_set_transmit_ifp(nh, attrs->rta_oif);
if (attrs->rta_pref_src != NULL) {
*perror = nl_set_nexthop_prefsrc(nh, attrs->rta_pref_src);
if (*perror != 0) {
nhop_free(nh);
return (NULL);
}
}
if (attrs->rtax_mtu != 0)
nhop_set_mtu(nh, attrs->rtax_mtu, true);
if (attrs->rta_expire > 0) {
nh_expire = attrs->rta_expire - time_second + time_uptime;
nhop_set_expire(nh, nh_expire);
}
if (attrs->rta_rtflags & RTF_BROADCAST)
nhop_set_broadcast(nh, true);
if (attrs->rtm_protocol > RTPROT_STATIC)
nhop_set_origin(nh, attrs->rtm_protocol);
nhop_set_metric(nh, attrs->rta_metric);
nhop_set_pxtype_flag(nh, get_pxflag(attrs));
nhop_set_rtflags(nh, attrs->rta_rtflags);
switch (attrs->rtm_type) {
case RTN_UNICAST:
break;
case RTN_BLACKHOLE:
nhop_set_blackhole(nh, RTF_BLACKHOLE);
break;
case RTN_PROHIBIT:
case RTN_UNREACHABLE:
nhop_set_blackhole(nh, RTF_REJECT);
break;
}
nh = finalize_nhop(nh, attrs->rta_dst, perror);
}
return (nh);
}
static inline void
old_linux_compat(struct nl_parsed_route *attrs)
{
if (attrs->rtm_table > 0 && attrs->rta_table == 0)
attrs->rta_table = attrs->rtm_table;
}
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);
if (attrs.rta_dst == NULL) {
NL_LOG(LOG_DEBUG, "missing RTA_DST");
return (EINVAL);
}
old_linux_compat(&attrs);
if (attrs.rta_table >= V_rt_numfibs || attrs.rtm_family > AF_MAX) {
NLMSG_REPORT_ERR_MSG(npt, "invalid fib");
return (EINVAL);
}
if (attrs.rta_nh_id != 0) {
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);
}
}
if (!NH_IS_NHGRP(nh) && attrs.rta_weight == 0)
attrs.rta_weight = RT_DEFAULT_WEIGHT;
struct route_nhop_data rnd = { .rnd_nhop = nh, .rnd_weight = attrs.rta_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_metric != 0 && attrs->rta_metric != nhop_get_metric(nh))
return (0);
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);
}
old_linux_compat(&attrs);
if (attrs.rta_table >= V_rt_numfibs || attrs.rtm_family > AF_MAX) {
NLMSG_REPORT_ERR_MSG(npt, "invalid fib");
return (EINVAL);
}
error = rib_del_route_px(attrs.rta_table, attrs.rta_dst,
attrs.rtm_dst_len, path_match_func, &attrs,
(attrs.rta_rtflags & RTF_PINNED) ? RTM_F_FORCE : 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);
old_linux_compat(&attrs);
if (attrs.rta_table >= V_rt_numfibs || attrs.rtm_family > AF_MAX) {
NLMSG_REPORT_ERR_MSG(npt, "invalid fib");
return (EINVAL);
}
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)
{
struct nl_writer nw;
int family, nlm_flags = 0;
family = rt_get_family(rc->rc_rt);
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 (!nl_writer_group(&nw, NLMSG_SMALL, NETLINK_ROUTE, group_id, 0,
false)) {
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,
.flags = RTNL_F_ALLOW_NONVNET_JAIL,
},
{
.cmd = NL_RTM_DELROUTE,
.name = "RTM_DELROUTE",
.cb = &rtnl_handle_delroute,
.priv = PRIV_NET_ROUTE,
.flags = RTNL_F_ALLOW_NONVNET_JAIL,
},
{
.cmd = NL_RTM_NEWROUTE,
.name = "RTM_NEWROUTE",
.cb = &rtnl_handle_newroute,
.priv = PRIV_NET_ROUTE,
.flags = RTNL_F_ALLOW_NONVNET_JAIL,
}
};
static const struct nlhdr_parser *all_parsers[] = {&mpath_parser, &metrics_parser, &rtm_parser};
void
rtnl_routes_init(void)
{
NL_VERIFY_PARSERS(all_parsers);
rtnl_register_messages(cmd_handlers, nitems(cmd_handlers));
}
