The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/net/ieee8023ad_lacp.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*      $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $     */
    2 
    3 /*-
    4  * Copyright (c)2005 YAMAMOTO Takashi,
    5  * Copyright (c)2008 Andrew Thompson <thompsa@FreeBSD.org>
    6  * All rights reserved.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  */
   29 
   30 #include <sys/cdefs.h>
   31 __FBSDID("$FreeBSD: releng/10.1/sys/net/ieee8023ad_lacp.c 260179 2014-01-02 01:51:54Z scottl $");
   32 
   33 #include <sys/param.h>
   34 #include <sys/callout.h>
   35 #include <sys/mbuf.h>
   36 #include <sys/systm.h>
   37 #include <sys/malloc.h>
   38 #include <sys/kernel.h> /* hz */
   39 #include <sys/socket.h> /* for net/if.h */
   40 #include <sys/sockio.h>
   41 #include <sys/sysctl.h>
   42 #include <machine/stdarg.h>
   43 #include <sys/lock.h>
   44 #include <sys/rwlock.h>
   45 
   46 #include <net/if.h>
   47 #include <net/if_dl.h>
   48 #include <net/ethernet.h>
   49 #include <net/if_media.h>
   50 #include <net/if_types.h>
   51 
   52 #include <net/if_lagg.h>
   53 #include <net/ieee8023ad_lacp.h>
   54 
   55 /*
   56  * actor system priority and port priority.
   57  * XXX should be configurable.
   58  */
   59 
   60 #define LACP_SYSTEM_PRIO        0x8000
   61 #define LACP_PORT_PRIO          0x8000
   62 
   63 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
   64     { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
   65 
   66 static const struct tlv_template lacp_info_tlv_template[] = {
   67         { LACP_TYPE_ACTORINFO,
   68             sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
   69         { LACP_TYPE_PARTNERINFO,
   70             sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) },
   71         { LACP_TYPE_COLLECTORINFO,
   72             sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) },
   73         { 0, 0 },
   74 };
   75 
   76 static const struct tlv_template marker_info_tlv_template[] = {
   77         { MARKER_TYPE_INFO,
   78             sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
   79         { 0, 0 },
   80 };
   81 
   82 static const struct tlv_template marker_response_tlv_template[] = {
   83         { MARKER_TYPE_RESPONSE,
   84             sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) },
   85         { 0, 0 },
   86 };
   87 
   88 typedef void (*lacp_timer_func_t)(struct lacp_port *);
   89 
   90 static void     lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
   91 static void     lacp_fill_markerinfo(struct lacp_port *,
   92                     struct lacp_markerinfo *);
   93 
   94 static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *);
   95 static void     lacp_suppress_distributing(struct lacp_softc *,
   96                     struct lacp_aggregator *);
   97 static void     lacp_transit_expire(void *);
   98 static void     lacp_update_portmap(struct lacp_softc *);
   99 static void     lacp_select_active_aggregator(struct lacp_softc *);
  100 static uint16_t lacp_compose_key(struct lacp_port *);
  101 static int      tlv_check(const void *, size_t, const struct tlvhdr *,
  102                     const struct tlv_template *, boolean_t);
  103 static void     lacp_tick(void *);
  104 
  105 static void     lacp_fill_aggregator_id(struct lacp_aggregator *,
  106                     const struct lacp_port *);
  107 static void     lacp_fill_aggregator_id_peer(struct lacp_peerinfo *,
  108                     const struct lacp_peerinfo *);
  109 static int      lacp_aggregator_is_compatible(const struct lacp_aggregator *,
  110                     const struct lacp_port *);
  111 static int      lacp_peerinfo_is_compatible(const struct lacp_peerinfo *,
  112                     const struct lacp_peerinfo *);
  113 
  114 static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *,
  115                     struct lacp_port *);
  116 static void     lacp_aggregator_addref(struct lacp_softc *,
  117                     struct lacp_aggregator *);
  118 static void     lacp_aggregator_delref(struct lacp_softc *,
  119                     struct lacp_aggregator *);
  120 
  121 /* receive machine */
  122 
  123 static int      lacp_pdu_input(struct lacp_port *, struct mbuf *);
  124 static int      lacp_marker_input(struct lacp_port *, struct mbuf *);
  125 static void     lacp_sm_rx(struct lacp_port *, const struct lacpdu *);
  126 static void     lacp_sm_rx_timer(struct lacp_port *);
  127 static void     lacp_sm_rx_set_expired(struct lacp_port *);
  128 static void     lacp_sm_rx_update_ntt(struct lacp_port *,
  129                     const struct lacpdu *);
  130 static void     lacp_sm_rx_record_pdu(struct lacp_port *,
  131                     const struct lacpdu *);
  132 static void     lacp_sm_rx_update_selected(struct lacp_port *,
  133                     const struct lacpdu *);
  134 static void     lacp_sm_rx_record_default(struct lacp_port *);
  135 static void     lacp_sm_rx_update_default_selected(struct lacp_port *);
  136 static void     lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *,
  137                     const struct lacp_peerinfo *);
  138 
  139 /* mux machine */
  140 
  141 static void     lacp_sm_mux(struct lacp_port *);
  142 static void     lacp_set_mux(struct lacp_port *, enum lacp_mux_state);
  143 static void     lacp_sm_mux_timer(struct lacp_port *);
  144 
  145 /* periodic transmit machine */
  146 
  147 static void     lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t);
  148 static void     lacp_sm_ptx_tx_schedule(struct lacp_port *);
  149 static void     lacp_sm_ptx_timer(struct lacp_port *);
  150 
  151 /* transmit machine */
  152 
  153 static void     lacp_sm_tx(struct lacp_port *);
  154 static void     lacp_sm_assert_ntt(struct lacp_port *);
  155 
  156 static void     lacp_run_timers(struct lacp_port *);
  157 static int      lacp_compare_peerinfo(const struct lacp_peerinfo *,
  158                     const struct lacp_peerinfo *);
  159 static int      lacp_compare_systemid(const struct lacp_systemid *,
  160                     const struct lacp_systemid *);
  161 static void     lacp_port_enable(struct lacp_port *);
  162 static void     lacp_port_disable(struct lacp_port *);
  163 static void     lacp_select(struct lacp_port *);
  164 static void     lacp_unselect(struct lacp_port *);
  165 static void     lacp_disable_collecting(struct lacp_port *);
  166 static void     lacp_enable_collecting(struct lacp_port *);
  167 static void     lacp_disable_distributing(struct lacp_port *);
  168 static void     lacp_enable_distributing(struct lacp_port *);
  169 static int      lacp_xmit_lacpdu(struct lacp_port *);
  170 static int      lacp_xmit_marker(struct lacp_port *);
  171 
  172 /* Debugging */
  173 
  174 static void     lacp_dump_lacpdu(const struct lacpdu *);
  175 static const char *lacp_format_partner(const struct lacp_peerinfo *, char *,
  176                     size_t);
  177 static const char *lacp_format_lagid(const struct lacp_peerinfo *,
  178                     const struct lacp_peerinfo *, char *, size_t);
  179 static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *,
  180                     char *, size_t);
  181 static const char *lacp_format_state(uint8_t, char *, size_t);
  182 static const char *lacp_format_mac(const uint8_t *, char *, size_t);
  183 static const char *lacp_format_systemid(const struct lacp_systemid *, char *,
  184                     size_t);
  185 static const char *lacp_format_portid(const struct lacp_portid *, char *,
  186                     size_t);
  187 static void     lacp_dprintf(const struct lacp_port *, const char *, ...)
  188                     __attribute__((__format__(__printf__, 2, 3)));
  189 
  190 static int lacp_debug = 0;
  191 SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad");
  192 SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_TUN,
  193     &lacp_debug, 0, "Enable LACP debug logging (1=debug, 2=trace)");
  194 TUNABLE_INT("net.link.lagg.lacp.debug", &lacp_debug);
  195 
  196 #define LACP_DPRINTF(a) if (lacp_debug & 0x01) { lacp_dprintf a ; }
  197 #define LACP_TRACE(a) if (lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
  198 #define LACP_TPRINTF(a) if (lacp_debug & 0x04) { lacp_dprintf a ; }
  199 
  200 /*
  201  * partner administration variables.
  202  * XXX should be configurable.
  203  */
  204 
  205 static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
  206         .lip_systemid = { .lsi_prio = 0xffff },
  207         .lip_portid = { .lpi_prio = 0xffff },
  208         .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
  209             LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
  210 };
  211 
  212 static const struct lacp_peerinfo lacp_partner_admin_strict = {
  213         .lip_systemid = { .lsi_prio = 0xffff },
  214         .lip_portid = { .lpi_prio = 0xffff },
  215         .lip_state = 0,
  216 };
  217 
  218 static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
  219         [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
  220         [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
  221         [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
  222 };
  223 
  224 struct mbuf *
  225 lacp_input(struct lagg_port *lgp, struct mbuf *m)
  226 {
  227         struct lacp_port *lp = LACP_PORT(lgp);
  228         uint8_t subtype;
  229 
  230         if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) {
  231                 m_freem(m);
  232                 return (NULL);
  233         }
  234 
  235         m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype);
  236         switch (subtype) {
  237                 case SLOWPROTOCOLS_SUBTYPE_LACP:
  238                         lacp_pdu_input(lp, m);
  239                         return (NULL);
  240 
  241                 case SLOWPROTOCOLS_SUBTYPE_MARKER:
  242                         lacp_marker_input(lp, m);
  243                         return (NULL);
  244         }
  245 
  246         /* Not a subtype we are interested in */
  247         return (m);
  248 }
  249 
  250 /*
  251  * lacp_pdu_input: process lacpdu
  252  */
  253 static int
  254 lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
  255 {
  256         struct lacp_softc *lsc = lp->lp_lsc;
  257         struct lacpdu *du;
  258         int error = 0;
  259 
  260         if (m->m_pkthdr.len != sizeof(*du)) {
  261                 goto bad;
  262         }
  263 
  264         if ((m->m_flags & M_MCAST) == 0) {
  265                 goto bad;
  266         }
  267 
  268         if (m->m_len < sizeof(*du)) {
  269                 m = m_pullup(m, sizeof(*du));
  270                 if (m == NULL) {
  271                         return (ENOMEM);
  272                 }
  273         }
  274 
  275         du = mtod(m, struct lacpdu *);
  276 
  277         if (memcmp(&du->ldu_eh.ether_dhost,
  278             &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
  279                 goto bad;
  280         }
  281 
  282         /*
  283          * ignore the version for compatibility with
  284          * the future protocol revisions.
  285          */
  286 #if 0
  287         if (du->ldu_sph.sph_version != 1) {
  288                 goto bad;
  289         }
  290 #endif
  291 
  292         /*
  293          * ignore tlv types for compatibility with
  294          * the future protocol revisions.
  295          */
  296         if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor,
  297             lacp_info_tlv_template, FALSE)) {
  298                 goto bad;
  299         }
  300 
  301         if (lacp_debug > 0) {
  302                 lacp_dprintf(lp, "lacpdu receive\n");
  303                 lacp_dump_lacpdu(du);
  304         }
  305 
  306         if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) {
  307                 LACP_TPRINTF((lp, "Dropping RX PDU\n"));
  308                 goto bad;
  309         }
  310 
  311         LACP_LOCK(lsc);
  312         lacp_sm_rx(lp, du);
  313         LACP_UNLOCK(lsc);
  314 
  315         m_freem(m);
  316         return (error);
  317 
  318 bad:
  319         m_freem(m);
  320         return (EINVAL);
  321 }
  322 
  323 static void
  324 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
  325 {
  326         struct lagg_port *lgp = lp->lp_lagg;
  327         struct lagg_softc *sc = lgp->lp_softc;
  328 
  329         info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
  330         memcpy(&info->lip_systemid.lsi_mac,
  331             IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
  332         info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
  333         info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
  334         info->lip_state = lp->lp_state;
  335 }
  336 
  337 static void
  338 lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
  339 {
  340         struct ifnet *ifp = lp->lp_ifp;
  341 
  342         /* Fill in the port index and system id (encoded as the MAC) */
  343         info->mi_rq_port = htons(ifp->if_index);
  344         memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN);
  345         info->mi_rq_xid = htonl(0);
  346 }
  347 
  348 static int
  349 lacp_xmit_lacpdu(struct lacp_port *lp)
  350 {
  351         struct lagg_port *lgp = lp->lp_lagg;
  352         struct mbuf *m;
  353         struct lacpdu *du;
  354         int error;
  355 
  356         LACP_LOCK_ASSERT(lp->lp_lsc);
  357 
  358         m = m_gethdr(M_NOWAIT, MT_DATA);
  359         if (m == NULL) {
  360                 return (ENOMEM);
  361         }
  362         m->m_len = m->m_pkthdr.len = sizeof(*du);
  363 
  364         du = mtod(m, struct lacpdu *);
  365         memset(du, 0, sizeof(*du));
  366 
  367         memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
  368             ETHER_ADDR_LEN);
  369         memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
  370         du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW);
  371 
  372         du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP;
  373         du->ldu_sph.sph_version = 1;
  374 
  375         TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor));
  376         du->ldu_actor = lp->lp_actor;
  377 
  378         TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO,
  379             sizeof(du->ldu_partner));
  380         du->ldu_partner = lp->lp_partner;
  381 
  382         TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO,
  383             sizeof(du->ldu_collector));
  384         du->ldu_collector.lci_maxdelay = 0;
  385 
  386         if (lacp_debug > 0) {
  387                 lacp_dprintf(lp, "lacpdu transmit\n");
  388                 lacp_dump_lacpdu(du);
  389         }
  390 
  391         m->m_flags |= M_MCAST;
  392 
  393         /*
  394          * XXX should use higher priority queue.
  395          * otherwise network congestion can break aggregation.
  396          */
  397 
  398         error = lagg_enqueue(lp->lp_ifp, m);
  399         return (error);
  400 }
  401 
  402 static int
  403 lacp_xmit_marker(struct lacp_port *lp)
  404 {
  405         struct lagg_port *lgp = lp->lp_lagg;
  406         struct mbuf *m;
  407         struct markerdu *mdu;
  408         int error;
  409 
  410         LACP_LOCK_ASSERT(lp->lp_lsc);
  411 
  412         m = m_gethdr(M_NOWAIT, MT_DATA);
  413         if (m == NULL) {
  414                 return (ENOMEM);
  415         }
  416         m->m_len = m->m_pkthdr.len = sizeof(*mdu);
  417 
  418         mdu = mtod(m, struct markerdu *);
  419         memset(mdu, 0, sizeof(*mdu));
  420 
  421         memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols,
  422             ETHER_ADDR_LEN);
  423         memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN);
  424         mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW);
  425 
  426         mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER;
  427         mdu->mdu_sph.sph_version = 1;
  428 
  429         /* Bump the transaction id and copy over the marker info */
  430         lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1);
  431         TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info));
  432         mdu->mdu_info = lp->lp_marker;
  433 
  434         LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n",
  435             ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":",
  436             ntohl(mdu->mdu_info.mi_rq_xid)));
  437 
  438         m->m_flags |= M_MCAST;
  439         error = lagg_enqueue(lp->lp_ifp, m);
  440         return (error);
  441 }
  442 
  443 void
  444 lacp_linkstate(struct lagg_port *lgp)
  445 {
  446         struct lacp_port *lp = LACP_PORT(lgp);
  447         struct lacp_softc *lsc = lp->lp_lsc;
  448         struct ifnet *ifp = lgp->lp_ifp;
  449         struct ifmediareq ifmr;
  450         int error = 0;
  451         u_int media;
  452         uint8_t old_state;
  453         uint16_t old_key;
  454 
  455         bzero((char *)&ifmr, sizeof(ifmr));
  456         error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr);
  457         if (error != 0)
  458                 return;
  459 
  460         LACP_LOCK(lsc);
  461         media = ifmr.ifm_active;
  462         LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, "
  463             "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER,
  464             (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP));
  465         old_state = lp->lp_state;
  466         old_key = lp->lp_key;
  467 
  468         lp->lp_media = media;
  469         /*
  470          * If the port is not an active full duplex Ethernet link then it can
  471          * not be aggregated.
  472          */
  473         if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 ||
  474             ifp->if_link_state != LINK_STATE_UP) {
  475                 lacp_port_disable(lp);
  476         } else {
  477                 lacp_port_enable(lp);
  478         }
  479         lp->lp_key = lacp_compose_key(lp);
  480 
  481         if (old_state != lp->lp_state || old_key != lp->lp_key) {
  482                 LACP_DPRINTF((lp, "-> UNSELECTED\n"));
  483                 lp->lp_selected = LACP_UNSELECTED;
  484         }
  485         LACP_UNLOCK(lsc);
  486 }
  487 
  488 static void
  489 lacp_tick(void *arg)
  490 {
  491         struct lacp_softc *lsc = arg;
  492         struct lacp_port *lp;
  493 
  494         LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
  495                 if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0)
  496                         continue;
  497 
  498                 lacp_run_timers(lp);
  499 
  500                 lacp_select(lp);
  501                 lacp_sm_mux(lp);
  502                 lacp_sm_tx(lp);
  503                 lacp_sm_ptx_tx_schedule(lp);
  504         }
  505         callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
  506 }
  507 
  508 int
  509 lacp_port_create(struct lagg_port *lgp)
  510 {
  511         struct lagg_softc *sc = lgp->lp_softc;
  512         struct lacp_softc *lsc = LACP_SOFTC(sc);
  513         struct lacp_port *lp;
  514         struct ifnet *ifp = lgp->lp_ifp;
  515         struct sockaddr_dl sdl;
  516         struct ifmultiaddr *rifma = NULL;
  517         int error;
  518 
  519         boolean_t active = TRUE; /* XXX should be configurable */
  520         boolean_t fast = FALSE; /* XXX should be configurable */
  521 
  522         bzero((char *)&sdl, sizeof(sdl));
  523         sdl.sdl_len = sizeof(sdl);
  524         sdl.sdl_family = AF_LINK;
  525         sdl.sdl_index = ifp->if_index;
  526         sdl.sdl_type = IFT_ETHER;
  527         sdl.sdl_alen = ETHER_ADDR_LEN;
  528 
  529         bcopy(&ethermulticastaddr_slowprotocols,
  530             LLADDR(&sdl), ETHER_ADDR_LEN);
  531         error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma);
  532         if (error) {
  533                 printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifname);
  534                 return (error);
  535         }
  536 
  537         lp = malloc(sizeof(struct lacp_port),
  538             M_DEVBUF, M_NOWAIT|M_ZERO);
  539         if (lp == NULL)
  540                 return (ENOMEM);
  541 
  542         LACP_LOCK(lsc);
  543         lgp->lp_psc = (caddr_t)lp;
  544         lp->lp_ifp = ifp;
  545         lp->lp_lagg = lgp;
  546         lp->lp_lsc = lsc;
  547         lp->lp_ifma = rifma;
  548 
  549         LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next);
  550 
  551         lacp_fill_actorinfo(lp, &lp->lp_actor);
  552         lacp_fill_markerinfo(lp, &lp->lp_marker);
  553         lp->lp_state =
  554             (active ? LACP_STATE_ACTIVITY : 0) |
  555             (fast ? LACP_STATE_TIMEOUT : 0);
  556         lp->lp_aggregator = NULL;
  557         lacp_sm_rx_set_expired(lp);
  558         LACP_UNLOCK(lsc);
  559         lacp_linkstate(lgp);
  560 
  561         return (0);
  562 }
  563 
  564 void
  565 lacp_port_destroy(struct lagg_port *lgp)
  566 {
  567         struct lacp_port *lp = LACP_PORT(lgp);
  568         struct lacp_softc *lsc = lp->lp_lsc;
  569         int i;
  570 
  571         LACP_LOCK(lsc);
  572         for (i = 0; i < LACP_NTIMER; i++) {
  573                 LACP_TIMER_DISARM(lp, i);
  574         }
  575 
  576         lacp_disable_collecting(lp);
  577         lacp_disable_distributing(lp);
  578         lacp_unselect(lp);
  579 
  580         /* The address may have already been removed by if_purgemaddrs() */
  581         if (!lgp->lp_detaching)
  582                 if_delmulti_ifma(lp->lp_ifma);
  583 
  584         LIST_REMOVE(lp, lp_next);
  585         LACP_UNLOCK(lsc);
  586         free(lp, M_DEVBUF);
  587 }
  588 
  589 void
  590 lacp_req(struct lagg_softc *sc, caddr_t data)
  591 {
  592         struct lacp_opreq *req = (struct lacp_opreq *)data;
  593         struct lacp_softc *lsc = LACP_SOFTC(sc);
  594         struct lacp_aggregator *la = lsc->lsc_active_aggregator;
  595 
  596         LACP_LOCK(lsc);
  597         bzero(req, sizeof(struct lacp_opreq));
  598         if (la != NULL) {
  599                 req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio);
  600                 memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac,
  601                     ETHER_ADDR_LEN);
  602                 req->actor_key = ntohs(la->la_actor.lip_key);
  603                 req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio);
  604                 req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno);
  605                 req->actor_state = la->la_actor.lip_state;
  606 
  607                 req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio);
  608                 memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac,
  609                     ETHER_ADDR_LEN);
  610                 req->partner_key = ntohs(la->la_partner.lip_key);
  611                 req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio);
  612                 req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno);
  613                 req->partner_state = la->la_partner.lip_state;
  614         }
  615         LACP_UNLOCK(lsc);
  616 }
  617 
  618 void
  619 lacp_portreq(struct lagg_port *lgp, caddr_t data)
  620 {
  621         struct lacp_opreq *req = (struct lacp_opreq *)data;
  622         struct lacp_port *lp = LACP_PORT(lgp);
  623         struct lacp_softc *lsc = lp->lp_lsc;
  624 
  625         LACP_LOCK(lsc);
  626         req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio);
  627         memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac,
  628             ETHER_ADDR_LEN);
  629         req->actor_key = ntohs(lp->lp_actor.lip_key);
  630         req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio);
  631         req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno);
  632         req->actor_state = lp->lp_actor.lip_state;
  633 
  634         req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio);
  635         memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac,
  636             ETHER_ADDR_LEN);
  637         req->partner_key = ntohs(lp->lp_partner.lip_key);
  638         req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio);
  639         req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno);
  640         req->partner_state = lp->lp_partner.lip_state;
  641         LACP_UNLOCK(lsc);
  642 }
  643 
  644 static void
  645 lacp_disable_collecting(struct lacp_port *lp)
  646 {
  647         LACP_DPRINTF((lp, "collecting disabled\n"));
  648         lp->lp_state &= ~LACP_STATE_COLLECTING;
  649 }
  650 
  651 static void
  652 lacp_enable_collecting(struct lacp_port *lp)
  653 {
  654         LACP_DPRINTF((lp, "collecting enabled\n"));
  655         lp->lp_state |= LACP_STATE_COLLECTING;
  656 }
  657 
  658 static void
  659 lacp_disable_distributing(struct lacp_port *lp)
  660 {
  661         struct lacp_aggregator *la = lp->lp_aggregator;
  662         struct lacp_softc *lsc = lp->lp_lsc;
  663         struct lagg_softc *sc = lsc->lsc_softc;
  664         char buf[LACP_LAGIDSTR_MAX+1];
  665 
  666         LACP_LOCK_ASSERT(lsc);
  667 
  668         if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) {
  669                 return;
  670         }
  671 
  672         KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports"));
  673         KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports));
  674         KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid"));
  675 
  676         LACP_DPRINTF((lp, "disable distributing on aggregator %s, "
  677             "nports %d -> %d\n",
  678             lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
  679             la->la_nports, la->la_nports - 1));
  680 
  681         TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
  682         la->la_nports--;
  683         sc->sc_active = la->la_nports;
  684 
  685         if (lsc->lsc_active_aggregator == la) {
  686                 lacp_suppress_distributing(lsc, la);
  687                 lacp_select_active_aggregator(lsc);
  688                 /* regenerate the port map, the active aggregator has changed */
  689                 lacp_update_portmap(lsc);
  690         }
  691 
  692         lp->lp_state &= ~LACP_STATE_DISTRIBUTING;
  693 }
  694 
  695 static void
  696 lacp_enable_distributing(struct lacp_port *lp)
  697 {
  698         struct lacp_aggregator *la = lp->lp_aggregator;
  699         struct lacp_softc *lsc = lp->lp_lsc;
  700         struct lagg_softc *sc = lsc->lsc_softc;
  701         char buf[LACP_LAGIDSTR_MAX+1];
  702 
  703         LACP_LOCK_ASSERT(lsc);
  704 
  705         if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) {
  706                 return;
  707         }
  708 
  709         LACP_DPRINTF((lp, "enable distributing on aggregator %s, "
  710             "nports %d -> %d\n",
  711             lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
  712             la->la_nports, la->la_nports + 1));
  713 
  714         KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
  715         TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
  716         la->la_nports++;
  717         sc->sc_active = la->la_nports;
  718 
  719         lp->lp_state |= LACP_STATE_DISTRIBUTING;
  720 
  721         if (lsc->lsc_active_aggregator == la) {
  722                 lacp_suppress_distributing(lsc, la);
  723                 lacp_update_portmap(lsc);
  724         } else
  725                 /* try to become the active aggregator */
  726                 lacp_select_active_aggregator(lsc);
  727 }
  728 
  729 static void
  730 lacp_transit_expire(void *vp)
  731 {
  732         struct lacp_softc *lsc = vp;
  733 
  734         LACP_LOCK_ASSERT(lsc);
  735 
  736         LACP_TRACE(NULL);
  737 
  738         lsc->lsc_suppress_distributing = FALSE;
  739 }
  740 
  741 static void
  742 lacp_attach_sysctl(struct lacp_softc *lsc, struct sysctl_oid *p_oid)
  743 {
  744         struct lagg_softc *sc = lsc->lsc_softc;
  745 
  746         SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(p_oid), OID_AUTO,
  747             "lacp_strict_mode",
  748             CTLFLAG_RW,
  749             &lsc->lsc_strict_mode,
  750             lsc->lsc_strict_mode,
  751             "Enable LACP strict mode");
  752 }
  753 
  754 static void
  755 lacp_attach_sysctl_debug(struct lacp_softc *lsc, struct sysctl_oid *p_oid)
  756 {
  757         struct lagg_softc *sc = lsc->lsc_softc;
  758         struct sysctl_oid *oid;
  759 
  760         /* Create a child of the parent lagg interface */
  761         oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(p_oid),
  762             OID_AUTO, "debug", CTLFLAG_RD, NULL, "DEBUG");
  763 
  764         SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
  765             "rx_test",
  766             CTLFLAG_RW,
  767             &lsc->lsc_debug.lsc_rx_test,
  768             lsc->lsc_debug.lsc_rx_test,
  769             "Bitmap of if_dunit entries to drop RX frames for");
  770         SYSCTL_ADD_UINT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
  771             "tx_test",
  772             CTLFLAG_RW,
  773             &lsc->lsc_debug.lsc_tx_test,
  774             lsc->lsc_debug.lsc_tx_test,
  775             "Bitmap of if_dunit entries to drop TX frames for");
  776 }
  777 
  778 int
  779 lacp_attach(struct lagg_softc *sc)
  780 {
  781         struct lacp_softc *lsc;
  782         struct sysctl_oid *oid;
  783 
  784         lsc = malloc(sizeof(struct lacp_softc),
  785             M_DEVBUF, M_NOWAIT|M_ZERO);
  786         if (lsc == NULL)
  787                 return (ENOMEM);
  788 
  789         sc->sc_psc = (caddr_t)lsc;
  790         lsc->lsc_softc = sc;
  791 
  792         lsc->lsc_hashkey = arc4random();
  793         lsc->lsc_active_aggregator = NULL;
  794         lsc->lsc_strict_mode = 1;
  795         LACP_LOCK_INIT(lsc);
  796         TAILQ_INIT(&lsc->lsc_aggregators);
  797         LIST_INIT(&lsc->lsc_ports);
  798 
  799         /* Create a child of the parent lagg interface */
  800         oid = SYSCTL_ADD_NODE(&sc->ctx, SYSCTL_CHILDREN(sc->sc_oid),
  801             OID_AUTO, "lacp", CTLFLAG_RD, NULL, "LACP");
  802 
  803         /* Attach sysctl nodes */
  804         lacp_attach_sysctl(lsc, oid);
  805         lacp_attach_sysctl_debug(lsc, oid);
  806 
  807         callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0);
  808         callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0);
  809 
  810         /* if the lagg is already up then do the same */
  811         if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING)
  812                 lacp_init(sc);
  813 
  814         return (0);
  815 }
  816 
  817 int
  818 lacp_detach(struct lagg_softc *sc)
  819 {
  820         struct lacp_softc *lsc = LACP_SOFTC(sc);
  821 
  822         KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators),
  823             ("aggregators still active"));
  824         KASSERT(lsc->lsc_active_aggregator == NULL,
  825             ("aggregator still attached"));
  826 
  827         sc->sc_psc = NULL;
  828         callout_drain(&lsc->lsc_transit_callout);
  829         callout_drain(&lsc->lsc_callout);
  830 
  831         LACP_LOCK_DESTROY(lsc);
  832         free(lsc, M_DEVBUF);
  833         return (0);
  834 }
  835 
  836 void
  837 lacp_init(struct lagg_softc *sc)
  838 {
  839         struct lacp_softc *lsc = LACP_SOFTC(sc);
  840 
  841         LACP_LOCK(lsc);
  842         callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc);
  843         LACP_UNLOCK(lsc);
  844 }
  845 
  846 void
  847 lacp_stop(struct lagg_softc *sc)
  848 {
  849         struct lacp_softc *lsc = LACP_SOFTC(sc);
  850 
  851         LACP_LOCK(lsc);
  852         callout_stop(&lsc->lsc_transit_callout);
  853         callout_stop(&lsc->lsc_callout);
  854         LACP_UNLOCK(lsc);
  855 }
  856 
  857 struct lagg_port *
  858 lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m)
  859 {
  860         struct lacp_softc *lsc = LACP_SOFTC(sc);
  861         struct lacp_portmap *pm;
  862         struct lacp_port *lp;
  863         uint32_t hash;
  864 
  865         if (__predict_false(lsc->lsc_suppress_distributing)) {
  866                 LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__));
  867                 return (NULL);
  868         }
  869 
  870         pm = &lsc->lsc_pmap[lsc->lsc_activemap];
  871         if (pm->pm_count == 0) {
  872                 LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__));
  873                 return (NULL);
  874         }
  875 
  876         if (sc->use_flowid && (m->m_flags & M_FLOWID))
  877                 hash = m->m_pkthdr.flowid >> sc->flowid_shift;
  878         else
  879                 hash = lagg_hashmbuf(sc, m, lsc->lsc_hashkey);
  880         hash %= pm->pm_count;
  881         lp = pm->pm_map[hash];
  882 
  883         KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0,
  884             ("aggregated port is not distributing"));
  885 
  886         return (lp->lp_lagg);
  887 }
  888 /*
  889  * lacp_suppress_distributing: drop transmit packets for a while
  890  * to preserve packet ordering.
  891  */
  892 
  893 static void
  894 lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la)
  895 {
  896         struct lacp_port *lp;
  897 
  898         if (lsc->lsc_active_aggregator != la) {
  899                 return;
  900         }
  901 
  902         LACP_TRACE(NULL);
  903 
  904         lsc->lsc_suppress_distributing = TRUE;
  905 
  906         /* send a marker frame down each port to verify the queues are empty */
  907         LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) {
  908                 lp->lp_flags |= LACP_PORT_MARK;
  909                 lacp_xmit_marker(lp);
  910         }
  911 
  912         /* set a timeout for the marker frames */
  913         callout_reset(&lsc->lsc_transit_callout,
  914             LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc);
  915 }
  916 
  917 static int
  918 lacp_compare_peerinfo(const struct lacp_peerinfo *a,
  919     const struct lacp_peerinfo *b)
  920 {
  921         return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state)));
  922 }
  923 
  924 static int
  925 lacp_compare_systemid(const struct lacp_systemid *a,
  926     const struct lacp_systemid *b)
  927 {
  928         return (memcmp(a, b, sizeof(*a)));
  929 }
  930 
  931 #if 0   /* unused */
  932 static int
  933 lacp_compare_portid(const struct lacp_portid *a,
  934     const struct lacp_portid *b)
  935 {
  936         return (memcmp(a, b, sizeof(*a)));
  937 }
  938 #endif
  939 
  940 static uint64_t
  941 lacp_aggregator_bandwidth(struct lacp_aggregator *la)
  942 {
  943         struct lacp_port *lp;
  944         uint64_t speed;
  945 
  946         lp = TAILQ_FIRST(&la->la_ports);
  947         if (lp == NULL) {
  948                 return (0);
  949         }
  950 
  951         speed = ifmedia_baudrate(lp->lp_media);
  952         speed *= la->la_nports;
  953         if (speed == 0) {
  954                 LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n",
  955                     lp->lp_media, la->la_nports));
  956         }
  957 
  958         return (speed);
  959 }
  960 
  961 /*
  962  * lacp_select_active_aggregator: select an aggregator to be used to transmit
  963  * packets from lagg(4) interface.
  964  */
  965 
  966 static void
  967 lacp_select_active_aggregator(struct lacp_softc *lsc)
  968 {
  969         struct lacp_aggregator *la;
  970         struct lacp_aggregator *best_la = NULL;
  971         uint64_t best_speed = 0;
  972         char buf[LACP_LAGIDSTR_MAX+1];
  973 
  974         LACP_TRACE(NULL);
  975 
  976         TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
  977                 uint64_t speed;
  978 
  979                 if (la->la_nports == 0) {
  980                         continue;
  981                 }
  982 
  983                 speed = lacp_aggregator_bandwidth(la);
  984                 LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n",
  985                     lacp_format_lagid_aggregator(la, buf, sizeof(buf)),
  986                     speed, la->la_nports));
  987 
  988                 /* This aggregator is chosen if
  989                  *      the partner has a better system priority
  990                  *  or, the total aggregated speed is higher
  991                  *  or, it is already the chosen aggregator
  992                  */
  993                 if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) <
  994                      LACP_SYS_PRI(best_la->la_partner)) ||
  995                     speed > best_speed ||
  996                     (speed == best_speed &&
  997                     la == lsc->lsc_active_aggregator)) {
  998                         best_la = la;
  999                         best_speed = speed;
 1000                 }
 1001         }
 1002 
 1003         KASSERT(best_la == NULL || best_la->la_nports > 0,
 1004             ("invalid aggregator refcnt"));
 1005         KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports),
 1006             ("invalid aggregator list"));
 1007 
 1008         if (lsc->lsc_active_aggregator != best_la) {
 1009                 LACP_DPRINTF((NULL, "active aggregator changed\n"));
 1010                 LACP_DPRINTF((NULL, "old %s\n",
 1011                     lacp_format_lagid_aggregator(lsc->lsc_active_aggregator,
 1012                     buf, sizeof(buf))));
 1013         } else {
 1014                 LACP_DPRINTF((NULL, "active aggregator not changed\n"));
 1015         }
 1016         LACP_DPRINTF((NULL, "new %s\n",
 1017             lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
 1018 
 1019         if (lsc->lsc_active_aggregator != best_la) {
 1020                 lsc->lsc_active_aggregator = best_la;
 1021                 lacp_update_portmap(lsc);
 1022                 if (best_la) {
 1023                         lacp_suppress_distributing(lsc, best_la);
 1024                 }
 1025         }
 1026 }
 1027 
 1028 /*
 1029  * Updated the inactive portmap array with the new list of ports and
 1030  * make it live.
 1031  */
 1032 static void
 1033 lacp_update_portmap(struct lacp_softc *lsc)
 1034 {
 1035         struct lagg_softc *sc = lsc->lsc_softc;
 1036         struct lacp_aggregator *la;
 1037         struct lacp_portmap *p;
 1038         struct lacp_port *lp;
 1039         uint64_t speed;
 1040         u_int newmap;
 1041         int i;
 1042 
 1043         newmap = lsc->lsc_activemap == 0 ? 1 : 0;
 1044         p = &lsc->lsc_pmap[newmap];
 1045         la = lsc->lsc_active_aggregator;
 1046         speed = 0;
 1047         bzero(p, sizeof(struct lacp_portmap));
 1048 
 1049         if (la != NULL && la->la_nports > 0) {
 1050                 p->pm_count = la->la_nports;
 1051                 i = 0;
 1052                 TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
 1053                         p->pm_map[i++] = lp;
 1054                 KASSERT(i == p->pm_count, ("Invalid port count"));
 1055                 speed = lacp_aggregator_bandwidth(la);
 1056         }
 1057         sc->sc_ifp->if_baudrate = speed;
 1058 
 1059         /* switch the active portmap over */
 1060         atomic_store_rel_int(&lsc->lsc_activemap, newmap);
 1061         LACP_DPRINTF((NULL, "Set table %d with %d ports\n",
 1062                     lsc->lsc_activemap,
 1063                     lsc->lsc_pmap[lsc->lsc_activemap].pm_count));
 1064 }
 1065 
 1066 static uint16_t
 1067 lacp_compose_key(struct lacp_port *lp)
 1068 {
 1069         struct lagg_port *lgp = lp->lp_lagg;
 1070         struct lagg_softc *sc = lgp->lp_softc;
 1071         u_int media = lp->lp_media;
 1072         uint16_t key;
 1073 
 1074         if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) {
 1075 
 1076                 /*
 1077                  * non-aggregatable links should have unique keys.
 1078                  *
 1079                  * XXX this isn't really unique as if_index is 16 bit.
 1080                  */
 1081 
 1082                 /* bit 0..14:   (some bits of) if_index of this port */
 1083                 key = lp->lp_ifp->if_index;
 1084                 /* bit 15:      1 */
 1085                 key |= 0x8000;
 1086         } else {
 1087                 u_int subtype = IFM_SUBTYPE(media);
 1088 
 1089                 KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type"));
 1090                 KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface"));
 1091 
 1092                 /* bit 0..4:    IFM_SUBTYPE modulo speed */
 1093                 switch (subtype) {
 1094                 case IFM_10_T:
 1095                 case IFM_10_2:
 1096                 case IFM_10_5:
 1097                 case IFM_10_STP:
 1098                 case IFM_10_FL:
 1099                         key = IFM_10_T;
 1100                         break;
 1101                 case IFM_100_TX:
 1102                 case IFM_100_FX:
 1103                 case IFM_100_T4:
 1104                 case IFM_100_VG:
 1105                 case IFM_100_T2:
 1106                         key = IFM_100_TX;
 1107                         break;
 1108                 case IFM_1000_SX:
 1109                 case IFM_1000_LX:
 1110                 case IFM_1000_CX:
 1111                 case IFM_1000_T:
 1112                         key = IFM_1000_SX;
 1113                         break;
 1114                 case IFM_10G_LR:
 1115                 case IFM_10G_SR:
 1116                 case IFM_10G_CX4:
 1117                 case IFM_10G_TWINAX:
 1118                 case IFM_10G_TWINAX_LONG:
 1119                 case IFM_10G_LRM:
 1120                 case IFM_10G_T:
 1121                         key = IFM_10G_LR;
 1122                         break;
 1123                 case IFM_40G_CR4:
 1124                 case IFM_40G_SR4:
 1125                 case IFM_40G_LR4:
 1126                         key = IFM_40G_CR4;
 1127                         break;
 1128                 default:
 1129                         key = subtype;
 1130                 }
 1131                 /* bit 5..14:   (some bits of) if_index of lagg device */
 1132                 key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5);
 1133                 /* bit 15:      0 */
 1134         }
 1135         return (htons(key));
 1136 }
 1137 
 1138 static void
 1139 lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la)
 1140 {
 1141         char buf[LACP_LAGIDSTR_MAX+1];
 1142 
 1143         LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
 1144             __func__,
 1145             lacp_format_lagid(&la->la_actor, &la->la_partner,
 1146             buf, sizeof(buf)),
 1147             la->la_refcnt, la->la_refcnt + 1));
 1148 
 1149         KASSERT(la->la_refcnt > 0, ("refcount <= 0"));
 1150         la->la_refcnt++;
 1151         KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount"));
 1152 }
 1153 
 1154 static void
 1155 lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la)
 1156 {
 1157         char buf[LACP_LAGIDSTR_MAX+1];
 1158 
 1159         LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n",
 1160             __func__,
 1161             lacp_format_lagid(&la->la_actor, &la->la_partner,
 1162             buf, sizeof(buf)),
 1163             la->la_refcnt, la->la_refcnt - 1));
 1164 
 1165         KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt"));
 1166         la->la_refcnt--;
 1167         if (la->la_refcnt > 0) {
 1168                 return;
 1169         }
 1170 
 1171         KASSERT(la->la_refcnt == 0, ("refcount not zero"));
 1172         KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active"));
 1173 
 1174         TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q);
 1175 
 1176         free(la, M_DEVBUF);
 1177 }
 1178 
 1179 /*
 1180  * lacp_aggregator_get: allocate an aggregator.
 1181  */
 1182 
 1183 static struct lacp_aggregator *
 1184 lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp)
 1185 {
 1186         struct lacp_aggregator *la;
 1187 
 1188         la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT);
 1189         if (la) {
 1190                 la->la_refcnt = 1;
 1191                 la->la_nports = 0;
 1192                 TAILQ_INIT(&la->la_ports);
 1193                 la->la_pending = 0;
 1194                 TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q);
 1195         }
 1196 
 1197         return (la);
 1198 }
 1199 
 1200 /*
 1201  * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port.
 1202  */
 1203 
 1204 static void
 1205 lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp)
 1206 {
 1207         lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner);
 1208         lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor);
 1209 
 1210         la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION;
 1211 }
 1212 
 1213 static void
 1214 lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr,
 1215     const struct lacp_peerinfo *lpi_port)
 1216 {
 1217         memset(lpi_aggr, 0, sizeof(*lpi_aggr));
 1218         lpi_aggr->lip_systemid = lpi_port->lip_systemid;
 1219         lpi_aggr->lip_key = lpi_port->lip_key;
 1220 }
 1221 
 1222 /*
 1223  * lacp_aggregator_is_compatible: check if a port can join to an aggregator.
 1224  */
 1225 
 1226 static int
 1227 lacp_aggregator_is_compatible(const struct lacp_aggregator *la,
 1228     const struct lacp_port *lp)
 1229 {
 1230         if (!(lp->lp_state & LACP_STATE_AGGREGATION) ||
 1231             !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) {
 1232                 return (0);
 1233         }
 1234 
 1235         if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) {
 1236                 return (0);
 1237         }
 1238 
 1239         if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) {
 1240                 return (0);
 1241         }
 1242 
 1243         if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) {
 1244                 return (0);
 1245         }
 1246 
 1247         return (1);
 1248 }
 1249 
 1250 static int
 1251 lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a,
 1252     const struct lacp_peerinfo *b)
 1253 {
 1254         if (memcmp(&a->lip_systemid, &b->lip_systemid,
 1255             sizeof(a->lip_systemid))) {
 1256                 return (0);
 1257         }
 1258 
 1259         if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) {
 1260                 return (0);
 1261         }
 1262 
 1263         return (1);
 1264 }
 1265 
 1266 static void
 1267 lacp_port_enable(struct lacp_port *lp)
 1268 {
 1269         lp->lp_state |= LACP_STATE_AGGREGATION;
 1270 }
 1271 
 1272 static void
 1273 lacp_port_disable(struct lacp_port *lp)
 1274 {
 1275         lacp_set_mux(lp, LACP_MUX_DETACHED);
 1276 
 1277         lp->lp_state &= ~LACP_STATE_AGGREGATION;
 1278         lp->lp_selected = LACP_UNSELECTED;
 1279         lacp_sm_rx_record_default(lp);
 1280         lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION;
 1281         lp->lp_state &= ~LACP_STATE_EXPIRED;
 1282 }
 1283 
 1284 /*
 1285  * lacp_select: select an aggregator.  create one if necessary.
 1286  */
 1287 static void
 1288 lacp_select(struct lacp_port *lp)
 1289 {
 1290         struct lacp_softc *lsc = lp->lp_lsc;
 1291         struct lacp_aggregator *la;
 1292         char buf[LACP_LAGIDSTR_MAX+1];
 1293 
 1294         if (lp->lp_aggregator) {
 1295                 return;
 1296         }
 1297 
 1298         KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
 1299             ("timer_wait_while still active"));
 1300 
 1301         LACP_DPRINTF((lp, "port lagid=%s\n",
 1302             lacp_format_lagid(&lp->lp_actor, &lp->lp_partner,
 1303             buf, sizeof(buf))));
 1304 
 1305         TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) {
 1306                 if (lacp_aggregator_is_compatible(la, lp)) {
 1307                         break;
 1308                 }
 1309         }
 1310 
 1311         if (la == NULL) {
 1312                 la = lacp_aggregator_get(lsc, lp);
 1313                 if (la == NULL) {
 1314                         LACP_DPRINTF((lp, "aggregator creation failed\n"));
 1315 
 1316                         /*
 1317                          * will retry on the next tick.
 1318                          */
 1319 
 1320                         return;
 1321                 }
 1322                 lacp_fill_aggregator_id(la, lp);
 1323                 LACP_DPRINTF((lp, "aggregator created\n"));
 1324         } else {
 1325                 LACP_DPRINTF((lp, "compatible aggregator found\n"));
 1326                 if (la->la_refcnt == LACP_MAX_PORTS)
 1327                         return;
 1328                 lacp_aggregator_addref(lsc, la);
 1329         }
 1330 
 1331         LACP_DPRINTF((lp, "aggregator lagid=%s\n",
 1332             lacp_format_lagid(&la->la_actor, &la->la_partner,
 1333             buf, sizeof(buf))));
 1334 
 1335         lp->lp_aggregator = la;
 1336         lp->lp_selected = LACP_SELECTED;
 1337 }
 1338 
 1339 /*
 1340  * lacp_unselect: finish unselect/detach process.
 1341  */
 1342 
 1343 static void
 1344 lacp_unselect(struct lacp_port *lp)
 1345 {
 1346         struct lacp_softc *lsc = lp->lp_lsc;
 1347         struct lacp_aggregator *la = lp->lp_aggregator;
 1348 
 1349         KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
 1350             ("timer_wait_while still active"));
 1351 
 1352         if (la == NULL) {
 1353                 return;
 1354         }
 1355 
 1356         lp->lp_aggregator = NULL;
 1357         lacp_aggregator_delref(lsc, la);
 1358 }
 1359 
 1360 /* mux machine */
 1361 
 1362 static void
 1363 lacp_sm_mux(struct lacp_port *lp)
 1364 {
 1365         struct lagg_port *lgp = lp->lp_lagg;
 1366         struct lagg_softc *sc = lgp->lp_softc;
 1367         enum lacp_mux_state new_state;
 1368         boolean_t p_sync =
 1369                     (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
 1370         boolean_t p_collecting =
 1371             (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0;
 1372         enum lacp_selected selected = lp->lp_selected;
 1373         struct lacp_aggregator *la;
 1374 
 1375         if (lacp_debug > 1)
 1376                 lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
 1377                     "p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
 1378                     lp->lp_mux_state, selected, p_sync, p_collecting);
 1379 
 1380 re_eval:
 1381         la = lp->lp_aggregator;
 1382         KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL,
 1383             ("MUX not detached"));
 1384         new_state = lp->lp_mux_state;
 1385         switch (lp->lp_mux_state) {
 1386         case LACP_MUX_DETACHED:
 1387                 if (selected != LACP_UNSELECTED) {
 1388                         new_state = LACP_MUX_WAITING;
 1389                 }
 1390                 break;
 1391         case LACP_MUX_WAITING:
 1392                 KASSERT(la->la_pending > 0 ||
 1393                     !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE),
 1394                     ("timer_wait_while still active"));
 1395                 if (selected == LACP_SELECTED && la->la_pending == 0) {
 1396                         new_state = LACP_MUX_ATTACHED;
 1397                 } else if (selected == LACP_UNSELECTED) {
 1398                         new_state = LACP_MUX_DETACHED;
 1399                 }
 1400                 break;
 1401         case LACP_MUX_ATTACHED:
 1402                 if (selected == LACP_SELECTED && p_sync) {
 1403                         new_state = LACP_MUX_COLLECTING;
 1404                 } else if (selected != LACP_SELECTED) {
 1405                         new_state = LACP_MUX_DETACHED;
 1406                 }
 1407                 break;
 1408         case LACP_MUX_COLLECTING:
 1409                 if (selected == LACP_SELECTED && p_sync && p_collecting) {
 1410                         new_state = LACP_MUX_DISTRIBUTING;
 1411                 } else if (selected != LACP_SELECTED || !p_sync) {
 1412                         new_state = LACP_MUX_ATTACHED;
 1413                 }
 1414                 break;
 1415         case LACP_MUX_DISTRIBUTING:
 1416                 if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
 1417                         new_state = LACP_MUX_COLLECTING;
 1418                         lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n");
 1419                         sc->sc_flapping++;
 1420                 }
 1421                 break;
 1422         default:
 1423                 panic("%s: unknown state", __func__);
 1424         }
 1425 
 1426         if (lp->lp_mux_state == new_state) {
 1427                 return;
 1428         }
 1429 
 1430         lacp_set_mux(lp, new_state);
 1431         goto re_eval;
 1432 }
 1433 
 1434 static void
 1435 lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state)
 1436 {
 1437         struct lacp_aggregator *la = lp->lp_aggregator;
 1438 
 1439         if (lp->lp_mux_state == new_state) {
 1440                 return;
 1441         }
 1442 
 1443         switch (new_state) {
 1444         case LACP_MUX_DETACHED:
 1445                 lp->lp_state &= ~LACP_STATE_SYNC;
 1446                 lacp_disable_distributing(lp);
 1447                 lacp_disable_collecting(lp);
 1448                 lacp_sm_assert_ntt(lp);
 1449                 /* cancel timer */
 1450                 if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) {
 1451                         KASSERT(la->la_pending > 0,
 1452                             ("timer_wait_while not active"));
 1453                         la->la_pending--;
 1454                 }
 1455                 LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE);
 1456                 lacp_unselect(lp);
 1457                 break;
 1458         case LACP_MUX_WAITING:
 1459                 LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE,
 1460                     LACP_AGGREGATE_WAIT_TIME);
 1461                 la->la_pending++;
 1462                 break;
 1463         case LACP_MUX_ATTACHED:
 1464                 lp->lp_state |= LACP_STATE_SYNC;
 1465                 lacp_disable_collecting(lp);
 1466                 lacp_sm_assert_ntt(lp);
 1467                 break;
 1468         case LACP_MUX_COLLECTING:
 1469                 lacp_enable_collecting(lp);
 1470                 lacp_disable_distributing(lp);
 1471                 lacp_sm_assert_ntt(lp);
 1472                 break;
 1473         case LACP_MUX_DISTRIBUTING:
 1474                 lacp_enable_distributing(lp);
 1475                 break;
 1476         default:
 1477                 panic("%s: unknown state", __func__);
 1478         }
 1479 
 1480         LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state));
 1481 
 1482         lp->lp_mux_state = new_state;
 1483 }
 1484 
 1485 static void
 1486 lacp_sm_mux_timer(struct lacp_port *lp)
 1487 {
 1488         struct lacp_aggregator *la = lp->lp_aggregator;
 1489         char buf[LACP_LAGIDSTR_MAX+1];
 1490 
 1491         KASSERT(la->la_pending > 0, ("no pending event"));
 1492 
 1493         LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__,
 1494             lacp_format_lagid(&la->la_actor, &la->la_partner,
 1495             buf, sizeof(buf)),
 1496             la->la_pending, la->la_pending - 1));
 1497 
 1498         la->la_pending--;
 1499 }
 1500 
 1501 /* periodic transmit machine */
 1502 
 1503 static void
 1504 lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate)
 1505 {
 1506         if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state,
 1507             LACP_STATE_TIMEOUT)) {
 1508                 return;
 1509         }
 1510 
 1511         LACP_DPRINTF((lp, "partner timeout changed\n"));
 1512 
 1513         /*
 1514          * FAST_PERIODIC -> SLOW_PERIODIC
 1515          * or
 1516          * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC
 1517          *
 1518          * let lacp_sm_ptx_tx_schedule to update timeout.
 1519          */
 1520 
 1521         LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
 1522 
 1523         /*
 1524          * if timeout has been shortened, assert NTT.
 1525          */
 1526 
 1527         if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) {
 1528                 lacp_sm_assert_ntt(lp);
 1529         }
 1530 }
 1531 
 1532 static void
 1533 lacp_sm_ptx_tx_schedule(struct lacp_port *lp)
 1534 {
 1535         int timeout;
 1536 
 1537         if (!(lp->lp_state & LACP_STATE_ACTIVITY) &&
 1538             !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) {
 1539 
 1540                 /*
 1541                  * NO_PERIODIC
 1542                  */
 1543 
 1544                 LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC);
 1545                 return;
 1546         }
 1547 
 1548         if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) {
 1549                 return;
 1550         }
 1551 
 1552         timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ?
 1553             LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME;
 1554 
 1555         LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout);
 1556 }
 1557 
 1558 static void
 1559 lacp_sm_ptx_timer(struct lacp_port *lp)
 1560 {
 1561         lacp_sm_assert_ntt(lp);
 1562 }
 1563 
 1564 static void
 1565 lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du)
 1566 {
 1567         int timeout;
 1568 
 1569         /*
 1570          * check LACP_DISABLED first
 1571          */
 1572 
 1573         if (!(lp->lp_state & LACP_STATE_AGGREGATION)) {
 1574                 return;
 1575         }
 1576 
 1577         /*
 1578          * check loopback condition.
 1579          */
 1580 
 1581         if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid,
 1582             &lp->lp_actor.lip_systemid)) {
 1583                 return;
 1584         }
 1585 
 1586         /*
 1587          * EXPIRED, DEFAULTED, CURRENT -> CURRENT
 1588          */
 1589 
 1590         lacp_sm_rx_update_selected(lp, du);
 1591         lacp_sm_rx_update_ntt(lp, du);
 1592         lacp_sm_rx_record_pdu(lp, du);
 1593 
 1594         timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ?
 1595             LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME;
 1596         LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout);
 1597 
 1598         lp->lp_state &= ~LACP_STATE_EXPIRED;
 1599 
 1600         /*
 1601          * kick transmit machine without waiting the next tick.
 1602          */
 1603 
 1604         lacp_sm_tx(lp);
 1605 }
 1606 
 1607 static void
 1608 lacp_sm_rx_set_expired(struct lacp_port *lp)
 1609 {
 1610         lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
 1611         lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT;
 1612         LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME);
 1613         lp->lp_state |= LACP_STATE_EXPIRED;
 1614 }
 1615 
 1616 static void
 1617 lacp_sm_rx_timer(struct lacp_port *lp)
 1618 {
 1619         if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) {
 1620                 /* CURRENT -> EXPIRED */
 1621                 LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__));
 1622                 lacp_sm_rx_set_expired(lp);
 1623         } else {
 1624                 /* EXPIRED -> DEFAULTED */
 1625                 LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__));
 1626                 lacp_sm_rx_update_default_selected(lp);
 1627                 lacp_sm_rx_record_default(lp);
 1628                 lp->lp_state &= ~LACP_STATE_EXPIRED;
 1629         }
 1630 }
 1631 
 1632 static void
 1633 lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
 1634 {
 1635         boolean_t active;
 1636         uint8_t oldpstate;
 1637         char buf[LACP_STATESTR_MAX+1];
 1638 
 1639         LACP_TRACE(lp);
 1640 
 1641         oldpstate = lp->lp_partner.lip_state;
 1642 
 1643         active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY)
 1644             || ((lp->lp_state & LACP_STATE_ACTIVITY) &&
 1645             (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY));
 1646 
 1647         lp->lp_partner = du->ldu_actor;
 1648         if (active &&
 1649             ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
 1650             LACP_STATE_AGGREGATION) &&
 1651             !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner))
 1652             || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) {
 1653                 /* XXX nothing? */
 1654         } else {
 1655                 lp->lp_partner.lip_state &= ~LACP_STATE_SYNC;
 1656         }
 1657 
 1658         lp->lp_state &= ~LACP_STATE_DEFAULTED;
 1659 
 1660         if (oldpstate != lp->lp_partner.lip_state) {
 1661                 LACP_DPRINTF((lp, "old pstate %s\n",
 1662                     lacp_format_state(oldpstate, buf, sizeof(buf))));
 1663                 LACP_DPRINTF((lp, "new pstate %s\n",
 1664                     lacp_format_state(lp->lp_partner.lip_state, buf,
 1665                     sizeof(buf))));
 1666         }
 1667 
 1668         /* XXX Hack, still need to implement 5.4.9 para 2,3,4 */
 1669         if (lp->lp_lsc->lsc_strict_mode)
 1670                 lp->lp_partner.lip_state |= LACP_STATE_SYNC;
 1671 
 1672         lacp_sm_ptx_update_timeout(lp, oldpstate);
 1673 }
 1674 
 1675 static void
 1676 lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
 1677 {
 1678 
 1679         LACP_TRACE(lp);
 1680 
 1681         if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) ||
 1682             !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state,
 1683             LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) {
 1684                 LACP_DPRINTF((lp, "%s: assert ntt\n", __func__));
 1685                 lacp_sm_assert_ntt(lp);
 1686         }
 1687 }
 1688 
 1689 static void
 1690 lacp_sm_rx_record_default(struct lacp_port *lp)
 1691 {
 1692         uint8_t oldpstate;
 1693 
 1694         LACP_TRACE(lp);
 1695 
 1696         oldpstate = lp->lp_partner.lip_state;
 1697         if (lp->lp_lsc->lsc_strict_mode)
 1698                 lp->lp_partner = lacp_partner_admin_strict;
 1699         else
 1700                 lp->lp_partner = lacp_partner_admin_optimistic;;
 1701         lp->lp_state |= LACP_STATE_DEFAULTED;
 1702         lacp_sm_ptx_update_timeout(lp, oldpstate);
 1703 }
 1704 
 1705 static void
 1706 lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp,
 1707     const struct lacp_peerinfo *info)
 1708 {
 1709 
 1710         LACP_TRACE(lp);
 1711 
 1712         if (lacp_compare_peerinfo(&lp->lp_partner, info) ||
 1713             !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state,
 1714             LACP_STATE_AGGREGATION)) {
 1715                 lp->lp_selected = LACP_UNSELECTED;
 1716                 /* mux machine will clean up lp->lp_aggregator */
 1717         }
 1718 }
 1719 
 1720 static void
 1721 lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
 1722 {
 1723 
 1724         LACP_TRACE(lp);
 1725 
 1726         lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor);
 1727 }
 1728 
 1729 static void
 1730 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
 1731 {
 1732 
 1733         LACP_TRACE(lp);
 1734 
 1735         if (lp->lp_lsc->lsc_strict_mode)
 1736                 lacp_sm_rx_update_selected_from_peerinfo(lp,
 1737                     &lacp_partner_admin_strict);
 1738         else
 1739                 lacp_sm_rx_update_selected_from_peerinfo(lp,
 1740                     &lacp_partner_admin_optimistic);
 1741 }
 1742 
 1743 /* transmit machine */
 1744 
 1745 static void
 1746 lacp_sm_tx(struct lacp_port *lp)
 1747 {
 1748         int error = 0;
 1749 
 1750         if (!(lp->lp_state & LACP_STATE_AGGREGATION)
 1751 #if 1
 1752             || (!(lp->lp_state & LACP_STATE_ACTIVITY)
 1753             && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY))
 1754 #endif
 1755             ) {
 1756                 lp->lp_flags &= ~LACP_PORT_NTT;
 1757         }
 1758 
 1759         if (!(lp->lp_flags & LACP_PORT_NTT)) {
 1760                 return;
 1761         }
 1762 
 1763         /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */
 1764         if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent,
 1765                     (3 / LACP_FAST_PERIODIC_TIME)) == 0) {
 1766                 LACP_DPRINTF((lp, "rate limited pdu\n"));
 1767                 return;
 1768         }
 1769 
 1770         if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) {
 1771                 error = lacp_xmit_lacpdu(lp);
 1772         } else {
 1773                 LACP_TPRINTF((lp, "Dropping TX PDU\n"));
 1774         }
 1775 
 1776         if (error == 0) {
 1777                 lp->lp_flags &= ~LACP_PORT_NTT;
 1778         } else {
 1779                 LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n",
 1780                     error));
 1781         }
 1782 }
 1783 
 1784 static void
 1785 lacp_sm_assert_ntt(struct lacp_port *lp)
 1786 {
 1787 
 1788         lp->lp_flags |= LACP_PORT_NTT;
 1789 }
 1790 
 1791 static void
 1792 lacp_run_timers(struct lacp_port *lp)
 1793 {
 1794         int i;
 1795 
 1796         for (i = 0; i < LACP_NTIMER; i++) {
 1797                 KASSERT(lp->lp_timer[i] >= 0,
 1798                     ("invalid timer value %d", lp->lp_timer[i]));
 1799                 if (lp->lp_timer[i] == 0) {
 1800                         continue;
 1801                 } else if (--lp->lp_timer[i] <= 0) {
 1802                         if (lacp_timer_funcs[i]) {
 1803                                 (*lacp_timer_funcs[i])(lp);
 1804                         }
 1805                 }
 1806         }
 1807 }
 1808 
 1809 int
 1810 lacp_marker_input(struct lacp_port *lp, struct mbuf *m)
 1811 {
 1812         struct lacp_softc *lsc = lp->lp_lsc;
 1813         struct lagg_port *lgp = lp->lp_lagg;
 1814         struct lacp_port *lp2;
 1815         struct markerdu *mdu;
 1816         int error = 0;
 1817         int pending = 0;
 1818 
 1819         if (m->m_pkthdr.len != sizeof(*mdu)) {
 1820                 goto bad;
 1821         }
 1822 
 1823         if ((m->m_flags & M_MCAST) == 0) {
 1824                 goto bad;
 1825         }
 1826 
 1827         if (m->m_len < sizeof(*mdu)) {
 1828                 m = m_pullup(m, sizeof(*mdu));
 1829                 if (m == NULL) {
 1830                         return (ENOMEM);
 1831                 }
 1832         }
 1833 
 1834         mdu = mtod(m, struct markerdu *);
 1835 
 1836         if (memcmp(&mdu->mdu_eh.ether_dhost,
 1837             &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) {
 1838                 goto bad;
 1839         }
 1840 
 1841         if (mdu->mdu_sph.sph_version != 1) {
 1842                 goto bad;
 1843         }
 1844 
 1845         switch (mdu->mdu_tlv.tlv_type) {
 1846         case MARKER_TYPE_INFO:
 1847                 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
 1848                     marker_info_tlv_template, TRUE)) {
 1849                         goto bad;
 1850                 }
 1851                 mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE;
 1852                 memcpy(&mdu->mdu_eh.ether_dhost,
 1853                     &ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN);
 1854                 memcpy(&mdu->mdu_eh.ether_shost,
 1855                     lgp->lp_lladdr, ETHER_ADDR_LEN);
 1856                 error = lagg_enqueue(lp->lp_ifp, m);
 1857                 break;
 1858 
 1859         case MARKER_TYPE_RESPONSE:
 1860                 if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv,
 1861                     marker_response_tlv_template, TRUE)) {
 1862                         goto bad;
 1863                 }
 1864                 LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n",
 1865                     ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system,
 1866                     ":", ntohl(mdu->mdu_info.mi_rq_xid)));
 1867 
 1868                 /* Verify that it is the last marker we sent out */
 1869                 if (memcmp(&mdu->mdu_info, &lp->lp_marker,
 1870                     sizeof(struct lacp_markerinfo)))
 1871                         goto bad;
 1872 
 1873                 LACP_LOCK(lsc);
 1874                 lp->lp_flags &= ~LACP_PORT_MARK;
 1875 
 1876                 if (lsc->lsc_suppress_distributing) {
 1877                         /* Check if any ports are waiting for a response */
 1878                         LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) {
 1879                                 if (lp2->lp_flags & LACP_PORT_MARK) {
 1880                                         pending = 1;
 1881                                         break;
 1882                                 }
 1883                         }
 1884 
 1885                         if (pending == 0) {
 1886                                 /* All interface queues are clear */
 1887                                 LACP_DPRINTF((NULL, "queue flush complete\n"));
 1888                                 lsc->lsc_suppress_distributing = FALSE;
 1889                         }
 1890                 }
 1891                 LACP_UNLOCK(lsc);
 1892                 m_freem(m);
 1893                 break;
 1894 
 1895         default:
 1896                 goto bad;
 1897         }
 1898 
 1899         return (error);
 1900 
 1901 bad:
 1902         LACP_DPRINTF((lp, "bad marker frame\n"));
 1903         m_freem(m);
 1904         return (EINVAL);
 1905 }
 1906 
 1907 static int
 1908 tlv_check(const void *p, size_t size, const struct tlvhdr *tlv,
 1909     const struct tlv_template *tmpl, boolean_t check_type)
 1910 {
 1911         while (/* CONSTCOND */ 1) {
 1912                 if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) {
 1913                         return (EINVAL);
 1914                 }
 1915                 if ((check_type && tlv->tlv_type != tmpl->tmpl_type) ||
 1916                     tlv->tlv_length != tmpl->tmpl_length) {
 1917                         return (EINVAL);
 1918                 }
 1919                 if (tmpl->tmpl_type == 0) {
 1920                         break;
 1921                 }
 1922                 tlv = (const struct tlvhdr *)
 1923                     ((const char *)tlv + tlv->tlv_length);
 1924                 tmpl++;
 1925         }
 1926 
 1927         return (0);
 1928 }
 1929 
 1930 /* Debugging */
 1931 const char *
 1932 lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen)
 1933 {
 1934         snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X",
 1935             (int)mac[0],
 1936             (int)mac[1],
 1937             (int)mac[2],
 1938             (int)mac[3],
 1939             (int)mac[4],
 1940             (int)mac[5]);
 1941 
 1942         return (buf);
 1943 }
 1944 
 1945 const char *
 1946 lacp_format_systemid(const struct lacp_systemid *sysid,
 1947     char *buf, size_t buflen)
 1948 {
 1949         char macbuf[LACP_MACSTR_MAX+1];
 1950 
 1951         snprintf(buf, buflen, "%04X,%s",
 1952             ntohs(sysid->lsi_prio),
 1953             lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf)));
 1954 
 1955         return (buf);
 1956 }
 1957 
 1958 const char *
 1959 lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen)
 1960 {
 1961         snprintf(buf, buflen, "%04X,%04X",
 1962             ntohs(portid->lpi_prio),
 1963             ntohs(portid->lpi_portno));
 1964 
 1965         return (buf);
 1966 }
 1967 
 1968 const char *
 1969 lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen)
 1970 {
 1971         char sysid[LACP_SYSTEMIDSTR_MAX+1];
 1972         char portid[LACP_PORTIDSTR_MAX+1];
 1973 
 1974         snprintf(buf, buflen, "(%s,%04X,%s)",
 1975             lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)),
 1976             ntohs(peer->lip_key),
 1977             lacp_format_portid(&peer->lip_portid, portid, sizeof(portid)));
 1978 
 1979         return (buf);
 1980 }
 1981 
 1982 const char *
 1983 lacp_format_lagid(const struct lacp_peerinfo *a,
 1984     const struct lacp_peerinfo *b, char *buf, size_t buflen)
 1985 {
 1986         char astr[LACP_PARTNERSTR_MAX+1];
 1987         char bstr[LACP_PARTNERSTR_MAX+1];
 1988 
 1989 #if 0
 1990         /*
 1991          * there's a convention to display small numbered peer
 1992          * in the left.
 1993          */
 1994 
 1995         if (lacp_compare_peerinfo(a, b) > 0) {
 1996                 const struct lacp_peerinfo *t;
 1997 
 1998                 t = a;
 1999                 a = b;
 2000                 b = t;
 2001         }
 2002 #endif
 2003 
 2004         snprintf(buf, buflen, "[%s,%s]",
 2005             lacp_format_partner(a, astr, sizeof(astr)),
 2006             lacp_format_partner(b, bstr, sizeof(bstr)));
 2007 
 2008         return (buf);
 2009 }
 2010 
 2011 const char *
 2012 lacp_format_lagid_aggregator(const struct lacp_aggregator *la,
 2013     char *buf, size_t buflen)
 2014 {
 2015         if (la == NULL) {
 2016                 return ("(none)");
 2017         }
 2018 
 2019         return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen));
 2020 }
 2021 
 2022 const char *
 2023 lacp_format_state(uint8_t state, char *buf, size_t buflen)
 2024 {
 2025         snprintf(buf, buflen, "%b", state, LACP_STATE_BITS);
 2026         return (buf);
 2027 }
 2028 
 2029 static void
 2030 lacp_dump_lacpdu(const struct lacpdu *du)
 2031 {
 2032         char buf[LACP_PARTNERSTR_MAX+1];
 2033         char buf2[LACP_STATESTR_MAX+1];
 2034 
 2035         printf("actor=%s\n",
 2036             lacp_format_partner(&du->ldu_actor, buf, sizeof(buf)));
 2037         printf("actor.state=%s\n",
 2038             lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2)));
 2039         printf("partner=%s\n",
 2040             lacp_format_partner(&du->ldu_partner, buf, sizeof(buf)));
 2041         printf("partner.state=%s\n",
 2042             lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2)));
 2043 
 2044         printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay));
 2045 }
 2046 
 2047 static void
 2048 lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...)
 2049 {
 2050         va_list va;
 2051 
 2052         if (lp) {
 2053                 printf("%s: ", lp->lp_ifp->if_xname);
 2054         }
 2055 
 2056         va_start(va, fmt);
 2057         vprintf(fmt, va);
 2058         va_end(va);
 2059 }

Cache object: e599e75dbad326983e204605feac6136


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.