/*
* TCP RST from the peer, timeout, or some other such critical error.
*/
static int
do_abort_req(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_abort_req_rss *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct sge_wrq *ofld_txq = toep->ofld_txq;
struct inpcb *inp;
struct tcpcb *tp;
#ifdef INVARIANTS
unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl)));
#endif
KASSERT(opcode == CPL_ABORT_REQ_RSS,
("%s: unexpected opcode 0x%x", __func__, opcode));
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
if (toep->flags & TPF_SYNQE)
return (do_abort_req_synqe(iq, rss, m));
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
if (negative_advice(cpl->status)) {
CTR4(KTR_CXGBE, "%s: negative advice %d for tid %d (0x%x)",
__func__, cpl->status, tid, toep->flags);
return (0); /* Ignore negative advice */
}
inp = toep->inp;
INP_INFO_RLOCK(&V_tcbinfo); /* for tcp_close */
INP_WLOCK(inp);
tp = intotcpcb(inp);
CTR6(KTR_CXGBE,
"%s: tid %d (%s), toep_flags 0x%x, inp_flags 0x%x, status %d",
__func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags,
inp->inp_flags, cpl->status);
/*
* If we'd initiated an abort earlier the reply to it is responsible for
* cleaning up resources. Otherwise we tear everything down right here
* right now. We owe the T4 a CPL_ABORT_RPL no matter what.
*/
if (toep->flags & TPF_ABORT_SHUTDOWN) {
INP_WUNLOCK(inp);
goto done;
}
toep->flags |= TPF_ABORT_SHUTDOWN;
if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) {
struct socket *so = inp->inp_socket;
if (so != NULL)
so_error_set(so, abort_status_to_errno(tp,
cpl->status));
tp = tcp_close(tp);
if (tp == NULL)
INP_WLOCK(inp); /* re-acquire */
}
final_cpl_received(toep);
done:
INP_INFO_RUNLOCK(&V_tcbinfo);
send_abort_rpl(sc, ofld_txq, tid, CPL_ABORT_NO_RST);
return (0);
}
/*
* Peer has ACK'd our FIN.
*/
static int
do_close_con_rpl(struct sge_iq *iq, const struct rss_header *rss,
struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_close_con_rpl *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp = toep->inp;
struct tcpcb *tp = NULL;
struct socket *so = NULL;
#ifdef INVARIANTS
unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl)));
#endif
KASSERT(opcode == CPL_CLOSE_CON_RPL,
("%s: unexpected opcode 0x%x", __func__, opcode));
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
INP_INFO_RLOCK(&V_tcbinfo);
INP_WLOCK(inp);
tp = intotcpcb(inp);
CTR4(KTR_CXGBE, "%s: tid %u (%s), toep_flags 0x%x",
__func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags);
if (toep->flags & TPF_ABORT_SHUTDOWN)
goto done;
so = inp->inp_socket;
tp->snd_una = be32toh(cpl->snd_nxt) - 1; /* exclude FIN */
switch (tp->t_state) {
case TCPS_CLOSING: /* see TCPS_FIN_WAIT_2 in do_peer_close too */
tcp_twstart(tp);
release:
INP_UNLOCK_ASSERT(inp); /* safe, we have a ref on the inp */
INP_INFO_RUNLOCK(&V_tcbinfo);
INP_WLOCK(inp);
final_cpl_received(toep); /* no more CPLs expected */
return (0);
case TCPS_LAST_ACK:
if (tcp_close(tp))
INP_WUNLOCK(inp);
goto release;
case TCPS_FIN_WAIT_1:
if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
soisdisconnected(so);
tp->t_state = TCPS_FIN_WAIT_2;
break;
default:
log(LOG_ERR,
"%s: TID %u received CPL_CLOSE_CON_RPL in state %s\n",
__func__, tid, tcpstates[tp->t_state]);
}
done:
INP_WUNLOCK(inp);
INP_INFO_RUNLOCK(&V_tcbinfo);
return (0);
}
static int
do_fw4_ack(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_fw4_ack *cpl = (const void *)(rss + 1);
unsigned int tid = G_CPL_FW4_ACK_FLOWID(be32toh(OPCODE_TID(cpl)));
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp;
struct tcpcb *tp;
struct socket *so;
uint8_t credits = cpl->credits;
struct ofld_tx_sdesc *txsd;
int plen;
#ifdef INVARIANTS
unsigned int opcode = G_CPL_FW4_ACK_OPCODE(be32toh(OPCODE_TID(cpl)));
#endif
/*
* Very unusual case: we'd sent a flowc + abort_req for a synq entry and
* now this comes back carrying the credits for the flowc.
*/
if (__predict_false(toep->flags & TPF_SYNQE)) {
KASSERT(toep->flags & TPF_ABORT_SHUTDOWN,
("%s: credits for a synq entry %p", __func__, toep));
return (0);
}
inp = toep->inp;
KASSERT(opcode == CPL_FW4_ACK,
("%s: unexpected opcode 0x%x", __func__, opcode));
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
INP_WLOCK(inp);
if (__predict_false(toep->flags & TPF_ABORT_SHUTDOWN)) {
INP_WUNLOCK(inp);
return (0);
}
KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0,
("%s: inp_flags 0x%x", __func__, inp->inp_flags));
tp = intotcpcb(inp);
if (cpl->flags & CPL_FW4_ACK_FLAGS_SEQVAL) {
tcp_seq snd_una = be32toh(cpl->snd_una);
#ifdef INVARIANTS
if (__predict_false(SEQ_LT(snd_una, tp->snd_una))) {
log(LOG_ERR,
"%s: unexpected seq# %x for TID %u, snd_una %x\n",
__func__, snd_una, toep->tid, tp->snd_una);
}
#endif
if (tp->snd_una != snd_una) {
tp->snd_una = snd_una;
tp->ts_recent_age = tcp_ts_getticks();
}
}
so = inp->inp_socket;
txsd = &toep->txsd[toep->txsd_cidx];
plen = 0;
while (credits) {
KASSERT(credits >= txsd->tx_credits,
("%s: too many (or partial) credits", __func__));
credits -= txsd->tx_credits;
toep->tx_credits += txsd->tx_credits;
plen += txsd->plen;
txsd++;
toep->txsd_avail++;
KASSERT(toep->txsd_avail <= toep->txsd_total,
("%s: txsd avail > total", __func__));
if (__predict_false(++toep->txsd_cidx == toep->txsd_total)) {
txsd = &toep->txsd[0];
toep->txsd_cidx = 0;
}
}
if (toep->tx_credits == toep->tx_total) {
toep->tx_nocompl = 0;
toep->plen_nocompl = 0;
}
if (toep->flags & TPF_TX_SUSPENDED &&
toep->tx_credits >= toep->tx_total / 4) {
toep->flags &= ~TPF_TX_SUSPENDED;
t4_push_frames(sc, toep, plen);
} else if (plen > 0) {
struct sockbuf *sb = &so->so_snd;
SOCKBUF_LOCK(sb);
sbdrop_locked(sb, plen);
sowwakeup_locked(so);
SOCKBUF_UNLOCK_ASSERT(sb);
}
INP_WUNLOCK(inp);
return (0);
}
/*
* Peer has sent us a FIN.
*/
static int
do_peer_close(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_peer_close *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp = toep->inp;
struct tcpcb *tp = NULL;
struct socket *so;
#ifdef INVARIANTS
unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl)));
#endif
KASSERT(opcode == CPL_PEER_CLOSE,
("%s: unexpected opcode 0x%x", __func__, opcode));
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
if (__predict_false(toep->flags & TPF_SYNQE)) {
#ifdef INVARIANTS
struct synq_entry *synqe = (void *)toep;
INP_WLOCK(synqe->lctx->inp);
if (synqe->flags & TPF_SYNQE_HAS_L2TE) {
KASSERT(synqe->flags & TPF_ABORT_SHUTDOWN,
("%s: listen socket closed but tid %u not aborted.",
__func__, tid));
} else {
/*
* do_pass_accept_req is still running and will
* eventually take care of this tid.
*/
}
INP_WUNLOCK(synqe->lctx->inp);
#endif
CTR4(KTR_CXGBE, "%s: tid %u, synqe %p (0x%x)", __func__, tid,
toep, toep->flags);
return (0);
}
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
INP_INFO_RLOCK(&V_tcbinfo);
INP_WLOCK(inp);
tp = intotcpcb(inp);
CTR5(KTR_CXGBE, "%s: tid %u (%s), toep_flags 0x%x, inp %p", __func__,
tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags, inp);
if (toep->flags & TPF_ABORT_SHUTDOWN)
goto done;
tp->rcv_nxt++; /* FIN */
so = inp->inp_socket;
if (toep->ulp_mode == ULP_MODE_TCPDDP) {
DDP_LOCK(toep);
if (__predict_false(toep->ddp_flags &
(DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE)))
handle_ddp_close(toep, tp, cpl->rcv_nxt);
DDP_UNLOCK(toep);
}
socantrcvmore(so);
if (toep->ulp_mode != ULP_MODE_RDMA) {
KASSERT(tp->rcv_nxt == be32toh(cpl->rcv_nxt),
("%s: rcv_nxt mismatch: %u %u", __func__, tp->rcv_nxt,
be32toh(cpl->rcv_nxt)));
}
switch (tp->t_state) {
case TCPS_SYN_RECEIVED:
tp->t_starttime = ticks;
/* FALLTHROUGH */
case TCPS_ESTABLISHED:
tp->t_state = TCPS_CLOSE_WAIT;
break;
case TCPS_FIN_WAIT_1:
tp->t_state = TCPS_CLOSING;
break;
case TCPS_FIN_WAIT_2:
tcp_twstart(tp);
INP_UNLOCK_ASSERT(inp); /* safe, we have a ref on the inp */
INP_INFO_RUNLOCK(&V_tcbinfo);
INP_WLOCK(inp);
final_cpl_received(toep);
return (0);
default:
log(LOG_ERR, "%s: TID %u received CPL_PEER_CLOSE in state %d\n",
__func__, tid, tp->t_state);
}
done:
INP_WUNLOCK(inp);
INP_INFO_RUNLOCK(&V_tcbinfo);
return (0);
}
/*
* active open (soconnect).
*
* State of affairs on entry:
* soisconnecting (so_state |= SS_ISCONNECTING)
* tcbinfo not locked (This has changed - used to be WLOCKed)
* inp WLOCKed
* tp->t_state = TCPS_SYN_SENT
* rtalloc1, RT_UNLOCK on rt.
*/
int
t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt,
struct sockaddr *nam)
{
struct adapter *sc = tod->tod_softc;
struct tom_data *td = tod_td(tod);
struct toepcb *toep = NULL;
struct wrqe *wr = NULL;
struct ifnet *rt_ifp = rt->rt_ifp;
struct port_info *pi;
int mtu_idx, rscale, qid_atid, rc, isipv6;
struct inpcb *inp = sotoinpcb(so);
struct tcpcb *tp = intotcpcb(inp);
int reason;
INP_WLOCK_ASSERT(inp);
KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6,
("%s: dest addr %p has family %u", __func__, nam, nam->sa_family));
if (rt_ifp->if_type == IFT_ETHER)
pi = rt_ifp->if_softc;
else if (rt_ifp->if_type == IFT_L2VLAN) {
struct ifnet *ifp = VLAN_COOKIE(rt_ifp);
pi = ifp->if_softc;
} else if (rt_ifp->if_type == IFT_IEEE8023ADLAG)
DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */
else
DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP);
toep = alloc_toepcb(pi, -1, -1, M_NOWAIT);
if (toep == NULL)
DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
toep->tid = alloc_atid(sc, toep);
if (toep->tid < 0)
DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
toep->l2te = t4_l2t_get(pi, rt_ifp,
rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam);
if (toep->l2te == NULL)
DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
isipv6 = nam->sa_family == AF_INET6;
wr = alloc_wrqe(isipv6 ? sizeof(struct cpl_act_open_req6) :
sizeof(struct cpl_act_open_req), toep->ctrlq);
if (wr == NULL)
DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM);
if (sc->tt.ddp && (so->so_options & SO_NO_DDP) == 0)
set_tcpddp_ulp_mode(toep);
else
toep->ulp_mode = ULP_MODE_NONE;
SOCKBUF_LOCK(&so->so_rcv);
/* opt0 rcv_bufsiz initially, assumes its normal meaning later */
toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ);
SOCKBUF_UNLOCK(&so->so_rcv);
/*
* The kernel sets request_r_scale based on sb_max whereas we need to
* take hardware's MAX_RCV_WND into account too. This is normally a
* no-op as MAX_RCV_WND is much larger than the default sb_max.
*/
if (tp->t_flags & TF_REQ_SCALE)
rscale = tp->request_r_scale = select_rcv_wscale();
else
rscale = 0;
mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, 0);
qid_atid = (toep->ofld_rxq->iq.abs_id << 14) | toep->tid;
if (isipv6) {
struct cpl_act_open_req6 *cpl = wrtod(wr);
if ((inp->inp_vflag & INP_IPV6) == 0) {
/* XXX think about this a bit more */
log(LOG_ERR,
"%s: time to think about AF_INET6 + vflag 0x%x.\n",
__func__, inp->inp_vflag);
DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP);
}
toep->ce = hold_lip(td, &inp->in6p_laddr);
if (toep->ce == NULL)
DONT_OFFLOAD_ACTIVE_OPEN(ENOENT);
INIT_TP_WR(cpl, 0);
OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
qid_atid));
cpl->local_port = inp->inp_lport;
cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0];
cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8];
cpl->peer_port = inp->inp_fport;
cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0];
cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8];
cpl->opt0 = calc_opt0(so, pi, toep->l2te, mtu_idx, rscale,
toep->rx_credits, toep->ulp_mode);
cpl->params = select_ntuple(pi, toep->l2te, sc->filter_mode);
cpl->opt2 = calc_opt2a(so, toep);
} else {
struct cpl_act_open_req *cpl = wrtod(wr);
INIT_TP_WR(cpl, 0);
OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
qid_atid));
inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port,
&cpl->peer_ip, &cpl->peer_port);
cpl->opt0 = calc_opt0(so, pi, toep->l2te, mtu_idx, rscale,
toep->rx_credits, toep->ulp_mode);
cpl->params = select_ntuple(pi, toep->l2te, sc->filter_mode);
cpl->opt2 = calc_opt2a(so, toep);
}
CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__,
toep->tid, tcpstates[tp->t_state], toep, inp);
offload_socket(so, toep);
rc = t4_l2t_send(sc, wr, toep->l2te);
if (rc == 0) {
toep->flags |= TPF_CPL_PENDING;
return (0);
}
undo_offload_socket(so);
reason = __LINE__;
failed:
CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc);
if (wr)
free_wrqe(wr);
if (toep) {
if (toep->tid >= 0)
free_atid(sc, toep->tid);
if (toep->l2te)
t4_l2t_release(toep->l2te);
if (toep->ce)
release_lip(td, toep->ce);
free_toepcb(toep);
}
return (rc);
}
