static int
do_rx_iscsi_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
struct cxgbei_data *ci = sc->iscsi_ulp_softc;
struct cpl_iscsi_data *cpl = mtod(m, struct cpl_iscsi_data *);
u_int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct icl_cxgbei_pdu *icp = toep->ulpcb2;
M_ASSERTPKTHDR(m);
MPASS(m->m_pkthdr.len == be16toh(cpl->len) + sizeof(*cpl));
/* Must already have received the header (but not the data). */
MPASS(icp != NULL);
MPASS(icp->icp_flags == ICPF_RX_HDR);
MPASS(icp->ip.ip_data_mbuf == NULL);
m_adj(m, sizeof(*cpl));
MPASS(icp->ip.ip_data_len == m->m_pkthdr.len);
icp->icp_flags |= ICPF_RX_FLBUF;
icp->ip.ip_data_mbuf = m;
counter_u64_add(ci->fl_pdus, 1);
counter_u64_add(ci->fl_bytes, m->m_pkthdr.len);
#if 0
CTR3(KTR_CXGBE, "%s: tid %u, cpl->len %u", __func__, tid,
be16toh(cpl->len));
#endif
return (0);
}
/*
* Reply to the CPL_ABORT_REQ (send_reset)
*/
static int
do_abort_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_abort_rpl_rss *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp = toep->inp;
#ifdef INVARIANTS
unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl)));
#endif
KASSERT(opcode == CPL_ABORT_RPL_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_rpl_synqe(iq, rss, m));
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
CTR5(KTR_CXGBE, "%s: tid %u, toep %p, inp %p, status %d",
__func__, tid, toep, inp, cpl->status);
KASSERT(toep->flags & TPF_ABORT_SHUTDOWN,
("%s: wasn't expecting abort reply", __func__));
INP_WLOCK(inp);
final_cpl_received(toep);
return (0);
}
static int
do_rx_iscsi_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
struct cpl_iscsi_data *cpl = mtod(m, struct cpl_iscsi_data *);
u_int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct icl_cxgbei_pdu *icp = toep->ulpcb2;
M_ASSERTPKTHDR(m);
/* Must already have received the header (but not the data). */
MPASS(icp != NULL);
MPASS(icp->pdu_flags == SBUF_ULP_FLAG_HDR_RCVD);
MPASS(icp->ip.ip_data_mbuf == NULL);
MPASS(icp->ip.ip_data_len == 0);
m_adj(m, sizeof(*cpl));
icp->pdu_flags |= SBUF_ULP_FLAG_DATA_RCVD;
icp->ip.ip_data_mbuf = m;
icp->ip.ip_data_len = m->m_pkthdr.len;
#if 0
CTR4(KTR_CXGBE, "%s: tid %u, cpl->len dlen %u, m->m_len dlen %u",
__func__, tid, ntohs(cpl->len), m->m_len);
#endif
return (0);
}
static int
do_abort_req_rss(struct t3cdev *dev, struct mbuf *m)
{
union opcode_tid *p = cplhdr(m);
unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
struct toe_tid_entry *toe_tid;
toe_tid = lookup_tid(&(T3C_DATA (dev))->tid_maps, hwtid);
if (toe_tid->ctx && toe_tid->client->handlers &&
toe_tid->client->handlers[p->opcode]) {
return toe_tid->client->handlers[p->opcode]
(dev, m, toe_tid->ctx);
} else {
struct cpl_abort_req_rss *req = cplhdr(m);
struct cpl_abort_rpl *rpl;
struct mbuf *m = m_get(M_NOWAIT, MT_DATA);
if (!m) {
log(LOG_NOTICE, "do_abort_req_rss: couldn't get mbuf!\n");
goto out;
}
m_set_priority(m, CPL_PRIORITY_DATA);
rpl = cplhdr(m);
rpl->wr.wr_hi =
htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
rpl->wr.wr_lo = htonl(V_WR_TID(GET_TID(req)));
OPCODE_TID(rpl) =
htonl(MK_OPCODE_TID(CPL_ABORT_RPL, GET_TID(req)));
rpl->cmd = req->status;
cxgb_ofld_send(dev, m);
out:
return (CPL_RET_BUF_DONE);
}
}
static int cxgb4_del_hash_filter(struct net_device *dev, int filter_id,
struct filter_ctx *ctx)
{
struct adapter *adapter = netdev2adap(dev);
struct tid_info *t = &adapter->tids;
struct cpl_abort_req *abort_req;
struct cpl_abort_rpl *abort_rpl;
struct cpl_set_tcb_field *req;
struct ulptx_idata *aligner;
struct work_request_hdr *wr;
struct filter_entry *f;
struct sk_buff *skb;
unsigned int wrlen;
int ret;
netdev_dbg(dev, "%s: filter_id = %d ; nftids = %d\n",
__func__, filter_id, adapter->tids.nftids);
if (filter_id > adapter->tids.ntids)
return -E2BIG;
f = lookup_tid(t, filter_id);
if (!f) {
netdev_err(dev, "%s: no filter entry for filter_id = %d",
__func__, filter_id);
return -EINVAL;
}
ret = writable_filter(f);
if (ret)
return ret;
if (!f->valid)
return -EINVAL;
f->ctx = ctx;
f->pending = 1;
wrlen = roundup(sizeof(*wr) + (sizeof(*req) + sizeof(*aligner))
+ sizeof(*abort_req) + sizeof(*abort_rpl), 16);
skb = alloc_skb(wrlen, GFP_KERNEL);
if (!skb) {
netdev_err(dev, "%s: could not allocate skb ..\n", __func__);
return -ENOMEM;
}
set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
req = (struct cpl_set_tcb_field *)__skb_put(skb, wrlen);
INIT_ULPTX_WR(req, wrlen, 0, 0);
wr = (struct work_request_hdr *)req;
wr++;
req = (struct cpl_set_tcb_field *)wr;
mk_set_tcb_ulp(f, req, TCB_RSS_INFO_W, TCB_RSS_INFO_V(TCB_RSS_INFO_M),
TCB_RSS_INFO_V(adapter->sge.fw_evtq.abs_id), 0, 1);
aligner = (struct ulptx_idata *)(req + 1);
abort_req = (struct cpl_abort_req *)(aligner + 1);
mk_abort_req_ulp(abort_req, f->tid);
abort_rpl = (struct cpl_abort_rpl *)(abort_req + 1);
mk_abort_rpl_ulp(abort_rpl, f->tid);
t4_ofld_send(adapter, skb);
return 0;
}
void hash_del_filter_rpl(struct adapter *adap,
const struct cpl_abort_rpl_rss *rpl)
{
unsigned int status = rpl->status;
struct tid_info *t = &adap->tids;
unsigned int tid = GET_TID(rpl);
struct filter_ctx *ctx = NULL;
struct filter_entry *f;
dev_dbg(adap->pdev_dev, "%s: status = %u; tid = %u\n",
__func__, status, tid);
f = lookup_tid(t, tid);
if (!f) {
dev_err(adap->pdev_dev, "%s:could not find filter entry",
__func__);
return;
}
ctx = f->ctx;
f->ctx = NULL;
clear_filter(adap, f);
cxgb4_remove_tid(t, 0, tid, 0);
kfree(f);
if (ctx) {
ctx->result = 0;
complete(&ctx->completion);
}
}
static int
do_rx_data_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_rx_data_ddp *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
uint32_t vld;
struct toepcb *toep = lookup_tid(sc, tid);
struct tom_data *td = toep->td;
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
KASSERT(!(toep->flags & TPF_SYNQE),
("%s: toep %p claims to be a synq entry", __func__, toep));
vld = be32toh(cpl->ddpvld);
if (__predict_false(vld & DDP_ERR)) {
panic("%s: DDP error 0x%x (tid %d, toep %p)",
__func__, vld, tid, toep);
}
if (toep->ulp_mode == ULP_MODE_ISCSI) {
m = m_get(M_NOWAIT, MT_DATA);
if (m == NULL)
CXGBE_UNIMPLEMENTED("mbuf alloc failure");
memcpy(mtod(m, unsigned char *), cpl,
sizeof(struct cpl_rx_data_ddp));
if (!t4_cpl_iscsi_callback(td, toep, m, CPL_RX_DATA_DDP))
return (0);
m_freem(m);
}
static int
do_rx_iscsi_hdr(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
struct cpl_iscsi_hdr *cpl = mtod(m, struct cpl_iscsi_hdr *);
u_int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct icl_pdu *ip;
struct icl_cxgbei_pdu *icp;
M_ASSERTPKTHDR(m);
ip = icl_cxgbei_new_pdu(M_NOWAIT);
if (ip == NULL)
CXGBE_UNIMPLEMENTED("PDU allocation failure");
icp = ip_to_icp(ip);
bcopy(mtod(m, caddr_t) + sizeof(*cpl), icp->ip.ip_bhs, sizeof(struct
iscsi_bhs));
icp->pdu_seq = ntohl(cpl->seq);
icp->pdu_flags = SBUF_ULP_FLAG_HDR_RCVD;
/* This is the start of a new PDU. There should be no old state. */
MPASS(toep->ulpcb2 == NULL);
toep->ulpcb2 = icp;
#if 0
CTR4(KTR_CXGBE, "%s: tid %u, cpl->len hlen %u, m->m_len hlen %u",
__func__, tid, ntohs(cpl->len), m->m_len);
#endif
m_freem(m);
return (0);
}
static int
do_rx_data_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_rx_data_ddp *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
uint32_t vld;
struct toepcb *toep = lookup_tid(sc, tid);
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
KASSERT(!(toep->flags & TPF_SYNQE),
("%s: toep %p claims to be a synq entry", __func__, toep));
vld = be32toh(cpl->ddpvld);
if (__predict_false(vld & DDP_ERR)) {
panic("%s: DDP error 0x%x (tid %d, toep %p)",
__func__, vld, tid, toep);
}
if (toep->ulp_mode == ULP_MODE_ISCSI) {
sc->cpl_handler[CPL_RX_ISCSI_DDP](iq, rss, m);
return (0);
}
handle_ddp_data(toep, cpl->u.ddp_report, cpl->seq, be16toh(cpl->len));
return (0);
}
static int
do_rx_ddp_complete(struct sge_iq *iq, const struct rss_header *rss,
struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_rx_ddp_complete *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__));
KASSERT(!(toep->flags & TPF_SYNQE),
("%s: toep %p claims to be a synq entry", __func__, toep));
handle_ddp_data(toep, cpl->ddp_report, cpl->rcv_nxt, 0);
return (0);
}
static int
do_term(struct t3cdev *dev, struct mbuf *m)
{
unsigned int hwtid = ntohl(m_get_priority(m)) >> 8 & 0xfffff;
unsigned int opcode = G_OPCODE(ntohl(m->m_pkthdr.csum_data));
struct toe_tid_entry *toe_tid;
toe_tid = lookup_tid(&(T3C_DATA (dev))->tid_maps, hwtid);
if (toe_tid && toe_tid->ctx && toe_tid->client->handlers &&
toe_tid->client->handlers[opcode]) {
return toe_tid->client->handlers[opcode](dev, m, toe_tid->ctx);
} else {
log(LOG_ERR, "%s: received clientless CPL command 0x%x\n",
dev->name, opcode);
return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
}
return (0);
}
static int
do_hwtid_rpl(struct t3cdev *dev, struct mbuf *m)
{
union opcode_tid *p = cplhdr(m);
unsigned int hwtid;
struct toe_tid_entry *toe_tid;
DPRINTF("do_hwtid_rpl opcode=0x%x\n", p->opcode);
hwtid = G_TID(ntohl(p->opcode_tid));
toe_tid = lookup_tid(&(T3C_DATA (dev))->tid_maps, hwtid);
if (toe_tid->ctx && toe_tid->client->handlers &&
toe_tid->client->handlers[p->opcode]) {
return toe_tid->client->handlers[p->opcode]
(dev, m, toe_tid->ctx);
} else {
log(LOG_ERR, "%s: received clientless CPL command 0x%x\n",
dev->name, p->opcode);
return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
}
}
int
do_set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_set_tcb_rpl *cpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep;
#ifdef INVARIANTS
unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl)));
#endif
KASSERT(opcode == CPL_SET_TCB_RPL,
("%s: unexpected opcode 0x%x", __func__, opcode));
KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__));
MPASS(iq != &sc->sge.fwq);
toep = lookup_tid(sc, tid);
if (toep->ulp_mode == ULP_MODE_TCPDDP) {
handle_ddp_tcb_rpl(toep, cpl);
return (0);
}
/*
* TOM and/or other ULPs don't request replies for CPL_SET_TCB or
* CPL_SET_TCB_FIELD requests. This can easily change and when it does
* the dispatch code will go here.
*/
#ifdef INVARIANTS
panic("%s: Unexpected CPL_SET_TCB_RPL for tid %u on iq %p", __func__,
tid, iq);
#else
log(LOG_ERR, "%s: Unexpected CPL_SET_TCB_RPL for tid %u on iq %p\n",
__func__, tid, iq);
#endif
return (0);
}
static int
do_rx_iscsi_hdr(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
struct cpl_iscsi_hdr *cpl = mtod(m, struct cpl_iscsi_hdr *);
u_int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct icl_pdu *ip;
struct icl_cxgbei_pdu *icp;
uint16_t len_ddp = be16toh(cpl->pdu_len_ddp);
uint16_t len = be16toh(cpl->len);
M_ASSERTPKTHDR(m);
MPASS(m->m_pkthdr.len == len + sizeof(*cpl));
ip = icl_cxgbei_new_pdu(M_NOWAIT);
if (ip == NULL)
CXGBE_UNIMPLEMENTED("PDU allocation failure");
m_copydata(m, sizeof(*cpl), ISCSI_BHS_SIZE, (caddr_t)ip->ip_bhs);
ip->ip_data_len = G_ISCSI_PDU_LEN(len_ddp) - len;
icp = ip_to_icp(ip);
icp->icp_seq = ntohl(cpl->seq);
icp->icp_flags = ICPF_RX_HDR;
/* This is the start of a new PDU. There should be no old state. */
MPASS(toep->ulpcb2 == NULL);
toep->ulpcb2 = icp;
#if 0
CTR5(KTR_CXGBE, "%s: tid %u, cpl->len %u, pdu_len_ddp 0x%04x, icp %p",
__func__, tid, len, len_ddp, icp);
#endif
m_freem(m);
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 (plen > 0) {
struct sockbuf *sb = &so->so_snd;
SOCKBUF_LOCK(sb);
sbdrop_locked(sb, plen);
sowwakeup_locked(so);
SOCKBUF_UNLOCK_ASSERT(sb);
}
/* XXX */
if ((toep->flags & TPF_TX_SUSPENDED &&
toep->tx_credits >= MIN_OFLD_TX_CREDITS) ||
toep->tx_credits == toep->txsd_total *
howmany((sizeof(struct fw_ofld_tx_data_wr) + 1), 16)) {
toep->flags &= ~TPF_TX_SUSPENDED;
t4_push_frames(sc, toep);
}
INP_WUNLOCK(inp);
return (0);
}
static int
do_rx_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_rx_data *cpl = mtod(m, const void *);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp = toep->inp;
struct tcpcb *tp;
struct socket *so;
struct sockbuf *sb;
int len;
uint32_t ddp_placed = 0;
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);
m_freem(m);
return (0);
}
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
/* strip off CPL header */
m_adj(m, sizeof(*cpl));
len = m->m_pkthdr.len;
INP_WLOCK(inp);
if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) {
CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x",
__func__, tid, len, inp->inp_flags);
INP_WUNLOCK(inp);
m_freem(m);
return (0);
}
tp = intotcpcb(inp);
if (__predict_false(tp->rcv_nxt != be32toh(cpl->seq)))
ddp_placed = be32toh(cpl->seq) - tp->rcv_nxt;
tp->rcv_nxt += len;
if (tp->rcv_wnd < len) {
KASSERT(toep->ulp_mode == ULP_MODE_RDMA,
("%s: negative window size", __func__));
}
tp->rcv_wnd -= len;
tp->t_rcvtime = ticks;
if (toep->ulp_mode == ULP_MODE_TCPDDP)
DDP_LOCK(toep);
so = inp_inpcbtosocket(inp);
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) {
CTR3(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes)",
__func__, tid, len);
m_freem(m);
SOCKBUF_UNLOCK(sb);
if (toep->ulp_mode == ULP_MODE_TCPDDP)
DDP_UNLOCK(toep);
INP_WUNLOCK(inp);
INP_INFO_RLOCK(&V_tcbinfo);
INP_WLOCK(inp);
tp = tcp_drop(tp, ECONNRESET);
if (tp)
INP_WUNLOCK(inp);
INP_INFO_RUNLOCK(&V_tcbinfo);
return (0);
}
/* receive buffer autosize */
CURVNET_SET(so->so_vnet);
if (sb->sb_flags & SB_AUTOSIZE &&
V_tcp_do_autorcvbuf &&
sb->sb_hiwat < V_tcp_autorcvbuf_max &&
len > (sbspace(sb) / 8 * 7)) {
unsigned int hiwat = sb->sb_hiwat;
unsigned int newsize = min(hiwat + V_tcp_autorcvbuf_inc,
V_tcp_autorcvbuf_max);
if (!sbreserve_locked(sb, newsize, so, NULL))
sb->sb_flags &= ~SB_AUTOSIZE;
else
toep->rx_credits += newsize - hiwat;
}
if (toep->ddp_waiting_count != 0 || toep->ddp_active_count != 0)
CTR3(KTR_CXGBE, "%s: tid %u, non-ddp rx (%d bytes)", __func__,
tid, len);
if (toep->ulp_mode == ULP_MODE_TCPDDP) {
int changed = !(toep->ddp_flags & DDP_ON) ^ cpl->ddp_off;
if (changed) {
if (toep->ddp_flags & DDP_SC_REQ)
toep->ddp_flags ^= DDP_ON | DDP_SC_REQ;
else {
KASSERT(cpl->ddp_off == 1,
("%s: DDP switched on by itself.",
__func__));
/* Fell out of DDP mode */
toep->ddp_flags &= ~DDP_ON;
CTR1(KTR_CXGBE, "%s: fell out of DDP mode",
__func__);
insert_ddp_data(toep, ddp_placed);
}
}
if (toep->ddp_flags & DDP_ON) {
/*
* CPL_RX_DATA with DDP on can only be an indicate.
* Start posting queued AIO requests via DDP. The
* payload that arrived in this indicate is appended
* to the socket buffer as usual.
*/
handle_ddp_indicate(toep);
}
}
KASSERT(toep->sb_cc >= sbused(sb),
("%s: sb %p has more data (%d) than last time (%d).",
__func__, sb, sbused(sb), toep->sb_cc));
toep->rx_credits += toep->sb_cc - sbused(sb);
sbappendstream_locked(sb, m, 0);
toep->sb_cc = sbused(sb);
if (toep->rx_credits > 0 && toep->sb_cc + tp->rcv_wnd < sb->sb_lowat) {
int credits;
credits = send_rx_credits(sc, toep, toep->rx_credits);
toep->rx_credits -= credits;
tp->rcv_wnd += credits;
tp->rcv_adv += credits;
}
if (toep->ddp_waiting_count > 0 && sbavail(sb) != 0) {
CTR2(KTR_CXGBE, "%s: tid %u queueing AIO task", __func__,
tid);
ddp_queue_toep(toep);
}
sorwakeup_locked(so);
SOCKBUF_UNLOCK_ASSERT(sb);
if (toep->ulp_mode == ULP_MODE_TCPDDP)
DDP_UNLOCK(toep);
INP_WUNLOCK(inp);
CURVNET_RESTORE();
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);
}
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;
if (toep->ulp_mode == ULP_MODE_ISCSI)
t4_push_pdus(sc, toep, plen);
else
t4_push_frames(sc, toep, plen);
} else if (plen > 0) {
struct sockbuf *sb = &so->so_snd;
int sbu;
SOCKBUF_LOCK(sb);
sbu = sbused(sb);
if (toep->ulp_mode == ULP_MODE_ISCSI) {
if (__predict_false(sbu > 0)) {
/*
* The data trasmitted before the tid's ULP mode
* changed to ISCSI is still in so_snd.
* Incoming credits should account for so_snd
* first.
*/
sbdrop_locked(sb, min(sbu, plen));
plen -= min(sbu, plen);
}
sowwakeup_locked(so); /* unlocks so_snd */
rqdrop_locked(&toep->ulp_pdu_reclaimq, plen);
} else {
sbdrop_locked(sb, plen);
sowwakeup_locked(so); /* unlocks so_snd */
}
SOCKBUF_UNLOCK_ASSERT(sb);
}
INP_WUNLOCK(inp);
return (0);
}
/**
* Delete the specified hash filter.
*/
static int cxgbe_del_hash_filter(struct rte_eth_dev *dev,
unsigned int filter_id,
struct filter_ctx *ctx)
{
struct adapter *adapter = ethdev2adap(dev);
struct tid_info *t = &adapter->tids;
struct filter_entry *f;
struct sge_ctrl_txq *ctrlq;
unsigned int port_id = ethdev2pinfo(dev)->port_id;
int ret;
if (filter_id > adapter->tids.ntids)
return -E2BIG;
f = lookup_tid(t, filter_id);
if (!f) {
dev_err(adapter, "%s: no filter entry for filter_id = %d\n",
__func__, filter_id);
return -EINVAL;
}
ret = writable_filter(f);
if (ret)
return ret;
if (f->valid) {
unsigned int wrlen;
struct rte_mbuf *mbuf;
struct work_request_hdr *wr;
struct ulptx_idata *aligner;
struct cpl_set_tcb_field *req;
struct cpl_abort_req *abort_req;
struct cpl_abort_rpl *abort_rpl;
f->ctx = ctx;
f->pending = 1;
wrlen = cxgbe_roundup(sizeof(*wr) +
(sizeof(*req) + sizeof(*aligner)) +
sizeof(*abort_req) + sizeof(*abort_rpl),
16);
ctrlq = &adapter->sge.ctrlq[port_id];
mbuf = rte_pktmbuf_alloc(ctrlq->mb_pool);
if (!mbuf) {
dev_err(adapter, "%s: could not allocate skb ..\n",
__func__);
goto out_err;
}
mbuf->data_len = wrlen;
mbuf->pkt_len = mbuf->data_len;
req = rte_pktmbuf_mtod(mbuf, struct cpl_set_tcb_field *);
INIT_ULPTX_WR(req, wrlen, 0, 0);
wr = (struct work_request_hdr *)req;
wr++;
req = (struct cpl_set_tcb_field *)wr;
mk_set_tcb_field_ulp(f, req, W_TCB_RSS_INFO,
V_TCB_RSS_INFO(M_TCB_RSS_INFO),
V_TCB_RSS_INFO(adapter->sge.fw_evtq.abs_id),
0, 1);
aligner = (struct ulptx_idata *)(req + 1);
abort_req = (struct cpl_abort_req *)(aligner + 1);
mk_abort_req_ulp(abort_req, f->tid);
abort_rpl = (struct cpl_abort_rpl *)(abort_req + 1);
mk_abort_rpl_ulp(abort_rpl, f->tid);
t4_mgmt_tx(ctrlq, mbuf);
}
static int
do_rx_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_rx_data *cpl = mtod(m, const void *);
unsigned int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp = toep->inp;
struct tcpcb *tp;
struct socket *so;
struct sockbuf *sb;
int len;
uint32_t ddp_placed = 0;
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);
m_freem(m);
return (0);
}
KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__));
/* strip off CPL header */
m_adj(m, sizeof(*cpl));
len = m->m_pkthdr.len;
INP_WLOCK(inp);
if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) {
CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x",
__func__, tid, len, inp->inp_flags);
INP_WUNLOCK(inp);
m_freem(m);
return (0);
}
tp = intotcpcb(inp);
if (__predict_false(tp->rcv_nxt != be32toh(cpl->seq)))
ddp_placed = be32toh(cpl->seq) - tp->rcv_nxt;
tp->rcv_nxt += len;
KASSERT(tp->rcv_wnd >= len, ("%s: negative window size", __func__));
tp->rcv_wnd -= len;
tp->t_rcvtime = ticks;
so = inp_inpcbtosocket(inp);
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) {
CTR3(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes)",
__func__, tid, len);
m_freem(m);
SOCKBUF_UNLOCK(sb);
INP_WUNLOCK(inp);
INP_INFO_WLOCK(&V_tcbinfo);
INP_WLOCK(inp);
tp = tcp_drop(tp, ECONNRESET);
if (tp)
INP_WUNLOCK(inp);
INP_INFO_WUNLOCK(&V_tcbinfo);
return (0);
}
/* receive buffer autosize */
if (sb->sb_flags & SB_AUTOSIZE &&
V_tcp_do_autorcvbuf &&
sb->sb_hiwat < V_tcp_autorcvbuf_max &&
len > (sbspace(sb) / 8 * 7)) {
unsigned int hiwat = sb->sb_hiwat;
unsigned int newsize = min(hiwat + V_tcp_autorcvbuf_inc,
V_tcp_autorcvbuf_max);
if (!sbreserve_locked(sb, newsize, so, NULL))
sb->sb_flags &= ~SB_AUTOSIZE;
else
toep->rx_credits += newsize - hiwat;
}
if (toep->ulp_mode == ULP_MODE_TCPDDP) {
int changed = !(toep->ddp_flags & DDP_ON) ^ cpl->ddp_off;
if (changed) {
if (toep->ddp_flags & DDP_SC_REQ)
toep->ddp_flags ^= DDP_ON | DDP_SC_REQ;
else {
KASSERT(cpl->ddp_off == 1,
("%s: DDP switched on by itself.",
__func__));
/* Fell out of DDP mode */
toep->ddp_flags &= ~(DDP_ON | DDP_BUF0_ACTIVE |
DDP_BUF1_ACTIVE);
if (ddp_placed)
insert_ddp_data(toep, ddp_placed);
}
}
if ((toep->ddp_flags & DDP_OK) == 0 &&
time_uptime >= toep->ddp_disabled + DDP_RETRY_WAIT) {
toep->ddp_score = DDP_LOW_SCORE;
toep->ddp_flags |= DDP_OK;
CTR3(KTR_CXGBE, "%s: tid %u DDP_OK @ %u",
__func__, tid, time_uptime);
}
if (toep->ddp_flags & DDP_ON) {
/*
* CPL_RX_DATA with DDP on can only be an indicate. Ask
* soreceive to post a buffer or disable DDP. The
* payload that arrived in this indicate is appended to
* the socket buffer as usual.
*/
#if 0
CTR5(KTR_CXGBE,
"%s: tid %u (0x%x) DDP indicate (seq 0x%x, len %d)",
__func__, tid, toep->flags, be32toh(cpl->seq), len);
#endif
sb->sb_flags |= SB_DDP_INDICATE;
} else if ((toep->ddp_flags & (DDP_OK|DDP_SC_REQ)) == DDP_OK &&
tp->rcv_wnd > DDP_RSVD_WIN && len >= sc->tt.ddp_thres) {
/*
* DDP allowed but isn't on (and a request to switch it
* on isn't pending either), and conditions are ripe for
* it to work. Switch it on.
*/
enable_ddp(sc, toep);
}
}
KASSERT(toep->sb_cc >= sb->sb_cc,
("%s: sb %p has more data (%d) than last time (%d).",
__func__, sb, sb->sb_cc, toep->sb_cc));
toep->rx_credits += toep->sb_cc - sb->sb_cc;
sbappendstream_locked(sb, m);
toep->sb_cc = sb->sb_cc;
sorwakeup_locked(so);
SOCKBUF_UNLOCK_ASSERT(sb);
INP_WUNLOCK(inp);
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_WLOCK(&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_WUNLOCK(&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_WUNLOCK(&V_tcbinfo);
return (0);
}
/*
* 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_WLOCK(&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_WUNLOCK(&V_tcbinfo);
send_abort_rpl(sc, ofld_txq, tid, CPL_ABORT_NO_RST);
return (0);
}
static int
do_rx_iscsi_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
struct cxgbei_data *ci = sc->iscsi_ulp_softc;
const struct cpl_rx_data_ddp *cpl = (const void *)(rss + 1);
u_int tid = GET_TID(cpl);
struct toepcb *toep = lookup_tid(sc, tid);
struct inpcb *inp = toep->inp;
struct socket *so;
struct sockbuf *sb;
struct tcpcb *tp;
struct icl_cxgbei_conn *icc;
struct icl_conn *ic;
struct icl_cxgbei_pdu *icp = toep->ulpcb2;
struct icl_pdu *ip;
u_int pdu_len, val;
MPASS(m == NULL);
/* Must already be assembling a PDU. */
MPASS(icp != NULL);
MPASS(icp->icp_flags & ICPF_RX_HDR); /* Data is optional. */
MPASS((icp->icp_flags & ICPF_RX_STATUS) == 0);
pdu_len = be16toh(cpl->len); /* includes everything. */
val = be32toh(cpl->ddpvld);
#if 0
CTR5(KTR_CXGBE,
"%s: tid %u, cpl->len %u, ddpvld 0x%08x, icp_flags 0x%08x",
__func__, tid, pdu_len, val, icp->icp_flags);
#endif
icp->icp_flags |= ICPF_RX_STATUS;
ip = &icp->ip;
if (val & F_DDP_PADDING_ERR)
icp->icp_flags |= ICPF_PAD_ERR;
if (val & F_DDP_HDRCRC_ERR)
icp->icp_flags |= ICPF_HCRC_ERR;
if (val & F_DDP_DATACRC_ERR)
icp->icp_flags |= ICPF_DCRC_ERR;
if (val & F_DDP_PDU && ip->ip_data_mbuf == NULL) {
MPASS((icp->icp_flags & ICPF_RX_FLBUF) == 0);
MPASS(ip->ip_data_len > 0);
icp->icp_flags |= ICPF_RX_DDP;
counter_u64_add(ci->ddp_pdus, 1);
counter_u64_add(ci->ddp_bytes, ip->ip_data_len);
}
INP_WLOCK(inp);
if (__predict_false(inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))) {
CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x",
__func__, tid, pdu_len, inp->inp_flags);
INP_WUNLOCK(inp);
icl_cxgbei_conn_pdu_free(NULL, ip);
#ifdef INVARIANTS
toep->ulpcb2 = NULL;
#endif
return (0);
}
tp = intotcpcb(inp);
MPASS(icp->icp_seq == tp->rcv_nxt);
MPASS(tp->rcv_wnd >= pdu_len);
tp->rcv_nxt += pdu_len;
tp->rcv_wnd -= pdu_len;
tp->t_rcvtime = ticks;
/* update rx credits */
toep->rx_credits += pdu_len;
t4_rcvd(&toep->td->tod, tp); /* XXX: sc->tom_softc.tod */
so = inp->inp_socket;
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
icc = toep->ulpcb;
if (__predict_false(icc == NULL || sb->sb_state & SBS_CANTRCVMORE)) {
CTR5(KTR_CXGBE,
"%s: tid %u, excess rx (%d bytes), icc %p, sb_state 0x%x",
__func__, tid, pdu_len, icc, sb->sb_state);
SOCKBUF_UNLOCK(sb);
INP_WUNLOCK(inp);
INP_INFO_RLOCK(&V_tcbinfo);
INP_WLOCK(inp);
tp = tcp_drop(tp, ECONNRESET);
if (tp)
INP_WUNLOCK(inp);
INP_INFO_RUNLOCK(&V_tcbinfo);
icl_cxgbei_conn_pdu_free(NULL, ip);
#ifdef INVARIANTS
toep->ulpcb2 = NULL;
#endif
return (0);
}
MPASS(icc->icc_signature == CXGBEI_CONN_SIGNATURE);
ic = &icc->ic;
icl_cxgbei_new_pdu_set_conn(ip, ic);
MPASS(m == NULL); /* was unused, we'll use it now. */
m = sbcut_locked(sb, sbused(sb)); /* XXXNP: toep->sb_cc accounting? */
if (__predict_false(m != NULL)) {
int len = m_length(m, NULL);
/*
* PDUs were received before the tid transitioned to ULP mode.
* Convert them to icl_cxgbei_pdus and send them to ICL before
* the PDU in icp/ip.
*/
CTR3(KTR_CXGBE, "%s: tid %u, %u bytes in so_rcv", __func__, tid,
len);
/* XXXNP: needs to be rewritten. */
if (len == sizeof(struct iscsi_bhs) || len == 4 + sizeof(struct
iscsi_bhs)) {
struct icl_cxgbei_pdu *icp0;
struct icl_pdu *ip0;
ip0 = icl_cxgbei_new_pdu(M_NOWAIT);
icl_cxgbei_new_pdu_set_conn(ip0, ic);
if (ip0 == NULL)
CXGBE_UNIMPLEMENTED("PDU allocation failure");
icp0 = ip_to_icp(ip0);
icp0->icp_seq = 0; /* XXX */
icp0->icp_flags = ICPF_RX_HDR | ICPF_RX_STATUS;
m_copydata(m, 0, sizeof(struct iscsi_bhs), (void *)ip0->ip_bhs);
STAILQ_INSERT_TAIL(&icc->rcvd_pdus, ip0, ip_next);
}
m_freem(m);
}
STAILQ_INSERT_TAIL(&icc->rcvd_pdus, ip, ip_next);
if ((icc->rx_flags & RXF_ACTIVE) == 0) {
struct cxgbei_worker_thread_softc *cwt = &cwt_softc[icc->cwt];
mtx_lock(&cwt->cwt_lock);
icc->rx_flags |= RXF_ACTIVE;
TAILQ_INSERT_TAIL(&cwt->rx_head, icc, rx_link);
if (cwt->cwt_state == CWT_SLEEPING) {
cwt->cwt_state = CWT_RUNNING;
cv_signal(&cwt->cwt_cv);
}
mtx_unlock(&cwt->cwt_lock);
}
SOCKBUF_UNLOCK(sb);
INP_WUNLOCK(inp);
#ifdef INVARIANTS
toep->ulpcb2 = NULL;
#endif
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;
struct sockbuf *sb;
#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_WLOCK(&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;
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
if (__predict_false(toep->ddp_flags & (DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE))) {
m = m_get(M_NOWAIT, MT_DATA);
if (m == NULL)
CXGBE_UNIMPLEMENTED("mbuf alloc failure");
m->m_len = be32toh(cpl->rcv_nxt) - tp->rcv_nxt;
m->m_flags |= M_DDP; /* Data is already where it should be */
m->m_data = "nothing to see here";
tp->rcv_nxt = be32toh(cpl->rcv_nxt);
toep->ddp_flags &= ~(DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE);
KASSERT(toep->sb_cc >= sb->sb_cc,
("%s: sb %p has more data (%d) than last time (%d).",
__func__, sb, sb->sb_cc, toep->sb_cc));
toep->rx_credits += toep->sb_cc - sb->sb_cc;
#ifdef USE_DDP_RX_FLOW_CONTROL
toep->rx_credits -= m->m_len; /* adjust for F_RX_FC_DDP */
#endif
sbappendstream_locked(sb, m);
toep->sb_cc = sb->sb_cc;
}
socantrcvmore_locked(so); /* unlocks the sockbuf */
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_WUNLOCK(&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_WUNLOCK(&V_tcbinfo);
return (0);
}
