freebsd-dev/sys/netinet/ip_carp.c

2298 lines
53 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2002 Michael Shalayeff.
* Copyright (c) 2003 Ryan McBride.
* Copyright (c) 2011 Gleb Smirnoff <glebius@FreeBSD.org>
* All rights reserved.
*
* 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 ``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 HIS RELATIVES 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 MIND, 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 <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_bpf.h"
#include "opt_inet.h"
#include "opt_inet6.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/jail.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/priv.h>
#include <sys/proc.h>
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/taskqueue.h>
#include <sys/counter.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_var.h>
2005-11-14 12:50:23 +00:00
#include <net/if_dl.h>
#include <net/if_llatbl.h>
#include <net/if_types.h>
#include <net/route.h>
#include <net/vnet.h>
#if defined(INET) || defined(INET6)
#include <netinet/in.h>
#include <netinet/in_var.h>
#include <netinet/ip_carp.h>
#include <netinet/ip.h>
#include <machine/in_cksum.h>
#endif
#ifdef INET
#include <netinet/ip_var.h>
#include <netinet/if_ether.h>
#endif
#ifdef INET6
#include <netinet/icmp6.h>
#include <netinet/ip6.h>
#include <netinet6/in6_var.h>
#include <netinet6/ip6_var.h>
2005-07-25 12:36:43 +00:00
#include <netinet6/scope6_var.h>
#include <netinet6/nd6.h>
#endif
#include <crypto/sha1.h>
static MALLOC_DEFINE(M_CARP, "CARP", "CARP addresses");
struct carp_softc {
struct ifnet *sc_carpdev; /* Pointer to parent ifnet. */
struct ifaddr **sc_ifas; /* Our ifaddrs. */
struct sockaddr_dl sc_addr; /* Our link level address. */
struct callout sc_ad_tmo; /* Advertising timeout. */
#ifdef INET
struct callout sc_md_tmo; /* Master down timeout. */
#endif
#ifdef INET6
struct callout sc_md6_tmo; /* XXX: Master down timeout. */
#endif
struct mtx sc_mtx;
int sc_vhid;
int sc_advskew;
int sc_advbase;
int sc_naddrs;
int sc_naddrs6;
int sc_ifasiz;
enum { INIT = 0, BACKUP, MASTER } sc_state;
int sc_suppress;
int sc_sendad_errors;
#define CARP_SENDAD_MAX_ERRORS 3
int sc_sendad_success;
#define CARP_SENDAD_MIN_SUCCESS 3
int sc_init_counter;
uint64_t sc_counter;
/* authentication */
#define CARP_HMAC_PAD 64
unsigned char sc_key[CARP_KEY_LEN];
unsigned char sc_pad[CARP_HMAC_PAD];
SHA1_CTX sc_sha1;
TAILQ_ENTRY(carp_softc) sc_list; /* On the carp_if list. */
LIST_ENTRY(carp_softc) sc_next; /* On the global list. */
};
struct carp_if {
#ifdef INET
int cif_naddrs;
#endif
#ifdef INET6
int cif_naddrs6;
#endif
TAILQ_HEAD(, carp_softc) cif_vrs;
#ifdef INET
struct ip_moptions cif_imo;
#endif
#ifdef INET6
struct ip6_moptions cif_im6o;
#endif
struct ifnet *cif_ifp;
struct mtx cif_mtx;
uint32_t cif_flags;
#define CIF_PROMISC 0x00000001
};
#define CARP_INET 0
#define CARP_INET6 1
static int proto_reg[] = {-1, -1};
/*
* Brief design of carp(4).
*
* Any carp-capable ifnet may have a list of carp softcs hanging off
* its ifp->if_carp pointer. Each softc represents one unique virtual
* host id, or vhid. The softc has a back pointer to the ifnet. All
* softcs are joined in a global list, which has quite limited use.
*
* Any interface address that takes part in CARP negotiation has a
* pointer to the softc of its vhid, ifa->ifa_carp. That could be either
* AF_INET or AF_INET6 address.
*
* Although, one can get the softc's backpointer to ifnet and traverse
* through its ifp->if_addrhead queue to find all interface addresses
* involved in CARP, we keep a growable array of ifaddr pointers. This
* allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that
* do calls into the network stack, thus avoiding LORs.
*
* Locking:
*
* Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(),
* callout-driven events and ioctl()s.
*
* To traverse the list of softcs on an ifnet we use CIF_LOCK() or carp_sx.
* To traverse the global list we use the mutex carp_mtx.
*
* Known issues with locking:
*
* - Sending ad, we put the pointer to the softc in an mtag, and no reference
* counting is done on the softc.
* - On module unload we may race (?) with packet processing thread
* dereferencing our function pointers.
*/
/* Accept incoming CARP packets. */
VNET_DEFINE_STATIC(int, carp_allow) = 1;
#define V_carp_allow VNET(carp_allow)
/* Set DSCP in outgoing CARP packets. */
VNET_DEFINE_STATIC(int, carp_dscp) = 56;
#define V_carp_dscp VNET(carp_dscp)
/* Preempt slower nodes. */
VNET_DEFINE_STATIC(int, carp_preempt) = 0;
#define V_carp_preempt VNET(carp_preempt)
/* Log level. */
VNET_DEFINE_STATIC(int, carp_log) = 1;
#define V_carp_log VNET(carp_log)
/* Global advskew demotion. */
VNET_DEFINE_STATIC(int, carp_demotion) = 0;
#define V_carp_demotion VNET(carp_demotion)
/* Send error demotion factor. */
VNET_DEFINE_STATIC(int, carp_senderr_adj) = CARP_MAXSKEW;
#define V_carp_senderr_adj VNET(carp_senderr_adj)
/* Iface down demotion factor. */
VNET_DEFINE_STATIC(int, carp_ifdown_adj) = CARP_MAXSKEW;
#define V_carp_ifdown_adj VNET(carp_ifdown_adj)
static int carp_allow_sysctl(SYSCTL_HANDLER_ARGS);
static int carp_dscp_sysctl(SYSCTL_HANDLER_ARGS);
static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS);
SYSCTL_NODE(_net_inet, IPPROTO_CARP, carp, CTLFLAG_RW, 0, "CARP");
SYSCTL_PROC(_net_inet_carp, OID_AUTO, allow,
CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_allow_sysctl, "I",
"Accept incoming CARP packets");
SYSCTL_PROC(_net_inet_carp, OID_AUTO, dscp,
CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_dscp_sysctl, "I",
"DSCP value for carp packets");
SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode");
SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(carp_log), 0, "CARP log level");
SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion,
CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
0, 0, carp_demote_adj_sysctl, "I",
"Adjust demotion factor (skew of advskew)");
SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor,
CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment");
SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor,
CTLFLAG_VNET | CTLFLAG_RW,
&VNET_NAME(carp_ifdown_adj), 0,
"Interface down demotion factor adjustment");
VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats);
VNET_PCPUSTAT_SYSINIT(carpstats);
VNET_PCPUSTAT_SYSUNINIT(carpstats);
#define CARPSTATS_ADD(name, val) \
counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \
sizeof(uint64_t)], (val))
#define CARPSTATS_INC(name) CARPSTATS_ADD(name, 1)
SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats,
carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)");
#define CARP_LOCK_INIT(sc) mtx_init(&(sc)->sc_mtx, "carp_softc", \
NULL, MTX_DEF)
#define CARP_LOCK_DESTROY(sc) mtx_destroy(&(sc)->sc_mtx)
#define CARP_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
#define CARP_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
#define CARP_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
#define CIF_LOCK_INIT(cif) mtx_init(&(cif)->cif_mtx, "carp_if", \
NULL, MTX_DEF)
#define CIF_LOCK_DESTROY(cif) mtx_destroy(&(cif)->cif_mtx)
#define CIF_LOCK_ASSERT(cif) mtx_assert(&(cif)->cif_mtx, MA_OWNED)
#define CIF_LOCK(cif) mtx_lock(&(cif)->cif_mtx)
#define CIF_UNLOCK(cif) mtx_unlock(&(cif)->cif_mtx)
#define CIF_FREE(cif) do { \
CIF_LOCK(cif); \
if (TAILQ_EMPTY(&(cif)->cif_vrs)) \
carp_free_if(cif); \
else \
CIF_UNLOCK(cif); \
} while (0)
#define CARP_LOG(...) do { \
if (V_carp_log > 0) \
log(LOG_INFO, "carp: " __VA_ARGS__); \
} while (0)
#define CARP_DEBUG(...) do { \
if (V_carp_log > 1) \
log(LOG_DEBUG, __VA_ARGS__); \
} while (0)
#define IFNET_FOREACH_IFA(ifp, ifa) \
ifnet: Replace if_addr_lock rwlock with epoch + mutex Run on LLNW canaries and tested by pho@ gallatin: Using a 14-core, 28-HTT single socket E5-2697 v3 with a 40GbE MLX5 based ConnectX 4-LX NIC, I see an almost 12% improvement in received packet rate, and a larger improvement in bytes delivered all the way to userspace. When the host receiving 64 streams of netperf -H $DUT -t UDP_STREAM -- -m 1, I see, using nstat -I mce0 1 before the patch: InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 4.98 0.00 4.42 0.00 4235592 33 83.80 4720653 2149771 1235 247.32 4.73 0.00 4.20 0.00 4025260 33 82.99 4724900 2139833 1204 247.32 4.72 0.00 4.20 0.00 4035252 33 82.14 4719162 2132023 1264 247.32 4.71 0.00 4.21 0.00 4073206 33 83.68 4744973 2123317 1347 247.32 4.72 0.00 4.21 0.00 4061118 33 80.82 4713615 2188091 1490 247.32 4.72 0.00 4.21 0.00 4051675 33 85.29 4727399 2109011 1205 247.32 4.73 0.00 4.21 0.00 4039056 33 84.65 4724735 2102603 1053 247.32 After the patch InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 5.43 0.00 4.20 0.00 3313143 33 84.96 5434214 1900162 2656 245.51 5.43 0.00 4.20 0.00 3308527 33 85.24 5439695 1809382 2521 245.51 5.42 0.00 4.19 0.00 3316778 33 87.54 5416028 1805835 2256 245.51 5.42 0.00 4.19 0.00 3317673 33 90.44 5426044 1763056 2332 245.51 5.42 0.00 4.19 0.00 3314839 33 88.11 5435732 1792218 2499 245.52 5.44 0.00 4.19 0.00 3293228 33 91.84 5426301 1668597 2121 245.52 Similarly, netperf reports 230Mb/s before the patch, and 270Mb/s after the patch Reviewed by: gallatin Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D15366
2018-05-18 20:13:34 +00:00
CK_STAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link) \
if ((ifa)->ifa_carp != NULL)
#define CARP_FOREACH_IFA(sc, ifa) \
CARP_LOCK_ASSERT(sc); \
for (int _i = 0; \
_i < (sc)->sc_naddrs + (sc)->sc_naddrs6 && \
((ifa) = sc->sc_ifas[_i]) != NULL; \
++_i)
#define IFNET_FOREACH_CARP(ifp, sc) \
KASSERT(mtx_owned(&ifp->if_carp->cif_mtx) || \
sx_xlocked(&carp_sx), ("cif_vrs not locked")); \
TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list)
#define DEMOTE_ADVSKEW(sc) \
(((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ? \
CARP_MAXSKEW : ((sc)->sc_advskew + V_carp_demotion))
2005-02-26 10:33:14 +00:00
static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t);
static struct carp_softc
*carp_alloc(struct ifnet *);
static void carp_destroy(struct carp_softc *);
static struct carp_if
*carp_alloc_if(struct ifnet *);
static void carp_free_if(struct carp_if *);
static void carp_set_state(struct carp_softc *, int, const char* reason);
static void carp_sc_state(struct carp_softc *);
static void carp_setrun(struct carp_softc *, sa_family_t);
2005-02-26 10:33:14 +00:00
static void carp_master_down(void *);
static void carp_master_down_locked(struct carp_softc *,
const char* reason);
static void carp_send_ad(void *);
static void carp_send_ad_locked(struct carp_softc *);
static void carp_addroute(struct carp_softc *);
static void carp_ifa_addroute(struct ifaddr *);
static void carp_delroute(struct carp_softc *);
static void carp_ifa_delroute(struct ifaddr *);
static void carp_send_ad_all(void *, int);
static void carp_demote_adj(int, char *);
static LIST_HEAD(, carp_softc) carp_list;
static struct mtx carp_mtx;
static struct sx carp_sx;
static struct task carp_sendall_task =
TASK_INITIALIZER(0, carp_send_ad_all, NULL);
2005-02-26 10:33:14 +00:00
static void
carp_hmac_prepare(struct carp_softc *sc)
{
uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
uint8_t vhid = sc->sc_vhid & 0xff;
struct ifaddr *ifa;
int i, found;
#ifdef INET
struct in_addr last, cur, in;
#endif
#ifdef INET6
struct in6_addr last6, cur6, in6;
#endif
CARP_LOCK_ASSERT(sc);
/* Compute ipad from key. */
bzero(sc->sc_pad, sizeof(sc->sc_pad));
bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
for (i = 0; i < sizeof(sc->sc_pad); i++)
sc->sc_pad[i] ^= 0x36;
/* Precompute first part of inner hash. */
SHA1Init(&sc->sc_sha1);
SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
#ifdef INET
cur.s_addr = 0;
do {
found = 0;
last = cur;
cur.s_addr = 0xffffffff;
CARP_FOREACH_IFA(sc, ifa) {
in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr;
if (ifa->ifa_addr->sa_family == AF_INET &&
ntohl(in.s_addr) > ntohl(last.s_addr) &&
ntohl(in.s_addr) < ntohl(cur.s_addr)) {
cur.s_addr = in.s_addr;
found++;
}
}
if (found)
SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur));
} while (found);
#endif /* INET */
#ifdef INET6
memset(&cur6, 0, sizeof(cur6));
do {
found = 0;
last6 = cur6;
memset(&cur6, 0xff, sizeof(cur6));
CARP_FOREACH_IFA(sc, ifa) {
in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
if (IN6_IS_SCOPE_EMBED(&in6))
in6.s6_addr16[1] = 0;
if (ifa->ifa_addr->sa_family == AF_INET6 &&
memcmp(&in6, &last6, sizeof(in6)) > 0 &&
memcmp(&in6, &cur6, sizeof(in6)) < 0) {
cur6 = in6;
found++;
}
}
if (found)
SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6));
} while (found);
#endif /* INET6 */
/* convert ipad to opad */
for (i = 0; i < sizeof(sc->sc_pad); i++)
sc->sc_pad[i] ^= 0x36 ^ 0x5c;
}
2005-02-26 10:33:14 +00:00
static void
carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
unsigned char md[20])
{
SHA1_CTX sha1ctx;
CARP_LOCK_ASSERT(sc);
/* fetch first half of inner hash */
bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
SHA1Final(md, &sha1ctx);
/* outer hash */
SHA1Init(&sha1ctx);
SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
SHA1Update(&sha1ctx, md, 20);
SHA1Final(md, &sha1ctx);
}
2005-02-26 10:33:14 +00:00
static int
carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
unsigned char md[20])
{
unsigned char md2[20];
CARP_LOCK_ASSERT(sc);
carp_hmac_generate(sc, counter, md2);
return (bcmp(md, md2, sizeof(md2)));
}
/*
* process input packet.
* we have rearranged checks order compared to the rfc,
* but it seems more efficient this way or not possible otherwise.
*/
#ifdef INET
int
carp_input(struct mbuf **mp, int *offp, int proto)
{
struct mbuf *m = *mp;
struct ip *ip = mtod(m, struct ip *);
struct carp_header *ch;
int iplen, len;
iplen = *offp;
*mp = NULL;
CARPSTATS_INC(carps_ipackets);
if (!V_carp_allow) {
m_freem(m);
return (IPPROTO_DONE);
}
/* verify that the IP TTL is 255. */
if (ip->ip_ttl != CARP_DFLTTL) {
CARPSTATS_INC(carps_badttl);
CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
ip->ip_ttl,
m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
iplen = ip->ip_hl << 2;
if (m->m_pkthdr.len < iplen + sizeof(*ch)) {
CARPSTATS_INC(carps_badlen);
CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) "
"on %s\n", __func__, m->m_len - sizeof(struct ip),
m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
if (iplen + sizeof(*ch) < m->m_len) {
if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) {
CARPSTATS_INC(carps_hdrops);
CARP_DEBUG("%s: pullup failed\n", __func__);
return (IPPROTO_DONE);
}
ip = mtod(m, struct ip *);
}
ch = (struct carp_header *)((char *)ip + iplen);
/*
* verify that the received packet length is
* equal to the CARP header
*/
len = iplen + sizeof(*ch);
if (len > m->m_pkthdr.len) {
CARPSTATS_INC(carps_badlen);
CARP_DEBUG("%s: packet too short %d on %s\n", __func__,
m->m_pkthdr.len,
m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
if ((m = m_pullup(m, len)) == NULL) {
CARPSTATS_INC(carps_hdrops);
return (IPPROTO_DONE);
}
ip = mtod(m, struct ip *);
ch = (struct carp_header *)((char *)ip + iplen);
/* verify the CARP checksum */
m->m_data += iplen;
2012-12-25 14:29:38 +00:00
if (in_cksum(m, len - iplen)) {
CARPSTATS_INC(carps_badsum);
CARP_DEBUG("%s: checksum failed on %s\n", __func__,
m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
m->m_data -= iplen;
carp_input_c(m, ch, AF_INET);
return (IPPROTO_DONE);
}
#endif
#ifdef INET6
int
carp6_input(struct mbuf **mp, int *offp, int proto)
{
struct mbuf *m = *mp;
struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
struct carp_header *ch;
u_int len;
CARPSTATS_INC(carps_ipackets6);
if (!V_carp_allow) {
m_freem(m);
return (IPPROTO_DONE);
}
/* check if received on a valid carp interface */
if (m->m_pkthdr.rcvif->if_carp == NULL) {
CARPSTATS_INC(carps_badif);
CARP_DEBUG("%s: packet received on non-carp interface: %s\n",
__func__, m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
/* verify that the IP TTL is 255 */
if (ip6->ip6_hlim != CARP_DFLTTL) {
CARPSTATS_INC(carps_badttl);
CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
/* verify that we have a complete carp packet */
len = m->m_len;
IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch));
if (ch == NULL) {
CARPSTATS_INC(carps_badlen);
CARP_DEBUG("%s: packet size %u too small\n", __func__, len);
return (IPPROTO_DONE);
}
/* verify the CARP checksum */
m->m_data += *offp;
2012-12-25 14:29:38 +00:00
if (in_cksum(m, sizeof(*ch))) {
CARPSTATS_INC(carps_badsum);
CARP_DEBUG("%s: checksum failed, on %s\n", __func__,
m->m_pkthdr.rcvif->if_xname);
m_freem(m);
return (IPPROTO_DONE);
}
m->m_data -= *offp;
carp_input_c(m, ch, AF_INET6);
return (IPPROTO_DONE);
}
#endif /* INET6 */
/*
* This routine should not be necessary at all, but some switches
* (VMWare ESX vswitches) can echo our own packets back at us,
* and we must ignore them or they will cause us to drop out of
* MASTER mode.
*
* We cannot catch all cases of network loops. Instead, what we
* do here is catch any packet that arrives with a carp header
* with a VHID of 0, that comes from an address that is our own.
* These packets are by definition "from us" (even if they are from
* a misconfigured host that is pretending to be us).
*
* The VHID test is outside this mini-function.
*/
static int
carp_source_is_self(struct mbuf *m, struct ifaddr *ifa, sa_family_t af)
{
#ifdef INET
struct ip *ip4;
struct in_addr in4;
#endif
#ifdef INET6
struct ip6_hdr *ip6;
struct in6_addr in6;
#endif
switch (af) {
#ifdef INET
case AF_INET:
ip4 = mtod(m, struct ip *);
in4 = ifatoia(ifa)->ia_addr.sin_addr;
return (in4.s_addr == ip4->ip_src.s_addr);
#endif
#ifdef INET6
case AF_INET6:
ip6 = mtod(m, struct ip6_hdr *);
in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
return (memcmp(&in6, &ip6->ip6_src, sizeof(in6)) == 0);
#endif
default:
break;
}
return (0);
}
2005-02-26 10:33:14 +00:00
static void
carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af)
{
struct ifnet *ifp = m->m_pkthdr.rcvif;
struct ifaddr *ifa, *match;
struct carp_softc *sc;
uint64_t tmp_counter;
struct timeval sc_tv, ch_tv;
int error;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
/*
* Verify that the VHID is valid on the receiving interface.
*
* There should be just one match. If there are none
* the VHID is not valid and we drop the packet. If
* there are multiple VHID matches, take just the first
* one, for compatibility with previous code. While we're
* scanning, check for obvious loops in the network topology
* (these should never happen, and as noted above, we may
* miss real loops; this is just a double-check).
*/
error = 0;
match = NULL;
IFNET_FOREACH_IFA(ifp, ifa) {
if (match == NULL && ifa->ifa_carp != NULL &&
ifa->ifa_addr->sa_family == af &&
ifa->ifa_carp->sc_vhid == ch->carp_vhid)
match = ifa;
if (ch->carp_vhid == 0 && carp_source_is_self(m, ifa, af))
error = ELOOP;
}
ifa = error ? NULL : match;
if (ifa != NULL)
ifa_ref(ifa);
if (ifa == NULL) {
if (error == ELOOP) {
CARP_DEBUG("dropping looped packet on interface %s\n",
ifp->if_xname);
CARPSTATS_INC(carps_badif); /* ??? */
} else {
CARPSTATS_INC(carps_badvhid);
}
m_freem(m);
return;
}
/* verify the CARP version. */
if (ch->carp_version != CARP_VERSION) {
CARPSTATS_INC(carps_badver);
CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname,
ch->carp_version);
ifa_free(ifa);
m_freem(m);
return;
}
sc = ifa->ifa_carp;
CARP_LOCK(sc);
ifa_free(ifa);
if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
CARPSTATS_INC(carps_badauth);
CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__,
sc->sc_vhid, ifp->if_xname);
goto out;
}
tmp_counter = ntohl(ch->carp_counter[0]);
tmp_counter = tmp_counter<<32;
tmp_counter += ntohl(ch->carp_counter[1]);
/* XXX Replay protection goes here */
sc->sc_init_counter = 0;
sc->sc_counter = tmp_counter;
sc_tv.tv_sec = sc->sc_advbase;
sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256;
ch_tv.tv_sec = ch->carp_advbase;
ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
switch (sc->sc_state) {
case INIT:
break;
case MASTER:
/*
* If we receive an advertisement from a master who's going to
* be more frequent than us, go into BACKUP state.
*/
if (timevalcmp(&sc_tv, &ch_tv, >) ||
timevalcmp(&sc_tv, &ch_tv, ==)) {
callout_stop(&sc->sc_ad_tmo);
carp_set_state(sc, BACKUP,
"more frequent advertisement received");
carp_setrun(sc, 0);
carp_delroute(sc);
}
break;
case BACKUP:
/*
* If we're pre-empting masters who advertise slower than us,
* and this one claims to be slower, treat him as down.
*/
if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) {
carp_master_down_locked(sc,
"preempting a slower master");
break;
}
/*
* If the master is going to advertise at such a low frequency
* that he's guaranteed to time out, we'd might as well just
* treat him as timed out now.
*/
sc_tv.tv_sec = sc->sc_advbase * 3;
if (timevalcmp(&sc_tv, &ch_tv, <)) {
carp_master_down_locked(sc, "master will time out");
break;
}
/*
* Otherwise, we reset the counter and wait for the next
* advertisement.
*/
carp_setrun(sc, af);
break;
}
out:
CARP_UNLOCK(sc);
m_freem(m);
}
2005-02-26 10:33:14 +00:00
static int
carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch)
{
struct m_tag *mtag;
if (sc->sc_init_counter) {
/* this could also be seconds since unix epoch */
sc->sc_counter = arc4random();
sc->sc_counter = sc->sc_counter << 32;
sc->sc_counter += arc4random();
} else
sc->sc_counter++;
ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff);
ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff);
carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
/* Tag packet for carp_output */
if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *),
M_NOWAIT)) == NULL) {
m_freem(m);
CARPSTATS_INC(carps_onomem);
return (ENOMEM);
}
bcopy(&sc, mtag + 1, sizeof(sc));
m_tag_prepend(m, mtag);
return (0);
}
/*
* To avoid LORs and possible recursions this function shouldn't
* be called directly, but scheduled via taskqueue.
*/
2005-02-26 10:33:14 +00:00
static void
carp_send_ad_all(void *ctx __unused, int pending __unused)
{
struct carp_softc *sc;
mtx_lock(&carp_mtx);
LIST_FOREACH(sc, &carp_list, sc_next)
if (sc->sc_state == MASTER) {
CARP_LOCK(sc);
CURVNET_SET(sc->sc_carpdev->if_vnet);
carp_send_ad_locked(sc);
CURVNET_RESTORE();
CARP_UNLOCK(sc);
}
mtx_unlock(&carp_mtx);
}
/* Send a periodic advertisement, executed in callout context. */
2005-02-26 10:33:14 +00:00
static void
carp_send_ad(void *v)
{
struct carp_softc *sc = v;
CARP_LOCK_ASSERT(sc);
CURVNET_SET(sc->sc_carpdev->if_vnet);
carp_send_ad_locked(sc);
CURVNET_RESTORE();
CARP_UNLOCK(sc);
}
static void
carp_send_ad_error(struct carp_softc *sc, int error)
{
if (error) {
if (sc->sc_sendad_errors < INT_MAX)
sc->sc_sendad_errors++;
if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
static const char fmt[] = "send error %d on %s";
char msg[sizeof(fmt) + IFNAMSIZ];
sprintf(msg, fmt, error, sc->sc_carpdev->if_xname);
carp_demote_adj(V_carp_senderr_adj, msg);
}
sc->sc_sendad_success = 0;
} else {
if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS &&
++sc->sc_sendad_success >= CARP_SENDAD_MIN_SUCCESS) {
static const char fmt[] = "send ok on %s";
char msg[sizeof(fmt) + IFNAMSIZ];
sprintf(msg, fmt, sc->sc_carpdev->if_xname);
carp_demote_adj(-V_carp_senderr_adj, msg);
sc->sc_sendad_errors = 0;
} else
sc->sc_sendad_errors = 0;
}
}
/*
* Pick the best ifaddr on the given ifp for sending CARP
* advertisements.
*
* "Best" here is defined by ifa_preferred(). This function is much
* much like ifaof_ifpforaddr() except that we just use ifa_preferred().
*
* (This could be simplified to return the actual address, except that
* it has a different format in AF_INET and AF_INET6.)
*/
static struct ifaddr *
carp_best_ifa(int af, struct ifnet *ifp)
{
struct ifaddr *ifa, *best;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
if (af >= AF_MAX)
return (NULL);
best = NULL;
ifnet: Replace if_addr_lock rwlock with epoch + mutex Run on LLNW canaries and tested by pho@ gallatin: Using a 14-core, 28-HTT single socket E5-2697 v3 with a 40GbE MLX5 based ConnectX 4-LX NIC, I see an almost 12% improvement in received packet rate, and a larger improvement in bytes delivered all the way to userspace. When the host receiving 64 streams of netperf -H $DUT -t UDP_STREAM -- -m 1, I see, using nstat -I mce0 1 before the patch: InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 4.98 0.00 4.42 0.00 4235592 33 83.80 4720653 2149771 1235 247.32 4.73 0.00 4.20 0.00 4025260 33 82.99 4724900 2139833 1204 247.32 4.72 0.00 4.20 0.00 4035252 33 82.14 4719162 2132023 1264 247.32 4.71 0.00 4.21 0.00 4073206 33 83.68 4744973 2123317 1347 247.32 4.72 0.00 4.21 0.00 4061118 33 80.82 4713615 2188091 1490 247.32 4.72 0.00 4.21 0.00 4051675 33 85.29 4727399 2109011 1205 247.32 4.73 0.00 4.21 0.00 4039056 33 84.65 4724735 2102603 1053 247.32 After the patch InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 5.43 0.00 4.20 0.00 3313143 33 84.96 5434214 1900162 2656 245.51 5.43 0.00 4.20 0.00 3308527 33 85.24 5439695 1809382 2521 245.51 5.42 0.00 4.19 0.00 3316778 33 87.54 5416028 1805835 2256 245.51 5.42 0.00 4.19 0.00 3317673 33 90.44 5426044 1763056 2332 245.51 5.42 0.00 4.19 0.00 3314839 33 88.11 5435732 1792218 2499 245.52 5.44 0.00 4.19 0.00 3293228 33 91.84 5426301 1668597 2121 245.52 Similarly, netperf reports 230Mb/s before the patch, and 270Mb/s after the patch Reviewed by: gallatin Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D15366
2018-05-18 20:13:34 +00:00
CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
if (ifa->ifa_addr->sa_family == af &&
(best == NULL || ifa_preferred(best, ifa)))
best = ifa;
}
if (best != NULL)
ifa_ref(best);
return (best);
}
static void
carp_send_ad_locked(struct carp_softc *sc)
{
struct carp_header ch;
struct timeval tv;
struct ifaddr *ifa;
struct carp_header *ch_ptr;
struct mbuf *m;
int len, advskew;
CARP_LOCK_ASSERT(sc);
advskew = DEMOTE_ADVSKEW(sc);
tv.tv_sec = sc->sc_advbase;
tv.tv_usec = advskew * 1000000 / 256;
ch.carp_version = CARP_VERSION;
ch.carp_type = CARP_ADVERTISEMENT;
ch.carp_vhid = sc->sc_vhid;
ch.carp_advbase = sc->sc_advbase;
ch.carp_advskew = advskew;
ch.carp_authlen = 7; /* XXX DEFINE */
ch.carp_pad1 = 0; /* must be zero */
ch.carp_cksum = 0;
/* XXXGL: OpenBSD picks first ifaddr with needed family. */
#ifdef INET
if (sc->sc_naddrs) {
struct ip *ip;
m = m_gethdr(M_NOWAIT, MT_DATA);
if (m == NULL) {
CARPSTATS_INC(carps_onomem);
goto resched;
}
len = sizeof(*ip) + sizeof(ch);
m->m_pkthdr.len = len;
m->m_pkthdr.rcvif = NULL;
m->m_len = len;
M_ALIGN(m, m->m_len);
m->m_flags |= M_MCAST;
ip = mtod(m, struct ip *);
ip->ip_v = IPVERSION;
ip->ip_hl = sizeof(*ip) >> 2;
ip->ip_tos = V_carp_dscp << IPTOS_DSCP_OFFSET;
ip->ip_len = htons(len);
ip->ip_off = htons(IP_DF);
ip->ip_ttl = CARP_DFLTTL;
ip->ip_p = IPPROTO_CARP;
ip->ip_sum = 0;
ip_fillid(ip);
ifa = carp_best_ifa(AF_INET, sc->sc_carpdev);
if (ifa != NULL) {
ip->ip_src.s_addr =
ifatoia(ifa)->ia_addr.sin_addr.s_addr;
ifa_free(ifa);
} else
ip->ip_src.s_addr = 0;
ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
ch_ptr = (struct carp_header *)(&ip[1]);
bcopy(&ch, ch_ptr, sizeof(ch));
if (carp_prepare_ad(m, sc, ch_ptr))
goto resched;
m->m_data += sizeof(*ip);
2012-12-25 14:29:38 +00:00
ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip));
m->m_data -= sizeof(*ip);
CARPSTATS_INC(carps_opackets);
carp_send_ad_error(sc, ip_output(m, NULL, NULL, IP_RAWOUTPUT,
&sc->sc_carpdev->if_carp->cif_imo, NULL));
}
#endif /* INET */
#ifdef INET6
if (sc->sc_naddrs6) {
struct ip6_hdr *ip6;
m = m_gethdr(M_NOWAIT, MT_DATA);
if (m == NULL) {
CARPSTATS_INC(carps_onomem);
goto resched;
}
len = sizeof(*ip6) + sizeof(ch);
m->m_pkthdr.len = len;
m->m_pkthdr.rcvif = NULL;
m->m_len = len;
M_ALIGN(m, m->m_len);
m->m_flags |= M_MCAST;
ip6 = mtod(m, struct ip6_hdr *);
bzero(ip6, sizeof(*ip6));
ip6->ip6_vfc |= IPV6_VERSION;
/* Traffic class isn't defined in ip6 struct instead
* it gets offset into flowid field */
ip6->ip6_flow |= htonl(V_carp_dscp << (IPV6_FLOWLABEL_LEN +
IPTOS_DSCP_OFFSET));
ip6->ip6_hlim = CARP_DFLTTL;
ip6->ip6_nxt = IPPROTO_CARP;
/* set the source address */
ifa = carp_best_ifa(AF_INET6, sc->sc_carpdev);
if (ifa != NULL) {
bcopy(IFA_IN6(ifa), &ip6->ip6_src,
sizeof(struct in6_addr));
ifa_free(ifa);
} else
/* This should never happen with IPv6. */
bzero(&ip6->ip6_src, sizeof(struct in6_addr));
/* Set the multicast destination. */
ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
ip6->ip6_dst.s6_addr8[15] = 0x12;
if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
m_freem(m);
CARP_DEBUG("%s: in6_setscope failed\n", __func__);
goto resched;
}
ch_ptr = (struct carp_header *)(&ip6[1]);
bcopy(&ch, ch_ptr, sizeof(ch));
if (carp_prepare_ad(m, sc, ch_ptr))
goto resched;
m->m_data += sizeof(*ip6);
2012-12-25 14:29:38 +00:00
ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6));
m->m_data -= sizeof(*ip6);
CARPSTATS_INC(carps_opackets6);
carp_send_ad_error(sc, ip6_output(m, NULL, NULL, 0,
&sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL));
}
#endif /* INET6 */
resched:
callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc);
}
static void
carp_addroute(struct carp_softc *sc)
{
struct ifaddr *ifa;
CARP_FOREACH_IFA(sc, ifa)
carp_ifa_addroute(ifa);
}
static void
carp_ifa_addroute(struct ifaddr *ifa)
{
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
in_addprefix(ifatoia(ifa), RTF_UP);
ifa_add_loopback_route(ifa,
(struct sockaddr *)&ifatoia(ifa)->ia_addr);
break;
#endif
#ifdef INET6
case AF_INET6:
ifa_add_loopback_route(ifa,
(struct sockaddr *)&ifatoia6(ifa)->ia_addr);
nd6_add_ifa_lle(ifatoia6(ifa));
break;
#endif
}
}
static void
carp_delroute(struct carp_softc *sc)
{
struct ifaddr *ifa;
CARP_FOREACH_IFA(sc, ifa)
carp_ifa_delroute(ifa);
}
static void
carp_ifa_delroute(struct ifaddr *ifa)
{
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
ifa_del_loopback_route(ifa,
(struct sockaddr *)&ifatoia(ifa)->ia_addr);
in_scrubprefix(ifatoia(ifa), LLE_STATIC);
break;
#endif
#ifdef INET6
case AF_INET6:
ifa_del_loopback_route(ifa,
(struct sockaddr *)&ifatoia6(ifa)->ia_addr);
nd6_rem_ifa_lle(ifatoia6(ifa), 1);
break;
#endif
}
}
int
carp_master(struct ifaddr *ifa)
{
struct carp_softc *sc = ifa->ifa_carp;
return (sc->sc_state == MASTER);
}
#ifdef INET
/*
* Broadcast a gratuitous ARP request containing
* the virtual router MAC address for each IP address
* associated with the virtual router.
*/
2005-02-26 10:33:14 +00:00
static void
carp_send_arp(struct carp_softc *sc)
{
struct ifaddr *ifa;
struct in_addr addr;
CARP_FOREACH_IFA(sc, ifa) {
if (ifa->ifa_addr->sa_family != AF_INET)
continue;
addr = ((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
arp_announce_ifaddr(sc->sc_carpdev, addr, LLADDR(&sc->sc_addr));
}
}
int
carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr)
{
struct carp_softc *sc = ifa->ifa_carp;
if (sc->sc_state == MASTER) {
*enaddr = LLADDR(&sc->sc_addr);
return (1);
}
return (0);
}
#endif
#ifdef INET6
2005-02-26 10:33:14 +00:00
static void
carp_send_na(struct carp_softc *sc)
{
static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
struct ifaddr *ifa;
struct in6_addr *in6;
CARP_FOREACH_IFA(sc, ifa) {
if (ifa->ifa_addr->sa_family != AF_INET6)
continue;
in6 = IFA_IN6(ifa);
nd6_na_output(sc->sc_carpdev, &mcast, in6,
ND_NA_FLAG_OVERRIDE, 1, NULL);
DELAY(1000); /* XXX */
}
}
/*
* Returns ifa in case it's a carp address and it is MASTER, or if the address
* matches and is not a carp address. Returns NULL otherwise.
*/
struct ifaddr *
carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr)
{
struct ifaddr *ifa;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
ifa = NULL;
ifnet: Replace if_addr_lock rwlock with epoch + mutex Run on LLNW canaries and tested by pho@ gallatin: Using a 14-core, 28-HTT single socket E5-2697 v3 with a 40GbE MLX5 based ConnectX 4-LX NIC, I see an almost 12% improvement in received packet rate, and a larger improvement in bytes delivered all the way to userspace. When the host receiving 64 streams of netperf -H $DUT -t UDP_STREAM -- -m 1, I see, using nstat -I mce0 1 before the patch: InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 4.98 0.00 4.42 0.00 4235592 33 83.80 4720653 2149771 1235 247.32 4.73 0.00 4.20 0.00 4025260 33 82.99 4724900 2139833 1204 247.32 4.72 0.00 4.20 0.00 4035252 33 82.14 4719162 2132023 1264 247.32 4.71 0.00 4.21 0.00 4073206 33 83.68 4744973 2123317 1347 247.32 4.72 0.00 4.21 0.00 4061118 33 80.82 4713615 2188091 1490 247.32 4.72 0.00 4.21 0.00 4051675 33 85.29 4727399 2109011 1205 247.32 4.73 0.00 4.21 0.00 4039056 33 84.65 4724735 2102603 1053 247.32 After the patch InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 5.43 0.00 4.20 0.00 3313143 33 84.96 5434214 1900162 2656 245.51 5.43 0.00 4.20 0.00 3308527 33 85.24 5439695 1809382 2521 245.51 5.42 0.00 4.19 0.00 3316778 33 87.54 5416028 1805835 2256 245.51 5.42 0.00 4.19 0.00 3317673 33 90.44 5426044 1763056 2332 245.51 5.42 0.00 4.19 0.00 3314839 33 88.11 5435732 1792218 2499 245.52 5.44 0.00 4.19 0.00 3293228 33 91.84 5426301 1668597 2121 245.52 Similarly, netperf reports 230Mb/s before the patch, and 270Mb/s after the patch Reviewed by: gallatin Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D15366
2018-05-18 20:13:34 +00:00
CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
if (ifa->ifa_addr->sa_family != AF_INET6)
continue;
if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa)))
continue;
if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER)
ifa = NULL;
else
ifa_ref(ifa);
break;
}
return (ifa);
}
caddr_t
carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr)
{
struct ifaddr *ifa;
Widen NET_EPOCH coverage. When epoch(9) was introduced to network stack, it was basically dropped in place of existing locking, which was mutexes and rwlocks. For the sake of performance mutex covered areas were as small as possible, so became epoch covered areas. However, epoch doesn't introduce any contention, it just delays memory reclaim. So, there is no point to minimise epoch covered areas in sense of performance. Meanwhile entering/exiting epoch also has non-zero CPU usage, so doing this less often is a win. Not the least is also code maintainability. In the new paradigm we can assume that at any stage of processing a packet, we are inside network epoch. This makes coding both input and output path way easier. On output path we already enter epoch quite early - in the ip_output(), in the ip6_output(). This patch does the same for the input path. All ISR processing, network related callouts, other ways of packet injection to the network stack shall be performed in net_epoch. Any leaf function that walks network configuration now asserts epoch. Tricky part is configuration code paths - ioctls, sysctls. They also call into leaf functions, so some need to be changed. This patch would introduce more epoch recursions (see EPOCH_TRACE) than we had before. They will be cleaned up separately, as several of them aren't trivial. Note, that unlike a lock recursion the epoch recursion is safe and just wastes a bit of resources. Reviewed by: gallatin, hselasky, cy, adrian, kristof Differential Revision: https://reviews.freebsd.org/D19111
2019-10-07 22:40:05 +00:00
NET_EPOCH_ASSERT();
IFNET_FOREACH_IFA(ifp, ifa)
if (ifa->ifa_addr->sa_family == AF_INET6 &&
IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) {
struct carp_softc *sc = ifa->ifa_carp;
struct m_tag *mtag;
mtag = m_tag_get(PACKET_TAG_CARP,
sizeof(struct carp_softc *), M_NOWAIT);
if (mtag == NULL)
/* Better a bit than nothing. */
return (LLADDR(&sc->sc_addr));
bcopy(&sc, mtag + 1, sizeof(sc));
m_tag_prepend(m, mtag);
return (LLADDR(&sc->sc_addr));
}
return (NULL);
}
#endif /* INET6 */
int
carp_forus(struct ifnet *ifp, u_char *dhost)
{
struct carp_softc *sc;
uint8_t *ena = dhost;
if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1)
return (0);
CIF_LOCK(ifp->if_carp);
IFNET_FOREACH_CARP(ifp, sc) {
CARP_LOCK(sc);
if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr),
ETHER_ADDR_LEN)) {
CARP_UNLOCK(sc);
CIF_UNLOCK(ifp->if_carp);
return (1);
}
CARP_UNLOCK(sc);
}
CIF_UNLOCK(ifp->if_carp);
return (0);
}
/* Master down timeout event, executed in callout context. */
2005-02-26 10:33:14 +00:00
static void
carp_master_down(void *v)
{
struct carp_softc *sc = v;
CARP_LOCK_ASSERT(sc);
CURVNET_SET(sc->sc_carpdev->if_vnet);
if (sc->sc_state == BACKUP) {
carp_master_down_locked(sc, "master timed out");
}
CURVNET_RESTORE();
CARP_UNLOCK(sc);
}
static void
carp_master_down_locked(struct carp_softc *sc, const char *reason)
{
CARP_LOCK_ASSERT(sc);
switch (sc->sc_state) {
case BACKUP:
carp_set_state(sc, MASTER, reason);
carp_send_ad_locked(sc);
#ifdef INET
carp_send_arp(sc);
#endif
#ifdef INET6
carp_send_na(sc);
#endif
carp_setrun(sc, 0);
carp_addroute(sc);
break;
case INIT:
case MASTER:
#ifdef INVARIANTS
panic("carp: VHID %u@%s: master_down event in %s state\n",
sc->sc_vhid,
sc->sc_carpdev->if_xname,
sc->sc_state ? "MASTER" : "INIT");
#endif
break;
}
}
/*
* When in backup state, af indicates whether to reset the master down timer
* for v4 or v6. If it's set to zero, reset the ones which are already pending.
*/
2005-02-26 10:33:14 +00:00
static void
carp_setrun(struct carp_softc *sc, sa_family_t af)
{
struct timeval tv;
CARP_LOCK_ASSERT(sc);
if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 ||
sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
(sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) ||
!V_carp_allow)
return;
switch (sc->sc_state) {
case INIT:
carp_set_state(sc, BACKUP, "initialization complete");
carp_setrun(sc, 0);
break;
case BACKUP:
callout_stop(&sc->sc_ad_tmo);
tv.tv_sec = 3 * sc->sc_advbase;
tv.tv_usec = sc->sc_advskew * 1000000 / 256;
switch (af) {
#ifdef INET
case AF_INET:
callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
carp_master_down, sc);
break;
#endif
#ifdef INET6
case AF_INET6:
callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
carp_master_down, sc);
break;
#endif
default:
#ifdef INET
if (sc->sc_naddrs)
callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
carp_master_down, sc);
#endif
#ifdef INET6
if (sc->sc_naddrs6)
callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
carp_master_down, sc);
#endif
break;
}
break;
case MASTER:
tv.tv_sec = sc->sc_advbase;
tv.tv_usec = sc->sc_advskew * 1000000 / 256;
callout_reset(&sc->sc_ad_tmo, tvtohz(&tv),
carp_send_ad, sc);
break;
}
}
/*
* Setup multicast structures.
*/
static int
carp_multicast_setup(struct carp_if *cif, sa_family_t sa)
{
struct ifnet *ifp = cif->cif_ifp;
int error = 0;
switch (sa) {
#ifdef INET
case AF_INET:
{
struct ip_moptions *imo = &cif->cif_imo;
struct in_mfilter *imf;
struct in_addr addr;
if (ip_mfilter_first(&imo->imo_head) != NULL)
return (0);
imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
ip_mfilter_init(&imo->imo_head);
imo->imo_multicast_vif = -1;
addr.s_addr = htonl(INADDR_CARP_GROUP);
if ((error = in_joingroup(ifp, &addr, NULL,
&imf->imf_inm)) != 0) {
ip_mfilter_free(imf);
break;
}
ip_mfilter_insert(&imo->imo_head, imf);
imo->imo_multicast_ifp = ifp;
imo->imo_multicast_ttl = CARP_DFLTTL;
imo->imo_multicast_loop = 0;
break;
}
#endif
#ifdef INET6
case AF_INET6:
{
struct ip6_moptions *im6o = &cif->cif_im6o;
struct in6_mfilter *im6f[2];
struct in6_addr in6;
if (ip6_mfilter_first(&im6o->im6o_head))
return (0);
im6f[0] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
im6f[1] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
ip6_mfilter_init(&im6o->im6o_head);
im6o->im6o_multicast_hlim = CARP_DFLTTL;
im6o->im6o_multicast_ifp = ifp;
/* Join IPv6 CARP multicast group. */
bzero(&in6, sizeof(in6));
in6.s6_addr16[0] = htons(0xff02);
in6.s6_addr8[15] = 0x12;
if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
ip6_mfilter_free(im6f[0]);
ip6_mfilter_free(im6f[1]);
break;
}
if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[0]->im6f_in6m, 0)) != 0) {
ip6_mfilter_free(im6f[0]);
ip6_mfilter_free(im6f[1]);
break;
}
/* Join solicited multicast address. */
bzero(&in6, sizeof(in6));
in6.s6_addr16[0] = htons(0xff02);
in6.s6_addr32[1] = 0;
in6.s6_addr32[2] = htonl(1);
in6.s6_addr32[3] = 0;
in6.s6_addr8[12] = 0xff;
if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
ip6_mfilter_free(im6f[0]);
ip6_mfilter_free(im6f[1]);
break;
}
if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[1]->im6f_in6m, 0)) != 0) {
in6_leavegroup(im6f[0]->im6f_in6m, NULL);
ip6_mfilter_free(im6f[0]);
ip6_mfilter_free(im6f[1]);
break;
}
ip6_mfilter_insert(&im6o->im6o_head, im6f[0]);
ip6_mfilter_insert(&im6o->im6o_head, im6f[1]);
break;
}
#endif
}
return (error);
}
/*
* Free multicast structures.
*/
static void
carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa)
{
#ifdef INET
struct ip_moptions *imo = &cif->cif_imo;
struct in_mfilter *imf;
#endif
#ifdef INET6
struct ip6_moptions *im6o = &cif->cif_im6o;
struct in6_mfilter *im6f;
#endif
sx_assert(&carp_sx, SA_XLOCKED);
switch (sa) {
#ifdef INET
case AF_INET:
if (cif->cif_naddrs != 0)
break;
while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
ip_mfilter_remove(&imo->imo_head, imf);
in_leavegroup(imf->imf_inm, NULL);
ip_mfilter_free(imf);
}
break;
#endif
#ifdef INET6
case AF_INET6:
if (cif->cif_naddrs6 != 0)
break;
while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
ip6_mfilter_remove(&im6o->im6o_head, im6f);
in6_leavegroup(im6f->im6f_in6m, NULL);
ip6_mfilter_free(im6f);
Bite the bullet, and make the IPv6 SSM and MLDv2 mega-commit: import from p4 bms_netdev. Summary of changes: * Connect netinet6/in6_mcast.c to build. The legacy KAME KPIs are mostly preserved. * Eliminate now dead code from ip6_output.c. Don't do mbuf bingo, we are not going to do RFC 2292 style CMSG tricks for multicast options as they are not required by any current IPv6 normative reference. * Refactor transports (UDP, raw_ip6) to do own mcast filtering. SCTP, TCP unaffected by this change. * Add ip6_msource, in6_msource structs to in6_var.h. * Hookup mld_ifinfo state to in6_ifextra, allocate from domifattach path. * Eliminate IN6_LOOKUP_MULTI(), it is no longer referenced. Kernel consumers which need this should use in6m_lookup(). * Refactor IPv6 socket group memberships to use a vector (like IPv4). * Update ifmcstat(8) for IPv6 SSM. * Add witness lock order for IN6_MULTI_LOCK. * Move IN6_MULTI_LOCK out of lower ip6_output()/ip6_input() paths. * Introduce IP6STAT_ADD/SUB/INC/DEC as per rwatson's IPv4 cleanup. * Update carp(4) for new IPv6 SSM KPIs. * Virtualize ip6_mrouter socket. Changes mostly localized to IPv6 MROUTING. * Don't do a local group lookup in MROUTING. * Kill unused KAME prototypes in6_purgemkludge(), in6_restoremkludge(). * Preserve KAME DAD timer jitter behaviour in MLDv1 compatibility mode. * Bump __FreeBSD_version to 800084. * Update UPDATING. NOTE WELL: * This code hasn't been tested against real MLDv2 queriers (yet), although the on-wire protocol has been verified in Wireshark. * There are a few unresolved issues in the socket layer APIs to do with scope ID propagation. * There is a LOR present in ip6_output()'s use of in6_setscope() which needs to be resolved. See comments in mld6.c. This is believed to be benign and can't be avoided for the moment without re-introducing an indirect netisr. This work was mostly derived from the IGMPv3 implementation, and has been sponsored by a third party.
2009-04-29 19:19:13 +00:00
}
break;
#endif
}
}
int
carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa)
{
struct m_tag *mtag;
struct carp_softc *sc;
if (!sa)
return (0);
switch (sa->sa_family) {
#ifdef INET
case AF_INET:
break;
#endif
#ifdef INET6
case AF_INET6:
break;
#endif
default:
return (0);
}
mtag = m_tag_find(m, PACKET_TAG_CARP, NULL);
if (mtag == NULL)
return (0);
bcopy(mtag + 1, &sc, sizeof(sc));
/* Set the source MAC address to the Virtual Router MAC Address. */
switch (ifp->if_type) {
case IFT_ETHER:
case IFT_BRIDGE:
case IFT_L2VLAN: {
struct ether_header *eh;
eh = mtod(m, struct ether_header *);
eh->ether_shost[0] = 0;
eh->ether_shost[1] = 0;
eh->ether_shost[2] = 0x5e;
eh->ether_shost[3] = 0;
eh->ether_shost[4] = 1;
eh->ether_shost[5] = sc->sc_vhid;
}
break;
default:
printf("%s: carp is not supported for the %d interface type\n",
ifp->if_xname, ifp->if_type);
return (EOPNOTSUPP);
}
return (0);
}
static struct carp_softc*
carp_alloc(struct ifnet *ifp)
{
struct carp_softc *sc;
struct carp_if *cif;
sx_assert(&carp_sx, SA_XLOCKED);
if ((cif = ifp->if_carp) == NULL)
cif = carp_alloc_if(ifp);
sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO);
sc->sc_advbase = CARP_DFLTINTV;
sc->sc_vhid = -1; /* required setting */
sc->sc_init_counter = 1;
sc->sc_state = INIT;
sc->sc_ifasiz = sizeof(struct ifaddr *);
sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO);
sc->sc_carpdev = ifp;
CARP_LOCK_INIT(sc);
#ifdef INET
callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
#endif
#ifdef INET6
callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
#endif
callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
CIF_LOCK(cif);
TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list);
CIF_UNLOCK(cif);
mtx_lock(&carp_mtx);
LIST_INSERT_HEAD(&carp_list, sc, sc_next);
mtx_unlock(&carp_mtx);
return (sc);
}
static void
carp_grow_ifas(struct carp_softc *sc)
{
struct ifaddr **new;
new = malloc(sc->sc_ifasiz * 2, M_CARP, M_WAITOK | M_ZERO);
CARP_LOCK(sc);
bcopy(sc->sc_ifas, new, sc->sc_ifasiz);
free(sc->sc_ifas, M_CARP);
sc->sc_ifas = new;
sc->sc_ifasiz *= 2;
CARP_UNLOCK(sc);
}
static void
carp_destroy(struct carp_softc *sc)
{
struct ifnet *ifp = sc->sc_carpdev;
struct carp_if *cif = ifp->if_carp;
sx_assert(&carp_sx, SA_XLOCKED);
if (sc->sc_suppress)
carp_demote_adj(-V_carp_ifdown_adj, "vhid removed");
CARP_UNLOCK(sc);
CIF_LOCK(cif);
TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list);
CIF_UNLOCK(cif);
mtx_lock(&carp_mtx);
LIST_REMOVE(sc, sc_next);
mtx_unlock(&carp_mtx);
callout_drain(&sc->sc_ad_tmo);
#ifdef INET
callout_drain(&sc->sc_md_tmo);
#endif
#ifdef INET6
callout_drain(&sc->sc_md6_tmo);
#endif
CARP_LOCK_DESTROY(sc);
Bite the bullet, and make the IPv6 SSM and MLDv2 mega-commit: import from p4 bms_netdev. Summary of changes: * Connect netinet6/in6_mcast.c to build. The legacy KAME KPIs are mostly preserved. * Eliminate now dead code from ip6_output.c. Don't do mbuf bingo, we are not going to do RFC 2292 style CMSG tricks for multicast options as they are not required by any current IPv6 normative reference. * Refactor transports (UDP, raw_ip6) to do own mcast filtering. SCTP, TCP unaffected by this change. * Add ip6_msource, in6_msource structs to in6_var.h. * Hookup mld_ifinfo state to in6_ifextra, allocate from domifattach path. * Eliminate IN6_LOOKUP_MULTI(), it is no longer referenced. Kernel consumers which need this should use in6m_lookup(). * Refactor IPv6 socket group memberships to use a vector (like IPv4). * Update ifmcstat(8) for IPv6 SSM. * Add witness lock order for IN6_MULTI_LOCK. * Move IN6_MULTI_LOCK out of lower ip6_output()/ip6_input() paths. * Introduce IP6STAT_ADD/SUB/INC/DEC as per rwatson's IPv4 cleanup. * Update carp(4) for new IPv6 SSM KPIs. * Virtualize ip6_mrouter socket. Changes mostly localized to IPv6 MROUTING. * Don't do a local group lookup in MROUTING. * Kill unused KAME prototypes in6_purgemkludge(), in6_restoremkludge(). * Preserve KAME DAD timer jitter behaviour in MLDv1 compatibility mode. * Bump __FreeBSD_version to 800084. * Update UPDATING. NOTE WELL: * This code hasn't been tested against real MLDv2 queriers (yet), although the on-wire protocol has been verified in Wireshark. * There are a few unresolved issues in the socket layer APIs to do with scope ID propagation. * There is a LOR present in ip6_output()'s use of in6_setscope() which needs to be resolved. See comments in mld6.c. This is believed to be benign and can't be avoided for the moment without re-introducing an indirect netisr. This work was mostly derived from the IGMPv3 implementation, and has been sponsored by a third party.
2009-04-29 19:19:13 +00:00
free(sc->sc_ifas, M_CARP);
free(sc, M_CARP);
}
static struct carp_if*
carp_alloc_if(struct ifnet *ifp)
{
struct carp_if *cif;
int error;
cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO);
if ((error = ifpromisc(ifp, 1)) != 0)
printf("%s: ifpromisc(%s) failed: %d\n",
__func__, ifp->if_xname, error);
else
cif->cif_flags |= CIF_PROMISC;
CIF_LOCK_INIT(cif);
cif->cif_ifp = ifp;
TAILQ_INIT(&cif->cif_vrs);
IF_ADDR_WLOCK(ifp);
ifp->if_carp = cif;
if_ref(ifp);
IF_ADDR_WUNLOCK(ifp);
return (cif);
}
static void
carp_free_if(struct carp_if *cif)
{
struct ifnet *ifp = cif->cif_ifp;
CIF_LOCK_ASSERT(cif);
KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty",
__func__));
IF_ADDR_WLOCK(ifp);
ifp->if_carp = NULL;
IF_ADDR_WUNLOCK(ifp);
CIF_LOCK_DESTROY(cif);
if (cif->cif_flags & CIF_PROMISC)
ifpromisc(ifp, 0);
if_rele(ifp);
free(cif, M_CARP);
}
static void
carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv)
{
CARP_LOCK(sc);
carpr->carpr_state = sc->sc_state;
carpr->carpr_vhid = sc->sc_vhid;
carpr->carpr_advbase = sc->sc_advbase;
carpr->carpr_advskew = sc->sc_advskew;
if (priv)
bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key));
else
bzero(carpr->carpr_key, sizeof(carpr->carpr_key));
CARP_UNLOCK(sc);
}
int
carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td)
{
struct carpreq carpr;
struct ifnet *ifp;
struct carp_softc *sc = NULL;
int error = 0, locked = 0;
if ((error = copyin(ifr_data_get_ptr(ifr), &carpr, sizeof carpr)))
return (error);
ifp = ifunit_ref(ifr->ifr_name);
if (ifp == NULL)
return (ENXIO);
switch (ifp->if_type) {
case IFT_ETHER:
case IFT_L2VLAN:
case IFT_BRIDGE:
break;
default:
error = EOPNOTSUPP;
goto out;
}
if ((ifp->if_flags & IFF_MULTICAST) == 0) {
error = EADDRNOTAVAIL;
goto out;
}
sx_xlock(&carp_sx);
switch (cmd) {
case SIOCSVH:
if ((error = priv_check(td, PRIV_NETINET_CARP)))
break;
if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID ||
carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) {
error = EINVAL;
break;
}
if (ifp->if_carp) {
IFNET_FOREACH_CARP(ifp, sc)
if (sc->sc_vhid == carpr.carpr_vhid)
break;
}
if (sc == NULL) {
sc = carp_alloc(ifp);
CARP_LOCK(sc);
sc->sc_vhid = carpr.carpr_vhid;
LLADDR(&sc->sc_addr)[0] = 0;
LLADDR(&sc->sc_addr)[1] = 0;
LLADDR(&sc->sc_addr)[2] = 0x5e;
LLADDR(&sc->sc_addr)[3] = 0;
LLADDR(&sc->sc_addr)[4] = 1;
LLADDR(&sc->sc_addr)[5] = sc->sc_vhid;
} else
CARP_LOCK(sc);
locked = 1;
if (carpr.carpr_advbase > 0) {
if (carpr.carpr_advbase > 255 ||
carpr.carpr_advbase < CARP_DFLTINTV) {
error = EINVAL;
break;
}
sc->sc_advbase = carpr.carpr_advbase;
}
if (carpr.carpr_advskew >= 255) {
error = EINVAL;
break;
}
sc->sc_advskew = carpr.carpr_advskew;
if (carpr.carpr_key[0] != '\0') {
bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
carp_hmac_prepare(sc);
}
if (sc->sc_state != INIT &&
carpr.carpr_state != sc->sc_state) {
switch (carpr.carpr_state) {
case BACKUP:
callout_stop(&sc->sc_ad_tmo);
carp_set_state(sc, BACKUP,
"user requested via ifconfig");
carp_setrun(sc, 0);
carp_delroute(sc);
break;
case MASTER:
carp_master_down_locked(sc,
"user requested via ifconfig");
break;
default:
break;
}
}
break;
case SIOCGVH:
{
int priveleged;
if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) {
error = EINVAL;
break;
}
if (carpr.carpr_count < 1) {
error = EMSGSIZE;
break;
}
if (ifp->if_carp == NULL) {
error = ENOENT;
break;
}
priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0);
if (carpr.carpr_vhid != 0) {
IFNET_FOREACH_CARP(ifp, sc)
if (sc->sc_vhid == carpr.carpr_vhid)
break;
if (sc == NULL) {
error = ENOENT;
break;
}
carp_carprcp(&carpr, sc, priveleged);
error = copyout(&carpr, ifr_data_get_ptr(ifr),
sizeof(carpr));
} else {
int i, count;
count = 0;
IFNET_FOREACH_CARP(ifp, sc)
count++;
if (count > carpr.carpr_count) {
CIF_UNLOCK(ifp->if_carp);
error = EMSGSIZE;
break;
}
i = 0;
IFNET_FOREACH_CARP(ifp, sc) {
carp_carprcp(&carpr, sc, priveleged);
carpr.carpr_count = count;
error = copyout(&carpr,
(caddr_t)ifr_data_get_ptr(ifr) +
(i * sizeof(carpr)), sizeof(carpr));
if (error) {
CIF_UNLOCK(ifp->if_carp);
break;
}
i++;
}
}
break;
}
default:
error = EINVAL;
}
sx_xunlock(&carp_sx);
out:
if (locked)
CARP_UNLOCK(sc);
if_rele(ifp);
return (error);
}
static int
carp_get_vhid(struct ifaddr *ifa)
{
if (ifa == NULL || ifa->ifa_carp == NULL)
return (0);
return (ifa->ifa_carp->sc_vhid);
}
int
carp_attach(struct ifaddr *ifa, int vhid)
{
struct ifnet *ifp = ifa->ifa_ifp;
struct carp_if *cif = ifp->if_carp;
struct carp_softc *sc;
int index, error;
KASSERT(ifa->ifa_carp == NULL, ("%s: ifa %p attached", __func__, ifa));
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
#endif
#ifdef INET6
case AF_INET6:
#endif
break;
default:
return (EPROTOTYPE);
}
sx_xlock(&carp_sx);
if (ifp->if_carp == NULL) {
sx_xunlock(&carp_sx);
return (ENOPROTOOPT);
}
IFNET_FOREACH_CARP(ifp, sc)
if (sc->sc_vhid == vhid)
break;
if (sc == NULL) {
sx_xunlock(&carp_sx);
return (ENOENT);
}
error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family);
if (error) {
CIF_FREE(cif);
sx_xunlock(&carp_sx);
return (error);
}
index = sc->sc_naddrs + sc->sc_naddrs6 + 1;
if (index > sc->sc_ifasiz / sizeof(struct ifaddr *))
carp_grow_ifas(sc);
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
cif->cif_naddrs++;
sc->sc_naddrs++;
break;
#endif
#ifdef INET6
case AF_INET6:
cif->cif_naddrs6++;
sc->sc_naddrs6++;
break;
#endif
}
ifa_ref(ifa);
CARP_LOCK(sc);
sc->sc_ifas[index - 1] = ifa;
ifa->ifa_carp = sc;
carp_hmac_prepare(sc);
carp_sc_state(sc);
CARP_UNLOCK(sc);
sx_xunlock(&carp_sx);
return (0);
}
void
carp_detach(struct ifaddr *ifa, bool keep_cif)
{
struct ifnet *ifp = ifa->ifa_ifp;
struct carp_if *cif = ifp->if_carp;
struct carp_softc *sc = ifa->ifa_carp;
int i, index;
KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa));
sx_xlock(&carp_sx);
CARP_LOCK(sc);
/* Shift array. */
index = sc->sc_naddrs + sc->sc_naddrs6;
for (i = 0; i < index; i++)
if (sc->sc_ifas[i] == ifa)
break;
KASSERT(i < index, ("%s: %p no backref", __func__, ifa));
for (; i < index - 1; i++)
sc->sc_ifas[i] = sc->sc_ifas[i+1];
sc->sc_ifas[index - 1] = NULL;
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
cif->cif_naddrs--;
sc->sc_naddrs--;
break;
#endif
#ifdef INET6
case AF_INET6:
cif->cif_naddrs6--;
sc->sc_naddrs6--;
break;
#endif
}
carp_ifa_delroute(ifa);
carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family);
ifa->ifa_carp = NULL;
ifa_free(ifa);
carp_hmac_prepare(sc);
carp_sc_state(sc);
if (!keep_cif && sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0)
carp_destroy(sc);
else
CARP_UNLOCK(sc);
if (!keep_cif)
CIF_FREE(cif);
sx_xunlock(&carp_sx);
}
static void
carp_set_state(struct carp_softc *sc, int state, const char *reason)
{
CARP_LOCK_ASSERT(sc);
if (sc->sc_state != state) {
const char *carp_states[] = { CARP_STATES };
char subsys[IFNAMSIZ+5];
snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid,
sc->sc_carpdev->if_xname);
CARP_LOG("%s: %s -> %s (%s)\n", subsys,
carp_states[sc->sc_state], carp_states[state], reason);
sc->sc_state = state;
devctl_notify("CARP", subsys, carp_states[state], NULL);
}
}
static void
carp_linkstate(struct ifnet *ifp)
{
struct carp_softc *sc;
CIF_LOCK(ifp->if_carp);
IFNET_FOREACH_CARP(ifp, sc) {
CARP_LOCK(sc);
carp_sc_state(sc);
CARP_UNLOCK(sc);
}
CIF_UNLOCK(ifp->if_carp);
}
static void
carp_sc_state(struct carp_softc *sc)
{
CARP_LOCK_ASSERT(sc);
if (sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
!(sc->sc_carpdev->if_flags & IFF_UP) ||
!V_carp_allow) {
callout_stop(&sc->sc_ad_tmo);
#ifdef INET
callout_stop(&sc->sc_md_tmo);
#endif
#ifdef INET6
callout_stop(&sc->sc_md6_tmo);
#endif
carp_set_state(sc, INIT, "hardware interface down");
carp_setrun(sc, 0);
if (!sc->sc_suppress)
carp_demote_adj(V_carp_ifdown_adj, "interface down");
sc->sc_suppress = 1;
} else {
carp_set_state(sc, INIT, "hardware interface up");
carp_setrun(sc, 0);
if (sc->sc_suppress)
carp_demote_adj(-V_carp_ifdown_adj, "interface up");
sc->sc_suppress = 0;
}
}
static void
carp_demote_adj(int adj, char *reason)
{
atomic_add_int(&V_carp_demotion, adj);
CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason);
taskqueue_enqueue(taskqueue_swi, &carp_sendall_task);
}
static int
carp_allow_sysctl(SYSCTL_HANDLER_ARGS)
{
int new, error;
struct carp_softc *sc;
new = V_carp_allow;
error = sysctl_handle_int(oidp, &new, 0, req);
if (error || !req->newptr)
return (error);
if (V_carp_allow != new) {
V_carp_allow = new;
mtx_lock(&carp_mtx);
LIST_FOREACH(sc, &carp_list, sc_next) {
CARP_LOCK(sc);
if (curvnet == sc->sc_carpdev->if_vnet)
carp_sc_state(sc);
CARP_UNLOCK(sc);
}
mtx_unlock(&carp_mtx);
}
return (0);
}
static int
carp_dscp_sysctl(SYSCTL_HANDLER_ARGS)
{
int new, error;
new = V_carp_dscp;
error = sysctl_handle_int(oidp, &new, 0, req);
if (error || !req->newptr)
return (error);
if (new < 0 || new > 63)
return (EINVAL);
V_carp_dscp = new;
return (0);
}
static int
carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS)
{
int new, error;
new = V_carp_demotion;
error = sysctl_handle_int(oidp, &new, 0, req);
if (error || !req->newptr)
return (error);
carp_demote_adj(new, "sysctl");
return (0);
}
#ifdef INET
extern struct domain inetdomain;
static struct protosw in_carp_protosw = {
.pr_type = SOCK_RAW,
.pr_domain = &inetdomain,
.pr_protocol = IPPROTO_CARP,
.pr_flags = PR_ATOMIC|PR_ADDR,
.pr_input = carp_input,
.pr_output = rip_output,
.pr_ctloutput = rip_ctloutput,
.pr_usrreqs = &rip_usrreqs
};
#endif
#ifdef INET6
extern struct domain inet6domain;
static struct protosw in6_carp_protosw = {
.pr_type = SOCK_RAW,
.pr_domain = &inet6domain,
.pr_protocol = IPPROTO_CARP,
.pr_flags = PR_ATOMIC|PR_ADDR,
.pr_input = carp6_input,
.pr_output = rip6_output,
.pr_ctloutput = rip6_ctloutput,
.pr_usrreqs = &rip6_usrreqs
};
#endif
static void
carp_mod_cleanup(void)
{
#ifdef INET
if (proto_reg[CARP_INET] == 0) {
(void)ipproto_unregister(IPPROTO_CARP);
pf_proto_unregister(PF_INET, IPPROTO_CARP, SOCK_RAW);
proto_reg[CARP_INET] = -1;
}
carp_iamatch_p = NULL;
#endif
#ifdef INET6
if (proto_reg[CARP_INET6] == 0) {
(void)ip6proto_unregister(IPPROTO_CARP);
pf_proto_unregister(PF_INET6, IPPROTO_CARP, SOCK_RAW);
proto_reg[CARP_INET6] = -1;
}
carp_iamatch6_p = NULL;
carp_macmatch6_p = NULL;
#endif
carp_ioctl_p = NULL;
carp_attach_p = NULL;
carp_detach_p = NULL;
carp_get_vhid_p = NULL;
carp_linkstate_p = NULL;
carp_forus_p = NULL;
carp_output_p = NULL;
carp_demote_adj_p = NULL;
carp_master_p = NULL;
mtx_unlock(&carp_mtx);
taskqueue_drain(taskqueue_swi, &carp_sendall_task);
mtx_destroy(&carp_mtx);
sx_destroy(&carp_sx);
}
static int
carp_mod_load(void)
{
int err;
mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF);
sx_init(&carp_sx, "carp_sx");
LIST_INIT(&carp_list);
carp_get_vhid_p = carp_get_vhid;
carp_forus_p = carp_forus;
carp_output_p = carp_output;
carp_linkstate_p = carp_linkstate;
carp_ioctl_p = carp_ioctl;
carp_attach_p = carp_attach;
carp_detach_p = carp_detach;
carp_demote_adj_p = carp_demote_adj;
carp_master_p = carp_master;
#ifdef INET6
carp_iamatch6_p = carp_iamatch6;
carp_macmatch6_p = carp_macmatch6;
proto_reg[CARP_INET6] = pf_proto_register(PF_INET6,
(struct protosw *)&in6_carp_protosw);
if (proto_reg[CARP_INET6]) {
printf("carp: error %d attaching to PF_INET6\n",
proto_reg[CARP_INET6]);
carp_mod_cleanup();
return (proto_reg[CARP_INET6]);
}
err = ip6proto_register(IPPROTO_CARP);
if (err) {
printf("carp: error %d registering with INET6\n", err);
carp_mod_cleanup();
return (err);
}
#endif
#ifdef INET
carp_iamatch_p = carp_iamatch;
proto_reg[CARP_INET] = pf_proto_register(PF_INET, &in_carp_protosw);
if (proto_reg[CARP_INET]) {
printf("carp: error %d attaching to PF_INET\n",
proto_reg[CARP_INET]);
carp_mod_cleanup();
return (proto_reg[CARP_INET]);
}
err = ipproto_register(IPPROTO_CARP);
if (err) {
printf("carp: error %d registering with INET\n", err);
carp_mod_cleanup();
return (err);
}
#endif
return (0);
}
static int
carp_modevent(module_t mod, int type, void *data)
{
switch (type) {
case MOD_LOAD:
return carp_mod_load();
/* NOTREACHED */
case MOD_UNLOAD:
mtx_lock(&carp_mtx);
if (LIST_EMPTY(&carp_list))
carp_mod_cleanup();
else {
mtx_unlock(&carp_mtx);
return (EBUSY);
}
break;
default:
return (EINVAL);
}
return (0);
}
static moduledata_t carp_mod = {
"carp",
carp_modevent,
0
};
DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);