mirror/freebsd
1
0
Fork 0
mirror of https://git.FreeBSD.org/src.git synced 2024-12-15 10:17:20 +00:00
Code Issues Releases Activity

Bring over some link aggregation / LACP protocol improvements and debugging

Browse Source 
additions.

* Add some new tracing events to aid in debugging.
* Add in a debugging mode to drop transmit and received frames, specifically
  to test whether seeing or hearing heartbeats correctly cause LACP to
  drop the port.
* Add in (and make default) a strict LACP mode, which requires the
  heartbeat on a port to be heard before it's used.  Sometimes vendor ports
  will hang but the link layer stays up, resulting in hung traffic.
* Add logging the number of link status flaps, again to aid in debugging
  badly behaving switch ports.
* Calculate the lagg interface port speed as the multiple of the
  configured ports, rather than the largest.

Obtained from:	Netflix
MFC after:	2 weeks
This commit is contained in:
Adrian Chadd 2013-07-13 04:25:03 +00:00
parent 8945bee0c7
commit 31402c27b8
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=253314
3 changed files with 98 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

81
sys/net/ieee8023ad_lacp.c
View File

@ -188,29 +188,43 @@ static void lacp_dprintf(const struct lacp_port *, const char *, ...)
__attribute__((__format__(__printf__, 2, 3)));
static int lacp_debug = 0;
SYSCTL_INT(_net, OID_AUTO, lacp_debug, CTLFLAG_RW | CTLFLAG_TUN,
SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad");
SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_TUN,
&lacp_debug, 0, "Enable LACP debug logging (1=debug, 2=trace)");
TUNABLE_INT("net.lacp_debug", &lacp_debug);
TUNABLE_INT("net.link.lagg.lacp.debug", &lacp_debug);
#define LACP_DPRINTF(a) if (lacp_debug > 0) { lacp_dprintf a ; }
#define LACP_TRACE(a) if (lacp_debug > 1) { lacp_dprintf(a,"%s\n",__func__); }
/* bitmap of ports */
static int lacp_rx_test = 0;
static int lacp_tx_test = 0;
SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, rxtest, CTLFLAG_RW, &lacp_rx_test, 0,
"RXTest");
SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, txtest, CTLFLAG_RW, &lacp_tx_test, 0,
"TXTest");
static int lacp_strict = 1;
SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, strict, CTLFLAG_RW, &lacp_strict,
0, "Strict spec compliance");
#define LACP_DPRINTF(a) if (lacp_debug & 0x01) { lacp_dprintf a ; }
#define LACP_TRACE(a) if (lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); }
#define LACP_TPRINTF(a) if (lacp_debug & 0x04) { lacp_dprintf a ; }
/*
* partner administration variables.
* XXX should be configurable.
*/
static const struct lacp_peerinfo lacp_partner_admin = {
static const struct lacp_peerinfo lacp_partner_admin_optimistic = {
.lip_systemid = { .lsi_prio = 0xffff },
.lip_portid = { .lpi_prio = 0xffff },
#if 1
/* optimistic */
.lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING,
#else
/* pessimistic */
};
static const struct lacp_peerinfo lacp_partner_admin_strict = {
.lip_systemid = { .lsi_prio = 0xffff },
.lip_portid = { .lpi_prio = 0xffff },
.lip_state = 0,
#endif
};
static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
@ -301,6 +315,11 @@ lacp_pdu_input(struct lacp_port *lp, struct mbuf *m)
lacp_dump_lacpdu(du);
}
if ((1 << lp->lp_ifp->if_dunit) & lacp_rx_test) {
LACP_TPRINTF((lp, "Dropping RX PDU\n"));
goto bad;
}
LACP_LOCK(lsc);
lacp_sm_rx(lp, du);
LACP_UNLOCK(lsc);
@ -653,6 +672,7 @@ lacp_disable_distributing(struct lacp_port *lp)
{
struct lacp_aggregator *la = lp->lp_aggregator;
struct lacp_softc *lsc = lp->lp_lsc;
struct lagg_softc *sc = lsc->lsc_softc;
char buf[LACP_LAGIDSTR_MAX+1];
LACP_LOCK_ASSERT(lsc);
@ -672,6 +692,7 @@ lacp_disable_distributing(struct lacp_port *lp)
TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q);
la->la_nports--;
sc->sc_active = la->la_nports;
if (lsc->lsc_active_aggregator == la) {
lacp_suppress_distributing(lsc, la);
@ -688,6 +709,7 @@ lacp_enable_distributing(struct lacp_port *lp)
{
struct lacp_aggregator *la = lp->lp_aggregator;
struct lacp_softc *lsc = lp->lp_lsc;
struct lagg_softc *sc = lsc->lsc_softc;
char buf[LACP_LAGIDSTR_MAX+1];
LACP_LOCK_ASSERT(lsc);
@ -704,6 +726,7 @@ lacp_enable_distributing(struct lacp_port *lp)
KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid"));
TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q);
la->la_nports++;
sc->sc_active = la->la_nports;
lp->lp_state |= LACP_STATE_DISTRIBUTING;
@ -908,7 +931,6 @@ lacp_aggregator_bandwidth(struct lacp_aggregator *la)
static void
lacp_select_active_aggregator(struct lacp_softc *lsc)
{
struct lagg_softc *sc = lsc->lsc_softc;
struct lacp_aggregator *la;
struct lacp_aggregator *best_la = NULL;
uint64_t best_speed = 0;
@ -960,7 +982,6 @@ lacp_select_active_aggregator(struct lacp_softc *lsc)
lacp_format_lagid_aggregator(best_la, buf, sizeof(buf))));
if (lsc->lsc_active_aggregator != best_la) {
sc->sc_ifp->if_baudrate = best_speed;
lsc->lsc_active_aggregator = best_la;
lacp_update_portmap(lsc);
if (best_la) {
@ -976,15 +997,18 @@ lacp_select_active_aggregator(struct lacp_softc *lsc)
static void
lacp_update_portmap(struct lacp_softc *lsc)
{
struct lagg_softc *sc = lsc->lsc_softc;
struct lacp_aggregator *la;
struct lacp_portmap *p;
struct lacp_port *lp;
uint64_t speed;
u_int newmap;
int i;
newmap = lsc->lsc_activemap == 0 ? 1 : 0;
p = &lsc->lsc_pmap[newmap];
la = lsc->lsc_active_aggregator;
speed = 0;
bzero(p, sizeof(struct lacp_portmap));
if (la != NULL && la->la_nports > 0) {
@ -993,7 +1017,9 @@ lacp_update_portmap(struct lacp_softc *lsc)
TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q)
p->pm_map[i++] = lp;
KASSERT(i == p->pm_count, ("Invalid port count"));
speed = lacp_aggregator_bandwidth(la);
}
sc->sc_ifp->if_baudrate = speed;
/* switch the active portmap over */
atomic_store_rel_int(&lsc->lsc_activemap, newmap);
@ -1264,6 +1290,8 @@ lacp_unselect(struct lacp_port *lp)
static void
lacp_sm_mux(struct lacp_port *lp)
{
struct lagg_port *lgp = lp->lp_lagg;
struct lagg_softc *sc = lgp->lp_softc;
enum lacp_mux_state new_state;
boolean_t p_sync =
(lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0;
@ -1273,7 +1301,9 @@ lacp_sm_mux(struct lacp_port *lp)
struct lacp_aggregator *la;
if (lacp_debug > 1)
lacp_dprintf(lp, "%s: state %d\n", __func__, lp->lp_mux_state);
lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, "
"p_sync= 0x%x, p_collecting= 0x%x\n", __func__,
lp->lp_mux_state, selected, p_sync, p_collecting);
re_eval:
la = lp->lp_aggregator;
@ -1313,6 +1343,8 @@ lacp_sm_mux(struct lacp_port *lp)
case LACP_MUX_DISTRIBUTING:
if (selected != LACP_SELECTED || !p_sync || !p_collecting) {
new_state = LACP_MUX_COLLECTING;
lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flaping\n");
sc->sc_flapping++;
}
break;
default:
@ -1561,6 +1593,10 @@ lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du)
sizeof(buf))));
}
/* XXX Hack, still need to implement 5.4.9 para 2,3,4 */
if (lacp_strict)
lp->lp_partner.lip_state |= LACP_STATE_SYNC;
lacp_sm_ptx_update_timeout(lp, oldpstate);
}
@ -1586,7 +1622,10 @@ lacp_sm_rx_record_default(struct lacp_port *lp)
LACP_TRACE(lp);
oldpstate = lp->lp_partner.lip_state;
lp->lp_partner = lacp_partner_admin;
if (lacp_strict)
lp->lp_partner = lacp_partner_admin_strict;
else
lp->lp_partner = lacp_partner_admin_optimistic;;
lp->lp_state |= LACP_STATE_DEFAULTED;
lacp_sm_ptx_update_timeout(lp, oldpstate);
}
@ -1621,7 +1660,12 @@ lacp_sm_rx_update_default_selected(struct lacp_port *lp)
LACP_TRACE(lp);
lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin);
if (lacp_strict)
lacp_sm_rx_update_selected_from_peerinfo(lp,
&lacp_partner_admin_strict);
else
lacp_sm_rx_update_selected_from_peerinfo(lp,
&lacp_partner_admin_optimistic);
}
/* transmit machine */
@ -1629,7 +1673,7 @@ lacp_sm_rx_update_default_selected(struct lacp_port *lp)
static void
lacp_sm_tx(struct lacp_port *lp)
{
int error;
int error = 0;
if (!(lp->lp_state & LACP_STATE_AGGREGATION)
#if 1
@ -1651,7 +1695,10 @@ lacp_sm_tx(struct lacp_port *lp)
return;
}
error = lacp_xmit_lacpdu(lp);
if (((1 << lp->lp_ifp->if_dunit) & lacp_tx_test) == 0)
error = lacp_xmit_lacpdu(lp);
else
LACP_TPRINTF((lp, "Dropping TX PDU\n"));
if (error == 0) {
lp->lp_flags &= ~LACP_PORT_NTT;

30
sys/net/if_lagg.c
View File

@ -122,6 +122,7 @@ static void lagg_media_status(struct ifnet *, struct ifmediareq *);
static struct lagg_port *lagg_link_active(struct lagg_softc *,
struct lagg_port *);
static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *);
static int lagg_sysctl_active(SYSCTL_HANDLER_ARGS);
/* Simple round robin */
static int lagg_rr_attach(struct lagg_softc *);
@ -171,7 +172,7 @@ static const struct {
};
SYSCTL_DECL(_net_link);
static SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0,
"Link Aggregation");
static int lagg_failover_rx_all = 0; /* Allow input on any failover links */
@ -298,6 +299,12 @@ lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params)
SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
"count", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_count, sc->sc_count,
"Total number of ports");
SYSCTL_ADD_PROC(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
"active", CTLTYPE_INT|CTLFLAG_RD, sc, 0, lagg_sysctl_active,
"I", "Total number of active ports");
SYSCTL_ADD_INT(&sc->ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
"flapping", CTLTYPE_INT|CTLFLAG_RD, &sc->sc_flapping,
sc->sc_flapping, "Total number of port change events");
/* Hash all layers by default */
sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4;
@ -1488,6 +1495,27 @@ lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf)
return (mtod(m, char *) + off);
}
static int
lagg_sysctl_active(SYSCTL_HANDLER_ARGS)
{
struct lagg_softc *sc = (struct lagg_softc *)arg1;
struct lagg_port *lp;
int error;
/* LACP tracks active links automatically, the others do not */
if (sc->sc_proto != LAGG_PROTO_LACP) {
sc->sc_active = 0;
SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
sc->sc_active += LAGG_PORTACTIVE(lp);
}
error = sysctl_handle_int(oidp, &sc->sc_active, 0, req);
if ((error) || (req->newptr == NULL))
return (error);
return (0);
}
uint32_t
lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key)
{

5
sys/net/if_lagg.h
View File

@ -190,6 +190,9 @@ struct lagg_softc {
struct rwlock sc_mtx;
int sc_proto; /* lagg protocol */
u_int sc_count; /* number of ports */
u_int sc_active; /* active port count */
u_int sc_flapping; /* number of flapping
* events */
struct lagg_port *sc_primary; /* primary port */
struct ifmedia sc_media; /* media config */
caddr_t sc_psc; /* protocol data */
@ -266,6 +269,8 @@ extern void (*lagg_linkstate_p)(struct ifnet *, int );
int lagg_enqueue(struct ifnet *, struct mbuf *);
uint32_t lagg_hashmbuf(struct lagg_softc *, struct mbuf *, uint32_t);
SYSCTL_DECL(_net_link_lagg);
#endif /* _KERNEL */
#endif /* _NET_LAGG_H */