diff --git a/sys/dev/cxgbe/tom/t4_tom.c b/sys/dev/cxgbe/tom/t4_tom.c
index cb2131a4c0c2..c5a7e9290666 100644
--- a/sys/dev/cxgbe/tom/t4_tom.c
+++ b/sys/dev/cxgbe/tom/t4_tom.c
@@ -1,2071 +1,2071 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2012 Chelsio Communications, Inc.
* All rights reserved.
* Written by: Navdeep Parhar <np@FreeBSD.org>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_kern_tls.h"
#include "opt_ratelimit.h"
#include <sys/param.h>
#include <sys/types.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/lock.h>
#include <sys/limits.h>
#include <sys/module.h>
#include <sys/protosw.h>
#include <sys/domain.h>
#include <sys/refcount.h>
#include <sys/rmlock.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <netinet/in.h>
#include <netinet/in_pcb.h>
#include <netinet/in_var.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#include <netinet6/scope6_var.h>
#define TCPSTATES
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_seq.h>
#include <netinet/tcp_timer.h>
#include <netinet/tcp_var.h>
#include <netinet/toecore.h>
#include <netinet/cc/cc.h>
#ifdef TCP_OFFLOAD
#include "common/common.h"
#include "common/t4_msg.h"
#include "common/t4_regs.h"
#include "common/t4_regs_values.h"
#include "common/t4_tcb.h"
#include "t4_clip.h"
#include "tom/t4_tom_l2t.h"
#include "tom/t4_tom.h"
#include "tom/t4_tls.h"
static struct protosw toe_protosw;
static struct protosw toe6_protosw;
/* Module ops */
static int t4_tom_mod_load(void);
static int t4_tom_mod_unload(void);
static int t4_tom_modevent(module_t, int, void *);
/* ULD ops and helpers */
static int t4_tom_activate(struct adapter *);
static int t4_tom_deactivate(struct adapter *);
static struct uld_info tom_uld_info = {
.uld_id = ULD_TOM,
.activate = t4_tom_activate,
.deactivate = t4_tom_deactivate,
};
static void release_offload_resources(struct toepcb *);
static int alloc_tid_tabs(struct tid_info *);
static void free_tid_tabs(struct tid_info *);
static void free_tom_data(struct adapter *, struct tom_data *);
static void reclaim_wr_resources(void *, int);
struct toepcb *
alloc_toepcb(struct vi_info *vi, int flags)
{
struct port_info *pi = vi->pi;
struct adapter *sc = pi->adapter;
struct toepcb *toep;
int tx_credits, txsd_total, len;
/*
* The firmware counts tx work request credits in units of 16 bytes
* each. Reserve room for an ABORT_REQ so the driver never has to worry
* about tx credits if it wants to abort a connection.
*/
tx_credits = sc->params.ofldq_wr_cred;
tx_credits -= howmany(sizeof(struct cpl_abort_req), 16);
/*
* Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte
* immediate payload, and firmware counts tx work request credits in
* units of 16 byte. Calculate the maximum work requests possible.
*/
txsd_total = tx_credits /
howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16);
len = offsetof(struct toepcb, txsd) +
txsd_total * sizeof(struct ofld_tx_sdesc);
toep = malloc(len, M_CXGBE, M_ZERO | flags);
if (toep == NULL)
return (NULL);
refcount_init(&toep->refcount, 1);
toep->td = sc->tom_softc;
toep->vi = vi;
toep->tid = -1;
toep->tx_total = tx_credits;
toep->tx_credits = tx_credits;
mbufq_init(&toep->ulp_pduq, INT_MAX);
mbufq_init(&toep->ulp_pdu_reclaimq, INT_MAX);
toep->txsd_total = txsd_total;
toep->txsd_avail = txsd_total;
toep->txsd_pidx = 0;
toep->txsd_cidx = 0;
aiotx_init_toep(toep);
return (toep);
}
/*
* Initialize a toepcb after its params have been filled out.
*/
int
init_toepcb(struct vi_info *vi, struct toepcb *toep)
{
struct conn_params *cp = &toep->params;
struct port_info *pi = vi->pi;
struct adapter *sc = pi->adapter;
struct tx_cl_rl_params *tc;
if (cp->tc_idx >= 0 && cp->tc_idx < sc->params.nsched_cls) {
tc = &pi->sched_params->cl_rl[cp->tc_idx];
mtx_lock(&sc->tc_lock);
if (tc->state != CS_HW_CONFIGURED) {
CH_ERR(vi, "tid %d cannot be bound to traffic class %d "
"because it is not configured (its state is %d)\n",
toep->tid, cp->tc_idx, tc->state);
cp->tc_idx = -1;
} else {
tc->refcount++;
}
mtx_unlock(&sc->tc_lock);
}
toep->ofld_txq = &sc->sge.ofld_txq[cp->txq_idx];
toep->ofld_rxq = &sc->sge.ofld_rxq[cp->rxq_idx];
toep->ctrlq = &sc->sge.ctrlq[pi->port_id];
tls_init_toep(toep);
if (ulp_mode(toep) == ULP_MODE_TCPDDP)
ddp_init_toep(toep);
toep->flags |= TPF_INITIALIZED;
return (0);
}
struct toepcb *
hold_toepcb(struct toepcb *toep)
{
refcount_acquire(&toep->refcount);
return (toep);
}
void
free_toepcb(struct toepcb *toep)
{
if (refcount_release(&toep->refcount) == 0)
return;
KASSERT(!(toep->flags & TPF_ATTACHED),
("%s: attached to an inpcb", __func__));
KASSERT(!(toep->flags & TPF_CPL_PENDING),
("%s: CPL pending", __func__));
if (toep->flags & TPF_INITIALIZED) {
if (ulp_mode(toep) == ULP_MODE_TCPDDP)
ddp_uninit_toep(toep);
tls_uninit_toep(toep);
}
free(toep, M_CXGBE);
}
/*
* Set up the socket for TCP offload.
*/
void
offload_socket(struct socket *so, struct toepcb *toep)
{
struct tom_data *td = toep->td;
struct inpcb *inp = sotoinpcb(so);
struct tcpcb *tp = intotcpcb(inp);
struct sockbuf *sb;
INP_WLOCK_ASSERT(inp);
/* Update socket */
sb = &so->so_snd;
SOCKBUF_LOCK(sb);
sb->sb_flags |= SB_NOCOALESCE;
SOCKBUF_UNLOCK(sb);
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
sb->sb_flags |= SB_NOCOALESCE;
if (inp->inp_vflag & INP_IPV6)
so->so_proto = &toe6_protosw;
else
so->so_proto = &toe_protosw;
SOCKBUF_UNLOCK(sb);
/* Update TCP PCB */
tp->tod = &td->tod;
tp->t_toe = toep;
tp->t_flags |= TF_TOE;
/* Install an extra hold on inp */
toep->inp = inp;
toep->flags |= TPF_ATTACHED;
in_pcbref(inp);
/* Add the TOE PCB to the active list */
mtx_lock(&td->toep_list_lock);
TAILQ_INSERT_HEAD(&td->toep_list, toep, link);
mtx_unlock(&td->toep_list_lock);
}
void
restore_so_proto(struct socket *so, bool v6)
{
if (v6)
so->so_proto = &tcp6_protosw;
else
so->so_proto = &tcp_protosw;
}
/* This is _not_ the normal way to "unoffload" a socket. */
void
undo_offload_socket(struct socket *so)
{
struct inpcb *inp = sotoinpcb(so);
struct tcpcb *tp = intotcpcb(inp);
struct toepcb *toep = tp->t_toe;
struct tom_data *td = toep->td;
struct sockbuf *sb;
INP_WLOCK_ASSERT(inp);
sb = &so->so_snd;
SOCKBUF_LOCK(sb);
sb->sb_flags &= ~SB_NOCOALESCE;
SOCKBUF_UNLOCK(sb);
sb = &so->so_rcv;
SOCKBUF_LOCK(sb);
sb->sb_flags &= ~SB_NOCOALESCE;
restore_so_proto(so, inp->inp_vflag & INP_IPV6);
SOCKBUF_UNLOCK(sb);
tp->tod = NULL;
tp->t_toe = NULL;
tp->t_flags &= ~TF_TOE;
toep->inp = NULL;
toep->flags &= ~TPF_ATTACHED;
if (in_pcbrele_wlocked(inp))
panic("%s: inp freed.", __func__);
mtx_lock(&td->toep_list_lock);
TAILQ_REMOVE(&td->toep_list, toep, link);
mtx_unlock(&td->toep_list_lock);
}
static void
release_offload_resources(struct toepcb *toep)
{
struct tom_data *td = toep->td;
struct adapter *sc = td_adapter(td);
int tid = toep->tid;
KASSERT(!(toep->flags & TPF_CPL_PENDING),
("%s: %p has CPL pending.", __func__, toep));
KASSERT(!(toep->flags & TPF_ATTACHED),
("%s: %p is still attached.", __func__, toep));
CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)",
__func__, toep, tid, toep->l2te, toep->ce);
/*
* These queues should have been emptied at approximately the same time
* that a normal connection's socket's so_snd would have been purged or
* drained. Do _not_ clean up here.
*/
MPASS(mbufq_len(&toep->ulp_pduq) == 0);
MPASS(mbufq_len(&toep->ulp_pdu_reclaimq) == 0);
#ifdef INVARIANTS
if (ulp_mode(toep) == ULP_MODE_TCPDDP)
ddp_assert_empty(toep);
#endif
MPASS(TAILQ_EMPTY(&toep->aiotx_jobq));
if (toep->l2te)
t4_l2t_release(toep->l2te);
if (tid >= 0) {
remove_tid(sc, tid, toep->ce ? 2 : 1);
release_tid(sc, tid, toep->ctrlq);
}
if (toep->ce)
t4_release_clip_entry(sc, toep->ce);
if (toep->params.tc_idx != -1)
t4_release_cl_rl(sc, toep->vi->pi->port_id, toep->params.tc_idx);
mtx_lock(&td->toep_list_lock);
TAILQ_REMOVE(&td->toep_list, toep, link);
mtx_unlock(&td->toep_list_lock);
free_toepcb(toep);
}
/*
* The kernel is done with the TCP PCB and this is our opportunity to unhook the
* toepcb hanging off of it. If the TOE driver is also done with the toepcb (no
* pending CPL) then it is time to release all resources tied to the toepcb.
*
* Also gets called when an offloaded active open fails and the TOM wants the
* kernel to take the TCP PCB back.
*/
static void
t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp)
{
#if defined(KTR) || defined(INVARIANTS)
struct inpcb *inp = tptoinpcb(tp);
#endif
struct toepcb *toep = tp->t_toe;
INP_WLOCK_ASSERT(inp);
KASSERT(toep != NULL, ("%s: toep is NULL", __func__));
KASSERT(toep->flags & TPF_ATTACHED,
("%s: not attached", __func__));
#ifdef KTR
if (tp->t_state == TCPS_SYN_SENT) {
CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)",
__func__, toep->tid, toep, toep->flags, inp,
inp->inp_flags);
} else {
CTR6(KTR_CXGBE,
"t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)",
toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp,
inp->inp_flags);
}
#endif
tp->tod = NULL;
tp->t_toe = NULL;
tp->t_flags &= ~TF_TOE;
toep->flags &= ~TPF_ATTACHED;
if (!(toep->flags & TPF_CPL_PENDING))
release_offload_resources(toep);
}
/*
* setsockopt handler.
*/
static void
t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name)
{
struct adapter *sc = tod->tod_softc;
struct toepcb *toep = tp->t_toe;
if (dir == SOPT_GET)
return;
CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name);
switch (name) {
case TCP_NODELAY:
if (tp->t_state != TCPS_ESTABLISHED)
break;
toep->params.nagle = tp->t_flags & TF_NODELAY ? 0 : 1;
t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_T_FLAGS,
V_TF_NAGLE(1), V_TF_NAGLE(toep->params.nagle), 0, 0);
break;
default:
break;
}
}
static inline uint64_t
get_tcb_tflags(const uint64_t *tcb)
{
return ((be64toh(tcb[14]) << 32) | (be64toh(tcb[15]) >> 32));
}
static inline uint32_t
get_tcb_field(const uint64_t *tcb, u_int word, uint32_t mask, u_int shift)
{
#define LAST_WORD ((TCB_SIZE / 4) - 1)
uint64_t t1, t2;
int flit_idx;
MPASS(mask != 0);
MPASS(word <= LAST_WORD);
MPASS(shift < 32);
flit_idx = (LAST_WORD - word) / 2;
if (word & 0x1)
shift += 32;
t1 = be64toh(tcb[flit_idx]) >> shift;
t2 = 0;
if (fls(mask) > 64 - shift) {
/*
* Will spill over into the next logical flit, which is the flit
* before this one. The flit_idx before this one must be valid.
*/
MPASS(flit_idx > 0);
t2 = be64toh(tcb[flit_idx - 1]) << (64 - shift);
}
return ((t2 | t1) & mask);
#undef LAST_WORD
}
#define GET_TCB_FIELD(tcb, F) \
get_tcb_field(tcb, W_TCB_##F, M_TCB_##F, S_TCB_##F)
/*
* Issues a CPL_GET_TCB to read the entire TCB for the tid.
*/
static int
send_get_tcb(struct adapter *sc, u_int tid)
{
struct cpl_get_tcb *cpl;
struct wrq_cookie cookie;
MPASS(tid >= sc->tids.tid_base);
MPASS(tid - sc->tids.tid_base < sc->tids.ntids);
cpl = start_wrq_wr(&sc->sge.ctrlq[0], howmany(sizeof(*cpl), 16),
&cookie);
if (__predict_false(cpl == NULL))
return (ENOMEM);
bzero(cpl, sizeof(*cpl));
INIT_TP_WR(cpl, tid);
OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_GET_TCB, tid));
cpl->reply_ctrl = htobe16(V_REPLY_CHAN(0) |
V_QUEUENO(sc->sge.ofld_rxq[0].iq.cntxt_id));
cpl->cookie = 0xff;
commit_wrq_wr(&sc->sge.ctrlq[0], cpl, &cookie);
return (0);
}
static struct tcb_histent *
alloc_tcb_histent(struct adapter *sc, u_int tid, int flags)
{
struct tcb_histent *te;
MPASS(flags == M_NOWAIT || flags == M_WAITOK);
te = malloc(sizeof(*te), M_CXGBE, M_ZERO | flags);
if (te == NULL)
return (NULL);
mtx_init(&te->te_lock, "TCB entry", NULL, MTX_DEF);
callout_init_mtx(&te->te_callout, &te->te_lock, 0);
te->te_adapter = sc;
te->te_tid = tid;
return (te);
}
static void
free_tcb_histent(struct tcb_histent *te)
{
mtx_destroy(&te->te_lock);
free(te, M_CXGBE);
}
/*
* Start tracking the tid in the TCB history.
*/
int
add_tid_to_history(struct adapter *sc, u_int tid)
{
struct tcb_histent *te = NULL;
struct tom_data *td = sc->tom_softc;
int rc;
MPASS(tid >= sc->tids.tid_base);
MPASS(tid - sc->tids.tid_base < sc->tids.ntids);
if (td->tcb_history == NULL)
return (ENXIO);
rw_wlock(&td->tcb_history_lock);
if (td->tcb_history[tid] != NULL) {
rc = EEXIST;
goto done;
}
te = alloc_tcb_histent(sc, tid, M_NOWAIT);
if (te == NULL) {
rc = ENOMEM;
goto done;
}
mtx_lock(&te->te_lock);
rc = send_get_tcb(sc, tid);
if (rc == 0) {
te->te_flags |= TE_RPL_PENDING;
td->tcb_history[tid] = te;
} else {
free(te, M_CXGBE);
}
mtx_unlock(&te->te_lock);
done:
rw_wunlock(&td->tcb_history_lock);
return (rc);
}
static void
remove_tcb_histent(struct tcb_histent *te)
{
struct adapter *sc = te->te_adapter;
struct tom_data *td = sc->tom_softc;
rw_assert(&td->tcb_history_lock, RA_WLOCKED);
mtx_assert(&te->te_lock, MA_OWNED);
MPASS(td->tcb_history[te->te_tid] == te);
td->tcb_history[te->te_tid] = NULL;
free_tcb_histent(te);
rw_wunlock(&td->tcb_history_lock);
}
static inline struct tcb_histent *
lookup_tcb_histent(struct adapter *sc, u_int tid, bool addrem)
{
struct tcb_histent *te;
struct tom_data *td = sc->tom_softc;
MPASS(tid >= sc->tids.tid_base);
MPASS(tid - sc->tids.tid_base < sc->tids.ntids);
if (td->tcb_history == NULL)
return (NULL);
if (addrem)
rw_wlock(&td->tcb_history_lock);
else
rw_rlock(&td->tcb_history_lock);
te = td->tcb_history[tid];
if (te != NULL) {
mtx_lock(&te->te_lock);
return (te); /* with both locks held */
}
if (addrem)
rw_wunlock(&td->tcb_history_lock);
else
rw_runlock(&td->tcb_history_lock);
return (te);
}
static inline void
release_tcb_histent(struct tcb_histent *te)
{
struct adapter *sc = te->te_adapter;
struct tom_data *td = sc->tom_softc;
mtx_assert(&te->te_lock, MA_OWNED);
mtx_unlock(&te->te_lock);
rw_assert(&td->tcb_history_lock, RA_RLOCKED);
rw_runlock(&td->tcb_history_lock);
}
static void
request_tcb(void *arg)
{
struct tcb_histent *te = arg;
mtx_assert(&te->te_lock, MA_OWNED);
/* Noone else is supposed to update the histent. */
MPASS(!(te->te_flags & TE_RPL_PENDING));
if (send_get_tcb(te->te_adapter, te->te_tid) == 0)
te->te_flags |= TE_RPL_PENDING;
else
callout_schedule(&te->te_callout, hz / 100);
}
static void
update_tcb_histent(struct tcb_histent *te, const uint64_t *tcb)
{
struct tom_data *td = te->te_adapter->tom_softc;
uint64_t tflags = get_tcb_tflags(tcb);
uint8_t sample = 0;
if (GET_TCB_FIELD(tcb, SND_MAX_RAW) != GET_TCB_FIELD(tcb, SND_UNA_RAW)) {
if (GET_TCB_FIELD(tcb, T_RXTSHIFT) != 0)
sample |= TS_RTO;
if (GET_TCB_FIELD(tcb, T_DUPACKS) != 0)
sample |= TS_DUPACKS;
if (GET_TCB_FIELD(tcb, T_DUPACKS) >= td->dupack_threshold)
sample |= TS_FASTREXMT;
}
if (GET_TCB_FIELD(tcb, SND_MAX_RAW) != 0) {
uint32_t snd_wnd;
sample |= TS_SND_BACKLOGGED; /* for whatever reason. */
snd_wnd = GET_TCB_FIELD(tcb, RCV_ADV);
if (tflags & V_TF_RECV_SCALE(1))
snd_wnd <<= GET_TCB_FIELD(tcb, RCV_SCALE);
if (GET_TCB_FIELD(tcb, SND_CWND) < snd_wnd)
sample |= TS_CWND_LIMITED; /* maybe due to CWND */
}
if (tflags & V_TF_CCTRL_ECN(1)) {
/*
* CE marker on incoming IP hdr, echoing ECE back in the TCP
* hdr. Indicates congestion somewhere on the way from the peer
* to this node.
*/
if (tflags & V_TF_CCTRL_ECE(1))
sample |= TS_ECN_ECE;
/*
* ECE seen and CWR sent (or about to be sent). Might indicate
* congestion on the way to the peer. This node is reducing its
* congestion window in response.
*/
if (tflags & (V_TF_CCTRL_CWR(1) | V_TF_CCTRL_RFR(1)))
sample |= TS_ECN_CWR;
}
te->te_sample[te->te_pidx] = sample;
if (++te->te_pidx == nitems(te->te_sample))
te->te_pidx = 0;
memcpy(te->te_tcb, tcb, TCB_SIZE);
te->te_flags |= TE_ACTIVE;
}
static int
do_get_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
{
struct adapter *sc = iq->adapter;
const struct cpl_get_tcb_rpl *cpl = mtod(m, const void *);
const uint64_t *tcb = (const uint64_t *)(const void *)(cpl + 1);
struct tcb_histent *te;
const u_int tid = GET_TID(cpl);
bool remove;
remove = GET_TCB_FIELD(tcb, T_STATE) == TCPS_CLOSED;
te = lookup_tcb_histent(sc, tid, remove);
if (te == NULL) {
/* Not in the history. Who issued the GET_TCB for this? */
device_printf(sc->dev, "tcb %u: flags 0x%016jx, state %u, "
"srtt %u, sscale %u, rscale %u, cookie 0x%x\n", tid,
(uintmax_t)get_tcb_tflags(tcb), GET_TCB_FIELD(tcb, T_STATE),
GET_TCB_FIELD(tcb, T_SRTT), GET_TCB_FIELD(tcb, SND_SCALE),
GET_TCB_FIELD(tcb, RCV_SCALE), cpl->cookie);
goto done;
}
MPASS(te->te_flags & TE_RPL_PENDING);
te->te_flags &= ~TE_RPL_PENDING;
if (remove) {
remove_tcb_histent(te);
} else {
update_tcb_histent(te, tcb);
callout_reset(&te->te_callout, hz / 10, request_tcb, te);
release_tcb_histent(te);
}
done:
m_freem(m);
return (0);
}
static void
fill_tcp_info_from_tcb(struct adapter *sc, uint64_t *tcb, struct tcp_info *ti)
{
uint32_t v;
ti->tcpi_state = GET_TCB_FIELD(tcb, T_STATE);
v = GET_TCB_FIELD(tcb, T_SRTT);
ti->tcpi_rtt = tcp_ticks_to_us(sc, v);
v = GET_TCB_FIELD(tcb, T_RTTVAR);
ti->tcpi_rttvar = tcp_ticks_to_us(sc, v);
ti->tcpi_snd_ssthresh = GET_TCB_FIELD(tcb, SND_SSTHRESH);
ti->tcpi_snd_cwnd = GET_TCB_FIELD(tcb, SND_CWND);
ti->tcpi_rcv_nxt = GET_TCB_FIELD(tcb, RCV_NXT);
v = GET_TCB_FIELD(tcb, TX_MAX);
ti->tcpi_snd_nxt = v - GET_TCB_FIELD(tcb, SND_NXT_RAW);
/* Receive window being advertised by us. */
ti->tcpi_rcv_wscale = GET_TCB_FIELD(tcb, SND_SCALE); /* Yes, SND. */
ti->tcpi_rcv_space = GET_TCB_FIELD(tcb, RCV_WND);
/* Send window */
ti->tcpi_snd_wscale = GET_TCB_FIELD(tcb, RCV_SCALE); /* Yes, RCV. */
ti->tcpi_snd_wnd = GET_TCB_FIELD(tcb, RCV_ADV);
if (get_tcb_tflags(tcb) & V_TF_RECV_SCALE(1))
ti->tcpi_snd_wnd <<= ti->tcpi_snd_wscale;
else
ti->tcpi_snd_wscale = 0;
}
static void
fill_tcp_info_from_history(struct adapter *sc, struct tcb_histent *te,
struct tcp_info *ti)
{
fill_tcp_info_from_tcb(sc, te->te_tcb, ti);
}
/*
* Reads the TCB for the given tid using a memory window and copies it to 'buf'
* in the same format as CPL_GET_TCB_RPL.
*/
static void
read_tcb_using_memwin(struct adapter *sc, u_int tid, uint64_t *buf)
{
int i, j, k, rc;
uint32_t addr;
u_char *tcb, tmp;
MPASS(tid >= sc->tids.tid_base);
MPASS(tid - sc->tids.tid_base < sc->tids.ntids);
addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) + tid * TCB_SIZE;
rc = read_via_memwin(sc, 2, addr, (uint32_t *)buf, TCB_SIZE);
if (rc != 0)
return;
tcb = (u_char *)buf;
for (i = 0, j = TCB_SIZE - 16; i < j; i += 16, j -= 16) {
for (k = 0; k < 16; k++) {
tmp = tcb[i + k];
tcb[i + k] = tcb[j + k];
tcb[j + k] = tmp;
}
}
}
static void
fill_tcp_info(struct adapter *sc, u_int tid, struct tcp_info *ti)
{
uint64_t tcb[TCB_SIZE / sizeof(uint64_t)];
struct tcb_histent *te;
ti->tcpi_toe_tid = tid;
te = lookup_tcb_histent(sc, tid, false);
if (te != NULL) {
fill_tcp_info_from_history(sc, te, ti);
release_tcb_histent(te);
} else {
if (!(sc->debug_flags & DF_DISABLE_TCB_CACHE)) {
/* XXX: tell firmware to flush TCB cache. */
}
read_tcb_using_memwin(sc, tid, tcb);
fill_tcp_info_from_tcb(sc, tcb, ti);
}
}
/*
* Called by the kernel to allow the TOE driver to "refine" values filled up in
* the tcp_info for an offloaded connection.
*/
static void
-t4_tcp_info(struct toedev *tod, struct tcpcb *tp, struct tcp_info *ti)
+t4_tcp_info(struct toedev *tod, const struct tcpcb *tp, struct tcp_info *ti)
{
struct adapter *sc = tod->tod_softc;
struct toepcb *toep = tp->t_toe;
- INP_WLOCK_ASSERT(tptoinpcb(tp));
+ INP_LOCK_ASSERT(tptoinpcb(tp));
MPASS(ti != NULL);
fill_tcp_info(sc, toep->tid, ti);
}
#ifdef KERN_TLS
static int
t4_alloc_tls_session(struct toedev *tod, struct tcpcb *tp,
struct ktls_session *tls, int direction)
{
struct toepcb *toep = tp->t_toe;
INP_WLOCK_ASSERT(tptoinpcb(tp));
MPASS(tls != NULL);
return (tls_alloc_ktls(toep, tls, direction));
}
#endif
/* SET_TCB_FIELD sent as a ULP command looks like this */
#define LEN__SET_TCB_FIELD_ULP (sizeof(struct ulp_txpkt) + \
sizeof(struct ulptx_idata) + sizeof(struct cpl_set_tcb_field_core))
static void *
mk_set_tcb_field_ulp(struct ulp_txpkt *ulpmc, uint64_t word, uint64_t mask,
uint64_t val, uint32_t tid)
{
struct ulptx_idata *ulpsc;
struct cpl_set_tcb_field_core *req;
ulpmc->cmd_dest = htonl(V_ULPTX_CMD(ULP_TX_PKT) | V_ULP_TXPKT_DEST(0));
ulpmc->len = htobe32(howmany(LEN__SET_TCB_FIELD_ULP, 16));
ulpsc = (struct ulptx_idata *)(ulpmc + 1);
ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_IMM));
ulpsc->len = htobe32(sizeof(*req));
req = (struct cpl_set_tcb_field_core *)(ulpsc + 1);
OPCODE_TID(req) = htobe32(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
req->reply_ctrl = htobe16(V_NO_REPLY(1));
req->word_cookie = htobe16(V_WORD(word) | V_COOKIE(0));
req->mask = htobe64(mask);
req->val = htobe64(val);
ulpsc = (struct ulptx_idata *)(req + 1);
if (LEN__SET_TCB_FIELD_ULP % 16) {
ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
ulpsc->len = htobe32(0);
return (ulpsc + 1);
}
return (ulpsc);
}
static void
send_mss_flowc_wr(struct adapter *sc, struct toepcb *toep)
{
struct wrq_cookie cookie;
struct fw_flowc_wr *flowc;
struct ofld_tx_sdesc *txsd;
const int flowclen = sizeof(*flowc) + sizeof(struct fw_flowc_mnemval);
const int flowclen16 = howmany(flowclen, 16);
if (toep->tx_credits < flowclen16 || toep->txsd_avail == 0) {
CH_ERR(sc, "%s: tid %u out of tx credits (%d, %d).\n", __func__,
toep->tid, toep->tx_credits, toep->txsd_avail);
return;
}
flowc = start_wrq_wr(&toep->ofld_txq->wrq, flowclen16, &cookie);
if (__predict_false(flowc == NULL)) {
CH_ERR(sc, "ENOMEM in %s for tid %u.\n", __func__, toep->tid);
return;
}
flowc->op_to_nparams = htobe32(V_FW_WR_OP(FW_FLOWC_WR) |
V_FW_FLOWC_WR_NPARAMS(1));
flowc->flowid_len16 = htonl(V_FW_WR_LEN16(flowclen16) |
V_FW_WR_FLOWID(toep->tid));
flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_MSS;
flowc->mnemval[0].val = htobe32(toep->params.emss);
txsd = &toep->txsd[toep->txsd_pidx];
txsd->tx_credits = flowclen16;
txsd->plen = 0;
toep->tx_credits -= txsd->tx_credits;
if (__predict_false(++toep->txsd_pidx == toep->txsd_total))
toep->txsd_pidx = 0;
toep->txsd_avail--;
commit_wrq_wr(&toep->ofld_txq->wrq, flowc, &cookie);
}
static void
t4_pmtu_update(struct toedev *tod, struct tcpcb *tp, tcp_seq seq, int mtu)
{
struct work_request_hdr *wrh;
struct ulp_txpkt *ulpmc;
int idx, len;
struct wrq_cookie cookie;
struct inpcb *inp = tptoinpcb(tp);
struct toepcb *toep = tp->t_toe;
struct adapter *sc = td_adapter(toep->td);
unsigned short *mtus = &sc->params.mtus[0];
INP_WLOCK_ASSERT(inp);
MPASS(mtu > 0); /* kernel is supposed to provide something usable. */
/* tp->snd_una and snd_max are in host byte order too. */
seq = be32toh(seq);
CTR6(KTR_CXGBE, "%s: tid %d, seq 0x%08x, mtu %u, mtu_idx %u (%d)",
__func__, toep->tid, seq, mtu, toep->params.mtu_idx,
mtus[toep->params.mtu_idx]);
if (ulp_mode(toep) == ULP_MODE_NONE && /* XXX: Read TCB otherwise? */
(SEQ_LT(seq, tp->snd_una) || SEQ_GEQ(seq, tp->snd_max))) {
CTR5(KTR_CXGBE,
"%s: tid %d, seq 0x%08x not in range [0x%08x, 0x%08x).",
__func__, toep->tid, seq, tp->snd_una, tp->snd_max);
return;
}
/* Find the best mtu_idx for the suggested MTU. */
for (idx = 0; idx < NMTUS - 1 && mtus[idx + 1] <= mtu; idx++)
continue;
if (idx >= toep->params.mtu_idx)
return; /* Never increase the PMTU (just like the kernel). */
/*
* We'll send a compound work request with 2 SET_TCB_FIELDs -- the first
* one updates the mtu_idx and the second one triggers a retransmit.
*/
len = sizeof(*wrh) + 2 * roundup2(LEN__SET_TCB_FIELD_ULP, 16);
wrh = start_wrq_wr(toep->ctrlq, howmany(len, 16), &cookie);
if (wrh == NULL) {
CH_ERR(sc, "failed to change mtu_idx of tid %d (%u -> %u).\n",
toep->tid, toep->params.mtu_idx, idx);
return;
}
INIT_ULPTX_WRH(wrh, len, 1, 0); /* atomic */
ulpmc = (struct ulp_txpkt *)(wrh + 1);
ulpmc = mk_set_tcb_field_ulp(ulpmc, W_TCB_T_MAXSEG,
V_TCB_T_MAXSEG(M_TCB_T_MAXSEG), V_TCB_T_MAXSEG(idx), toep->tid);
ulpmc = mk_set_tcb_field_ulp(ulpmc, W_TCB_TIMESTAMP,
V_TCB_TIMESTAMP(0x7FFFFULL << 11), 0, toep->tid);
commit_wrq_wr(toep->ctrlq, wrh, &cookie);
/* Update the software toepcb and tcpcb. */
toep->params.mtu_idx = idx;
tp->t_maxseg = mtus[toep->params.mtu_idx];
if (inp->inp_inc.inc_flags & INC_ISIPV6)
tp->t_maxseg -= sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
else
tp->t_maxseg -= sizeof(struct ip) + sizeof(struct tcphdr);
toep->params.emss = tp->t_maxseg;
if (tp->t_flags & TF_RCVD_TSTMP)
toep->params.emss -= TCPOLEN_TSTAMP_APPA;
/* Update the firmware flowc. */
send_mss_flowc_wr(sc, toep);
/* Update the MTU in the kernel's hostcache. */
if (sc->tt.update_hc_on_pmtu_change != 0) {
struct in_conninfo inc = {0};
inc.inc_fibnum = inp->inp_inc.inc_fibnum;
if (inp->inp_inc.inc_flags & INC_ISIPV6) {
inc.inc_flags |= INC_ISIPV6;
inc.inc6_faddr = inp->inp_inc.inc6_faddr;
} else {
inc.inc_faddr = inp->inp_inc.inc_faddr;
}
tcp_hc_updatemtu(&inc, mtu);
}
CTR6(KTR_CXGBE, "%s: tid %d, mtu_idx %u (%u), t_maxseg %u, emss %u",
__func__, toep->tid, toep->params.mtu_idx,
mtus[toep->params.mtu_idx], tp->t_maxseg, toep->params.emss);
}
/*
* The TOE driver will not receive any more CPLs for the tid associated with the
* toepcb; release the hold on the inpcb.
*/
void
final_cpl_received(struct toepcb *toep)
{
struct inpcb *inp = toep->inp;
bool need_wakeup;
KASSERT(inp != NULL, ("%s: inp is NULL", __func__));
INP_WLOCK_ASSERT(inp);
KASSERT(toep->flags & TPF_CPL_PENDING,
("%s: CPL not pending already?", __func__));
CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)",
__func__, toep->tid, toep, toep->flags, inp, inp->inp_flags);
if (ulp_mode(toep) == ULP_MODE_TCPDDP)
release_ddp_resources(toep);
toep->inp = NULL;
need_wakeup = (toep->flags & TPF_WAITING_FOR_FINAL) != 0;
toep->flags &= ~(TPF_CPL_PENDING | TPF_WAITING_FOR_FINAL);
mbufq_drain(&toep->ulp_pduq);
mbufq_drain(&toep->ulp_pdu_reclaimq);
if (!(toep->flags & TPF_ATTACHED))
release_offload_resources(toep);
if (!in_pcbrele_wlocked(inp))
INP_WUNLOCK(inp);
if (need_wakeup) {
struct mtx *lock = mtx_pool_find(mtxpool_sleep, toep);
mtx_lock(lock);
wakeup(toep);
mtx_unlock(lock);
}
}
void
insert_tid(struct adapter *sc, int tid, void *ctx, int ntids)
{
struct tid_info *t = &sc->tids;
MPASS(tid >= t->tid_base);
MPASS(tid - t->tid_base < t->ntids);
t->tid_tab[tid - t->tid_base] = ctx;
atomic_add_int(&t->tids_in_use, ntids);
}
void *
lookup_tid(struct adapter *sc, int tid)
{
struct tid_info *t = &sc->tids;
return (t->tid_tab[tid - t->tid_base]);
}
void
update_tid(struct adapter *sc, int tid, void *ctx)
{
struct tid_info *t = &sc->tids;
t->tid_tab[tid - t->tid_base] = ctx;
}
void
remove_tid(struct adapter *sc, int tid, int ntids)
{
struct tid_info *t = &sc->tids;
t->tid_tab[tid - t->tid_base] = NULL;
atomic_subtract_int(&t->tids_in_use, ntids);
}
/*
* What mtu_idx to use, given a 4-tuple. Note that both s->mss and tcp_mssopt
* have the MSS that we should advertise in our SYN. Advertised MSS doesn't
* account for any TCP options so the effective MSS (only payload, no headers or
* options) could be different.
*/
static int
find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc,
struct offload_settings *s)
{
unsigned short *mtus = &sc->params.mtus[0];
int i, mss, mtu;
MPASS(inc != NULL);
mss = s->mss > 0 ? s->mss : tcp_mssopt(inc);
if (inc->inc_flags & INC_ISIPV6)
mtu = mss + sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
else
mtu = mss + sizeof(struct ip) + sizeof(struct tcphdr);
for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mtu; i++)
continue;
return (i);
}
/*
* Determine the receive window size for a socket.
*/
u_long
select_rcv_wnd(struct socket *so)
{
unsigned long wnd;
SOCKBUF_LOCK_ASSERT(&so->so_rcv);
wnd = sbspace(&so->so_rcv);
if (wnd < MIN_RCV_WND)
wnd = MIN_RCV_WND;
return min(wnd, MAX_RCV_WND);
}
int
select_rcv_wscale(void)
{
int wscale = 0;
unsigned long space = sb_max;
if (space > MAX_RCV_WND)
space = MAX_RCV_WND;
while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space)
wscale++;
return (wscale);
}
__be64
calc_options0(struct vi_info *vi, struct conn_params *cp)
{
uint64_t opt0 = 0;
opt0 |= F_TCAM_BYPASS;
MPASS(cp->wscale >= 0 && cp->wscale <= M_WND_SCALE);
opt0 |= V_WND_SCALE(cp->wscale);
MPASS(cp->mtu_idx >= 0 && cp->mtu_idx < NMTUS);
opt0 |= V_MSS_IDX(cp->mtu_idx);
MPASS(cp->ulp_mode >= 0 && cp->ulp_mode <= M_ULP_MODE);
opt0 |= V_ULP_MODE(cp->ulp_mode);
MPASS(cp->opt0_bufsize >= 0 && cp->opt0_bufsize <= M_RCV_BUFSIZ);
opt0 |= V_RCV_BUFSIZ(cp->opt0_bufsize);
MPASS(cp->l2t_idx >= 0 && cp->l2t_idx < vi->adapter->vres.l2t.size);
opt0 |= V_L2T_IDX(cp->l2t_idx);
opt0 |= V_SMAC_SEL(vi->smt_idx);
opt0 |= V_TX_CHAN(vi->pi->tx_chan);
MPASS(cp->keepalive == 0 || cp->keepalive == 1);
opt0 |= V_KEEP_ALIVE(cp->keepalive);
MPASS(cp->nagle == 0 || cp->nagle == 1);
opt0 |= V_NAGLE(cp->nagle);
return (htobe64(opt0));
}
__be32
calc_options2(struct vi_info *vi, struct conn_params *cp)
{
uint32_t opt2 = 0;
struct port_info *pi = vi->pi;
struct adapter *sc = pi->adapter;
/*
* rx flow control, rx coalesce, congestion control, and tx pace are all
* explicitly set by the driver. On T5+ the ISS is also set by the
* driver to the value picked by the kernel.
*/
if (is_t4(sc)) {
opt2 |= F_RX_FC_VALID | F_RX_COALESCE_VALID;
opt2 |= F_CONG_CNTRL_VALID | F_PACE_VALID;
} else {
opt2 |= F_T5_OPT_2_VALID; /* all 4 valid */
opt2 |= F_T5_ISS; /* ISS provided in CPL */
}
MPASS(cp->sack == 0 || cp->sack == 1);
opt2 |= V_SACK_EN(cp->sack);
MPASS(cp->tstamp == 0 || cp->tstamp == 1);
opt2 |= V_TSTAMPS_EN(cp->tstamp);
if (cp->wscale > 0)
opt2 |= F_WND_SCALE_EN;
MPASS(cp->ecn == 0 || cp->ecn == 1);
opt2 |= V_CCTRL_ECN(cp->ecn);
/* XXX: F_RX_CHANNEL for multiple rx c-chan support goes here. */
opt2 |= V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]);
opt2 |= V_PACE(0);
opt2 |= F_RSS_QUEUE_VALID;
opt2 |= V_RSS_QUEUE(sc->sge.ofld_rxq[cp->rxq_idx].iq.abs_id);
MPASS(cp->cong_algo >= 0 && cp->cong_algo <= M_CONG_CNTRL);
opt2 |= V_CONG_CNTRL(cp->cong_algo);
MPASS(cp->rx_coalesce == 0 || cp->rx_coalesce == 1);
if (cp->rx_coalesce == 1)
opt2 |= V_RX_COALESCE(M_RX_COALESCE);
opt2 |= V_RX_FC_DDP(0) | V_RX_FC_DISABLE(0);
#ifdef USE_DDP_RX_FLOW_CONTROL
if (cp->ulp_mode == ULP_MODE_TCPDDP)
opt2 |= F_RX_FC_DDP;
#endif
return (htobe32(opt2));
}
uint64_t
select_ntuple(struct vi_info *vi, struct l2t_entry *e)
{
struct adapter *sc = vi->adapter;
struct tp_params *tp = &sc->params.tp;
uint64_t ntuple = 0;
/*
* Initialize each of the fields which we care about which are present
* in the Compressed Filter Tuple.
*/
if (tp->vlan_shift >= 0 && EVL_VLANOFTAG(e->vlan) != CPL_L2T_VLAN_NONE)
ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift;
if (tp->port_shift >= 0)
ntuple |= (uint64_t)e->lport << tp->port_shift;
if (tp->protocol_shift >= 0)
ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift;
if (tp->vnic_shift >= 0 && tp->vnic_mode == FW_VNIC_MODE_PF_VF) {
ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vi->vin) |
V_FT_VNID_ID_PF(sc->pf) | V_FT_VNID_ID_VLD(vi->vfvld)) <<
tp->vnic_shift;
}
if (is_t4(sc))
return (htobe32((uint32_t)ntuple));
else
return (htobe64(V_FILTER_TUPLE(ntuple)));
}
/*
* Initialize various connection parameters.
*/
void
init_conn_params(struct vi_info *vi , struct offload_settings *s,
struct in_conninfo *inc, struct socket *so,
const struct tcp_options *tcpopt, int16_t l2t_idx, struct conn_params *cp)
{
struct port_info *pi = vi->pi;
struct adapter *sc = pi->adapter;
struct tom_tunables *tt = &sc->tt;
struct inpcb *inp = sotoinpcb(so);
struct tcpcb *tp = intotcpcb(inp);
u_long wnd;
u_int q_idx;
MPASS(s->offload != 0);
/* Congestion control algorithm */
if (s->cong_algo >= 0)
cp->cong_algo = s->cong_algo & M_CONG_CNTRL;
else if (sc->tt.cong_algorithm >= 0)
cp->cong_algo = tt->cong_algorithm & M_CONG_CNTRL;
else {
struct cc_algo *cc = CC_ALGO(tp);
if (strcasecmp(cc->name, "reno") == 0)
cp->cong_algo = CONG_ALG_RENO;
else if (strcasecmp(cc->name, "tahoe") == 0)
cp->cong_algo = CONG_ALG_TAHOE;
if (strcasecmp(cc->name, "newreno") == 0)
cp->cong_algo = CONG_ALG_NEWRENO;
if (strcasecmp(cc->name, "highspeed") == 0)
cp->cong_algo = CONG_ALG_HIGHSPEED;
else {
/*
* Use newreno in case the algorithm selected by the
* host stack is not supported by the hardware.
*/
cp->cong_algo = CONG_ALG_NEWRENO;
}
}
/* Tx traffic scheduling class. */
if (s->sched_class >= 0 && s->sched_class < sc->params.nsched_cls)
cp->tc_idx = s->sched_class;
else
cp->tc_idx = -1;
/* Nagle's algorithm. */
if (s->nagle >= 0)
cp->nagle = s->nagle > 0 ? 1 : 0;
else
cp->nagle = tp->t_flags & TF_NODELAY ? 0 : 1;
/* TCP Keepalive. */
if (V_tcp_always_keepalive || so_options_get(so) & SO_KEEPALIVE)
cp->keepalive = 1;
else
cp->keepalive = 0;
/* Optimization that's specific to T5 @ 40G. */
if (tt->tx_align >= 0)
cp->tx_align = tt->tx_align > 0 ? 1 : 0;
else if (chip_id(sc) == CHELSIO_T5 &&
(port_top_speed(pi) > 10 || sc->params.nports > 2))
cp->tx_align = 1;
else
cp->tx_align = 0;
/* ULP mode. */
if (s->ddp > 0 ||
(s->ddp < 0 && sc->tt.ddp && (so_options_get(so) & SO_NO_DDP) == 0))
cp->ulp_mode = ULP_MODE_TCPDDP;
else
cp->ulp_mode = ULP_MODE_NONE;
/* Rx coalescing. */
if (s->rx_coalesce >= 0)
cp->rx_coalesce = s->rx_coalesce > 0 ? 1 : 0;
else if (tt->rx_coalesce >= 0)
cp->rx_coalesce = tt->rx_coalesce > 0 ? 1 : 0;
else
cp->rx_coalesce = 1; /* default */
/*
* Index in the PMTU table. This controls the MSS that we announce in
* our SYN initially, but after ESTABLISHED it controls the MSS that we
* use to send data.
*/
cp->mtu_idx = find_best_mtu_idx(sc, inc, s);
/* Tx queue for this connection. */
if (s->txq == QUEUE_RANDOM)
q_idx = arc4random();
else if (s->txq == QUEUE_ROUNDROBIN)
q_idx = atomic_fetchadd_int(&vi->txq_rr, 1);
else
q_idx = s->txq;
cp->txq_idx = vi->first_ofld_txq + q_idx % vi->nofldtxq;
/* Rx queue for this connection. */
if (s->rxq == QUEUE_RANDOM)
q_idx = arc4random();
else if (s->rxq == QUEUE_ROUNDROBIN)
q_idx = atomic_fetchadd_int(&vi->rxq_rr, 1);
else
q_idx = s->rxq;
cp->rxq_idx = vi->first_ofld_rxq + q_idx % vi->nofldrxq;
if (SOLISTENING(so)) {
/* Passive open */
MPASS(tcpopt != NULL);
/* TCP timestamp option */
if (tcpopt->tstamp &&
(s->tstamp > 0 || (s->tstamp < 0 && V_tcp_do_rfc1323)))
cp->tstamp = 1;
else
cp->tstamp = 0;
/* SACK */
if (tcpopt->sack &&
(s->sack > 0 || (s->sack < 0 && V_tcp_do_sack)))
cp->sack = 1;
else
cp->sack = 0;
/* Receive window scaling. */
if (tcpopt->wsf > 0 && tcpopt->wsf < 15 && V_tcp_do_rfc1323)
cp->wscale = select_rcv_wscale();
else
cp->wscale = 0;
/* ECN */
if (tcpopt->ecn && /* XXX: review. */
(s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn)))
cp->ecn = 1;
else
cp->ecn = 0;
wnd = max(so->sol_sbrcv_hiwat, MIN_RCV_WND);
cp->opt0_bufsize = min(wnd >> 10, M_RCV_BUFSIZ);
if (tt->sndbuf > 0)
cp->sndbuf = tt->sndbuf;
else if (so->sol_sbsnd_flags & SB_AUTOSIZE &&
V_tcp_do_autosndbuf)
cp->sndbuf = 256 * 1024;
else
cp->sndbuf = so->sol_sbsnd_hiwat;
} else {
/* Active open */
/* TCP timestamp option */
if (s->tstamp > 0 ||
(s->tstamp < 0 && (tp->t_flags & TF_REQ_TSTMP)))
cp->tstamp = 1;
else
cp->tstamp = 0;
/* SACK */
if (s->sack > 0 ||
(s->sack < 0 && (tp->t_flags & TF_SACK_PERMIT)))
cp->sack = 1;
else
cp->sack = 0;
/* Receive window scaling */
if (tp->t_flags & TF_REQ_SCALE)
cp->wscale = select_rcv_wscale();
else
cp->wscale = 0;
/* ECN */
if (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn == 1))
cp->ecn = 1;
else
cp->ecn = 0;
SOCKBUF_LOCK(&so->so_rcv);
wnd = max(select_rcv_wnd(so), MIN_RCV_WND);
SOCKBUF_UNLOCK(&so->so_rcv);
cp->opt0_bufsize = min(wnd >> 10, M_RCV_BUFSIZ);
if (tt->sndbuf > 0)
cp->sndbuf = tt->sndbuf;
else {
SOCKBUF_LOCK(&so->so_snd);
if (so->so_snd.sb_flags & SB_AUTOSIZE &&
V_tcp_do_autosndbuf)
cp->sndbuf = 256 * 1024;
else
cp->sndbuf = so->so_snd.sb_hiwat;
SOCKBUF_UNLOCK(&so->so_snd);
}
}
cp->l2t_idx = l2t_idx;
/* This will be initialized on ESTABLISHED. */
cp->emss = 0;
}
int
negative_advice(int status)
{
return (status == CPL_ERR_RTX_NEG_ADVICE ||
status == CPL_ERR_PERSIST_NEG_ADVICE ||
status == CPL_ERR_KEEPALV_NEG_ADVICE);
}
static int
alloc_tid_tab(struct tid_info *t, int flags)
{
MPASS(t->ntids > 0);
MPASS(t->tid_tab == NULL);
t->tid_tab = malloc(t->ntids * sizeof(*t->tid_tab), M_CXGBE,
M_ZERO | flags);
if (t->tid_tab == NULL)
return (ENOMEM);
atomic_store_rel_int(&t->tids_in_use, 0);
return (0);
}
static void
free_tid_tab(struct tid_info *t)
{
KASSERT(t->tids_in_use == 0,
("%s: %d tids still in use.", __func__, t->tids_in_use));
free(t->tid_tab, M_CXGBE);
t->tid_tab = NULL;
}
static int
alloc_stid_tab(struct tid_info *t, int flags)
{
MPASS(t->nstids > 0);
MPASS(t->stid_tab == NULL);
t->stid_tab = malloc(t->nstids * sizeof(*t->stid_tab), M_CXGBE,
M_ZERO | flags);
if (t->stid_tab == NULL)
return (ENOMEM);
mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF);
t->stids_in_use = 0;
TAILQ_INIT(&t->stids);
t->nstids_free_head = t->nstids;
return (0);
}
static void
free_stid_tab(struct tid_info *t)
{
KASSERT(t->stids_in_use == 0,
("%s: %d tids still in use.", __func__, t->stids_in_use));
if (mtx_initialized(&t->stid_lock))
mtx_destroy(&t->stid_lock);
free(t->stid_tab, M_CXGBE);
t->stid_tab = NULL;
}
static void
free_tid_tabs(struct tid_info *t)
{
free_tid_tab(t);
free_stid_tab(t);
}
static int
alloc_tid_tabs(struct tid_info *t)
{
int rc;
rc = alloc_tid_tab(t, M_NOWAIT);
if (rc != 0)
goto failed;
rc = alloc_stid_tab(t, M_NOWAIT);
if (rc != 0)
goto failed;
return (0);
failed:
free_tid_tabs(t);
return (rc);
}
static inline void
alloc_tcb_history(struct adapter *sc, struct tom_data *td)
{
if (sc->tids.ntids == 0 || sc->tids.ntids > 1024)
return;
rw_init(&td->tcb_history_lock, "TCB history");
td->tcb_history = malloc(sc->tids.ntids * sizeof(*td->tcb_history),
M_CXGBE, M_ZERO | M_NOWAIT);
td->dupack_threshold = G_DUPACKTHRESH(t4_read_reg(sc, A_TP_PARA_REG0));
}
static inline void
free_tcb_history(struct adapter *sc, struct tom_data *td)
{
#ifdef INVARIANTS
int i;
if (td->tcb_history != NULL) {
for (i = 0; i < sc->tids.ntids; i++) {
MPASS(td->tcb_history[i] == NULL);
}
}
#endif
free(td->tcb_history, M_CXGBE);
if (rw_initialized(&td->tcb_history_lock))
rw_destroy(&td->tcb_history_lock);
}
static void
free_tom_data(struct adapter *sc, struct tom_data *td)
{
ASSERT_SYNCHRONIZED_OP(sc);
KASSERT(TAILQ_EMPTY(&td->toep_list),
("%s: TOE PCB list is not empty.", __func__));
KASSERT(td->lctx_count == 0,
("%s: lctx hash table is not empty.", __func__));
t4_free_ppod_region(&td->pr);
if (td->listen_mask != 0)
hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask);
if (mtx_initialized(&td->unsent_wr_lock))
mtx_destroy(&td->unsent_wr_lock);
if (mtx_initialized(&td->lctx_hash_lock))
mtx_destroy(&td->lctx_hash_lock);
if (mtx_initialized(&td->toep_list_lock))
mtx_destroy(&td->toep_list_lock);
free_tcb_history(sc, td);
free_tid_tabs(&sc->tids);
free(td, M_CXGBE);
}
static char *
prepare_pkt(int open_type, uint16_t vtag, struct inpcb *inp, int *pktlen,
int *buflen)
{
char *pkt;
struct tcphdr *th;
int ipv6, len;
const int maxlen =
max(sizeof(struct ether_header), sizeof(struct ether_vlan_header)) +
max(sizeof(struct ip), sizeof(struct ip6_hdr)) +
sizeof(struct tcphdr);
MPASS(open_type == OPEN_TYPE_ACTIVE || open_type == OPEN_TYPE_LISTEN);
pkt = malloc(maxlen, M_CXGBE, M_ZERO | M_NOWAIT);
if (pkt == NULL)
return (NULL);
ipv6 = inp->inp_vflag & INP_IPV6;
len = 0;
if (EVL_VLANOFTAG(vtag) == 0xfff) {
struct ether_header *eh = (void *)pkt;
if (ipv6)
eh->ether_type = htons(ETHERTYPE_IPV6);
else
eh->ether_type = htons(ETHERTYPE_IP);
len += sizeof(*eh);
} else {
struct ether_vlan_header *evh = (void *)pkt;
evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
evh->evl_tag = htons(vtag);
if (ipv6)
evh->evl_proto = htons(ETHERTYPE_IPV6);
else
evh->evl_proto = htons(ETHERTYPE_IP);
len += sizeof(*evh);
}
if (ipv6) {
struct ip6_hdr *ip6 = (void *)&pkt[len];
ip6->ip6_vfc = IPV6_VERSION;
ip6->ip6_plen = htons(sizeof(struct tcphdr));
ip6->ip6_nxt = IPPROTO_TCP;
if (open_type == OPEN_TYPE_ACTIVE) {
ip6->ip6_src = inp->in6p_laddr;
ip6->ip6_dst = inp->in6p_faddr;
} else if (open_type == OPEN_TYPE_LISTEN) {
ip6->ip6_src = inp->in6p_laddr;
ip6->ip6_dst = ip6->ip6_src;
}
len += sizeof(*ip6);
} else {
struct ip *ip = (void *)&pkt[len];
ip->ip_v = IPVERSION;
ip->ip_hl = sizeof(*ip) >> 2;
ip->ip_tos = inp->inp_ip_tos;
ip->ip_len = htons(sizeof(struct ip) + sizeof(struct tcphdr));
ip->ip_ttl = inp->inp_ip_ttl;
ip->ip_p = IPPROTO_TCP;
if (open_type == OPEN_TYPE_ACTIVE) {
ip->ip_src = inp->inp_laddr;
ip->ip_dst = inp->inp_faddr;
} else if (open_type == OPEN_TYPE_LISTEN) {
ip->ip_src = inp->inp_laddr;
ip->ip_dst = ip->ip_src;
}
len += sizeof(*ip);
}
th = (void *)&pkt[len];
if (open_type == OPEN_TYPE_ACTIVE) {
th->th_sport = inp->inp_lport; /* network byte order already */
th->th_dport = inp->inp_fport; /* ditto */
} else if (open_type == OPEN_TYPE_LISTEN) {
th->th_sport = inp->inp_lport; /* network byte order already */
th->th_dport = th->th_sport;
}
len += sizeof(th);
*pktlen = *buflen = len;
return (pkt);
}
const struct offload_settings *
lookup_offload_policy(struct adapter *sc, int open_type, struct mbuf *m,
uint16_t vtag, struct inpcb *inp)
{
const struct t4_offload_policy *op;
char *pkt;
struct offload_rule *r;
int i, matched, pktlen, buflen;
static const struct offload_settings allow_offloading_settings = {
.offload = 1,
.rx_coalesce = -1,
.cong_algo = -1,
.sched_class = -1,
.tstamp = -1,
.sack = -1,
.nagle = -1,
.ecn = -1,
.ddp = -1,
.tls = -1,
.txq = QUEUE_RANDOM,
.rxq = QUEUE_RANDOM,
.mss = -1,
};
static const struct offload_settings disallow_offloading_settings = {
.offload = 0,
/* rest is irrelevant when offload is off. */
};
rw_assert(&sc->policy_lock, RA_LOCKED);
/*
* If there's no Connection Offloading Policy attached to the device
* then we need to return a default static policy. If
* "cop_managed_offloading" is true, then we need to disallow
* offloading until a COP is attached to the device. Otherwise we
* allow offloading ...
*/
op = sc->policy;
if (op == NULL) {
if (sc->tt.cop_managed_offloading)
return (&disallow_offloading_settings);
else
return (&allow_offloading_settings);
}
switch (open_type) {
case OPEN_TYPE_ACTIVE:
case OPEN_TYPE_LISTEN:
pkt = prepare_pkt(open_type, vtag, inp, &pktlen, &buflen);
break;
case OPEN_TYPE_PASSIVE:
MPASS(m != NULL);
pkt = mtod(m, char *);
MPASS(*pkt == CPL_PASS_ACCEPT_REQ);
pkt += sizeof(struct cpl_pass_accept_req);
pktlen = m->m_pkthdr.len - sizeof(struct cpl_pass_accept_req);
buflen = m->m_len - sizeof(struct cpl_pass_accept_req);
break;
default:
MPASS(0);
return (&disallow_offloading_settings);
}
if (pkt == NULL || pktlen == 0 || buflen == 0)
return (&disallow_offloading_settings);
matched = 0;
r = &op->rule[0];
for (i = 0; i < op->nrules; i++, r++) {
if (r->open_type != open_type &&
r->open_type != OPEN_TYPE_DONTCARE) {
continue;
}
matched = bpf_filter(r->bpf_prog.bf_insns, pkt, pktlen, buflen);
if (matched)
break;
}
if (open_type == OPEN_TYPE_ACTIVE || open_type == OPEN_TYPE_LISTEN)
free(pkt, M_CXGBE);
return (matched ? &r->settings : &disallow_offloading_settings);
}
static void
reclaim_wr_resources(void *arg, int count)
{
struct tom_data *td = arg;
STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list);
struct cpl_act_open_req *cpl;
u_int opcode, atid, tid;
struct wrqe *wr;
struct adapter *sc = td_adapter(td);
mtx_lock(&td->unsent_wr_lock);
STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe);
mtx_unlock(&td->unsent_wr_lock);
while ((wr = STAILQ_FIRST(&twr_list)) != NULL) {
STAILQ_REMOVE_HEAD(&twr_list, link);
cpl = wrtod(wr);
opcode = GET_OPCODE(cpl);
switch (opcode) {
case CPL_ACT_OPEN_REQ:
case CPL_ACT_OPEN_REQ6:
atid = G_TID_TID(be32toh(OPCODE_TID(cpl)));
CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid);
act_open_failure_cleanup(sc, atid, EHOSTUNREACH);
free(wr, M_CXGBE);
break;
case CPL_PASS_ACCEPT_RPL:
tid = GET_TID(cpl);
CTR2(KTR_CXGBE, "%s: tid %u ", __func__, tid);
synack_failure_cleanup(sc, tid);
free(wr, M_CXGBE);
break;
default:
log(LOG_ERR, "%s: leaked work request %p, wr_len %d, "
"opcode %x\n", __func__, wr, wr->wr_len, opcode);
/* WR not freed here; go look at it with a debugger. */
}
}
}
/*
* Ground control to Major TOM
* Commencing countdown, engines on
*/
static int
t4_tom_activate(struct adapter *sc)
{
struct tom_data *td;
struct toedev *tod;
struct vi_info *vi;
int i, rc, v;
ASSERT_SYNCHRONIZED_OP(sc);
/* per-adapter softc for TOM */
td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT);
if (td == NULL)
return (ENOMEM);
/* List of TOE PCBs and associated lock */
mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF);
TAILQ_INIT(&td->toep_list);
/* Listen context */
mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF);
td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE,
&td->listen_mask, HASH_NOWAIT);
/* List of WRs for which L2 resolution failed */
mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF);
STAILQ_INIT(&td->unsent_wr_list);
TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td);
/* TID tables */
rc = alloc_tid_tabs(&sc->tids);
if (rc != 0)
goto done;
rc = t4_init_ppod_region(&td->pr, &sc->vres.ddp,
t4_read_reg(sc, A_ULP_RX_TDDP_PSZ), "TDDP page pods");
if (rc != 0)
goto done;
t4_set_reg_field(sc, A_ULP_RX_TDDP_TAGMASK,
V_TDDPTAGMASK(M_TDDPTAGMASK), td->pr.pr_tag_mask);
alloc_tcb_history(sc, td);
/* toedev ops */
tod = &td->tod;
init_toedev(tod);
tod->tod_softc = sc;
tod->tod_connect = t4_connect;
tod->tod_listen_start = t4_listen_start;
tod->tod_listen_stop = t4_listen_stop;
tod->tod_rcvd = t4_rcvd;
tod->tod_output = t4_tod_output;
tod->tod_send_rst = t4_send_rst;
tod->tod_send_fin = t4_send_fin;
tod->tod_pcb_detach = t4_pcb_detach;
tod->tod_l2_update = t4_l2_update;
tod->tod_syncache_added = t4_syncache_added;
tod->tod_syncache_removed = t4_syncache_removed;
tod->tod_syncache_respond = t4_syncache_respond;
tod->tod_offload_socket = t4_offload_socket;
tod->tod_ctloutput = t4_ctloutput;
tod->tod_tcp_info = t4_tcp_info;
#ifdef KERN_TLS
tod->tod_alloc_tls_session = t4_alloc_tls_session;
#endif
tod->tod_pmtu_update = t4_pmtu_update;
for_each_port(sc, i) {
for_each_vi(sc->port[i], v, vi) {
SETTOEDEV(vi->ifp, &td->tod);
}
}
sc->tom_softc = td;
register_toedev(sc->tom_softc);
done:
if (rc != 0)
free_tom_data(sc, td);
return (rc);
}
static int
t4_tom_deactivate(struct adapter *sc)
{
int rc = 0;
struct tom_data *td = sc->tom_softc;
ASSERT_SYNCHRONIZED_OP(sc);
if (td == NULL)
return (0); /* XXX. KASSERT? */
if (sc->offload_map != 0)
return (EBUSY); /* at least one port has IFCAP_TOE enabled */
if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI))
return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */
mtx_lock(&td->toep_list_lock);
if (!TAILQ_EMPTY(&td->toep_list))
rc = EBUSY;
mtx_unlock(&td->toep_list_lock);
mtx_lock(&td->lctx_hash_lock);
if (td->lctx_count > 0)
rc = EBUSY;
mtx_unlock(&td->lctx_hash_lock);
taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources);
mtx_lock(&td->unsent_wr_lock);
if (!STAILQ_EMPTY(&td->unsent_wr_list))
rc = EBUSY;
mtx_unlock(&td->unsent_wr_lock);
if (rc == 0) {
unregister_toedev(sc->tom_softc);
free_tom_data(sc, td);
sc->tom_softc = NULL;
}
return (rc);
}
static int
t4_aio_queue_tom(struct socket *so, struct kaiocb *job)
{
struct tcpcb *tp = sototcpcb(so);
struct toepcb *toep = tp->t_toe;
int error;
/*
* No lock is needed as TOE sockets never change between
* active and passive.
*/
if (SOLISTENING(so))
return (EINVAL);
if (ulp_mode(toep) == ULP_MODE_TCPDDP) {
error = t4_aio_queue_ddp(so, job);
if (error != EOPNOTSUPP)
return (error);
}
return (t4_aio_queue_aiotx(so, job));
}
static int
t4_tom_mod_load(void)
{
/* CPL handlers */
t4_register_cpl_handler(CPL_GET_TCB_RPL, do_get_tcb_rpl);
t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL, do_l2t_write_rpl2,
CPL_COOKIE_TOM);
t4_init_connect_cpl_handlers();
t4_init_listen_cpl_handlers();
t4_init_cpl_io_handlers();
t4_ddp_mod_load();
t4_tls_mod_load();
bcopy(&tcp_protosw, &toe_protosw, sizeof(toe_protosw));
toe_protosw.pr_aio_queue = t4_aio_queue_tom;
bcopy(&tcp6_protosw, &toe6_protosw, sizeof(toe6_protosw));
toe6_protosw.pr_aio_queue = t4_aio_queue_tom;
return (t4_register_uld(&tom_uld_info));
}
static void
tom_uninit(struct adapter *sc, void *arg __unused)
{
if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun"))
return;
/* Try to free resources (works only if no port has IFCAP_TOE) */
if (uld_active(sc, ULD_TOM))
t4_deactivate_uld(sc, ULD_TOM);
end_synchronized_op(sc, 0);
}
static int
t4_tom_mod_unload(void)
{
t4_iterate(tom_uninit, NULL);
if (t4_unregister_uld(&tom_uld_info) == EBUSY)
return (EBUSY);
t4_tls_mod_unload();
t4_ddp_mod_unload();
t4_uninit_connect_cpl_handlers();
t4_uninit_listen_cpl_handlers();
t4_uninit_cpl_io_handlers();
t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL, NULL, CPL_COOKIE_TOM);
t4_register_cpl_handler(CPL_GET_TCB_RPL, NULL);
return (0);
}
#endif /* TCP_OFFLOAD */
static int
t4_tom_modevent(module_t mod, int cmd, void *arg)
{
int rc = 0;
#ifdef TCP_OFFLOAD
switch (cmd) {
case MOD_LOAD:
rc = t4_tom_mod_load();
break;
case MOD_UNLOAD:
rc = t4_tom_mod_unload();
break;
default:
rc = EINVAL;
}
#else
printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
rc = EOPNOTSUPP;
#endif
return (rc);
}
static moduledata_t t4_tom_moddata= {
"t4_tom",
t4_tom_modevent,
0
};
MODULE_VERSION(t4_tom, 1);
MODULE_DEPEND(t4_tom, toecore, 1, 1, 1);
MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1);
DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY);
diff --git a/sys/netinet/tcp_offload.c b/sys/netinet/tcp_offload.c
index 0a37d0a8ba30..6a362484d46f 100644
--- a/sys/netinet/tcp_offload.c
+++ b/sys/netinet/tcp_offload.c
@@ -1,231 +1,231 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2012 Chelsio Communications, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_inet6.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/eventhandler.h>
#include <sys/mbuf.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/sockopt.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_private.h>
#include <net/route.h>
#include <net/route/nhop.h>
#include <netinet/in.h>
#include <netinet/in_pcb.h>
#include <netinet/in_fib.h>
#include <netinet6/in6_fib.h>
#include <netinet/tcp.h>
#include <netinet/tcp_offload.h>
#define TCPOUTFLAGS
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_var.h>
#include <netinet/toecore.h>
int registered_toedevs;
/*
* Provide an opportunity for a TOE driver to offload.
*/
int
tcp_offload_connect(struct socket *so, struct sockaddr *nam)
{
struct ifnet *ifp;
struct toedev *tod;
struct nhop_object *nh;
struct epoch_tracker et;
int error = EOPNOTSUPP;
INP_WLOCK_ASSERT(sotoinpcb(so));
KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6,
("%s: called with sa_family %d", __func__, nam->sa_family));
if (registered_toedevs == 0)
return (error);
NET_EPOCH_ENTER(et);
nh = NULL;
#ifdef INET
if (nam->sa_family == AF_INET)
nh = fib4_lookup(0, ((struct sockaddr_in *)nam)->sin_addr,
NHR_NONE, 0, 0);
#endif
#if defined(INET) && defined(INET6)
else
#endif
#ifdef INET6
if (nam->sa_family == AF_INET6)
nh = fib6_lookup(0, &((struct sockaddr_in6 *)nam)->sin6_addr,
NHR_NONE, 0, 0);
#endif
if (nh == NULL) {
NET_EPOCH_EXIT(et);
return (EHOSTUNREACH);
}
ifp = nh->nh_ifp;
if (nam->sa_family == AF_INET && !(ifp->if_capenable & IFCAP_TOE4))
goto done;
if (nam->sa_family == AF_INET6 && !(ifp->if_capenable & IFCAP_TOE6))
goto done;
tod = TOEDEV(ifp);
if (tod != NULL)
error = tod->tod_connect(tod, so, nh, nam);
done:
NET_EPOCH_EXIT(et);
return (error);
}
void
tcp_offload_listen_start(struct tcpcb *tp)
{
INP_WLOCK_ASSERT(tptoinpcb(tp));
EVENTHANDLER_INVOKE(tcp_offload_listen_start, tp);
}
void
tcp_offload_listen_stop(struct tcpcb *tp)
{
INP_WLOCK_ASSERT(tptoinpcb(tp));
EVENTHANDLER_INVOKE(tcp_offload_listen_stop, tp);
}
void
tcp_offload_input(struct tcpcb *tp, struct mbuf *m)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
tod->tod_input(tod, tp, m);
}
int
tcp_offload_output(struct tcpcb *tp)
{
struct toedev *tod = tp->tod;
int error, flags;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
flags = tcp_outflags[tp->t_state];
if (flags & TH_RST) {
/* XXX: avoid repeated calls like we do for FIN */
error = tod->tod_send_rst(tod, tp);
} else if ((flags & TH_FIN || tp->t_flags & TF_NEEDFIN) &&
(tp->t_flags & TF_SENTFIN) == 0) {
error = tod->tod_send_fin(tod, tp);
if (error == 0)
tp->t_flags |= TF_SENTFIN;
} else
error = tod->tod_output(tod, tp);
return (error);
}
void
tcp_offload_rcvd(struct tcpcb *tp)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
tod->tod_rcvd(tod, tp);
}
void
tcp_offload_ctloutput(struct tcpcb *tp, int sopt_dir, int sopt_name)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
tod->tod_ctloutput(tod, tp, sopt_dir, sopt_name);
}
void
-tcp_offload_tcp_info(struct tcpcb *tp, struct tcp_info *ti)
+tcp_offload_tcp_info(const struct tcpcb *tp, struct tcp_info *ti)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
- INP_WLOCK_ASSERT(tptoinpcb(tp));
+ INP_LOCK_ASSERT(tptoinpcb(tp));
tod->tod_tcp_info(tod, tp, ti);
}
int
tcp_offload_alloc_tls_session(struct tcpcb *tp, struct ktls_session *tls,
int direction)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
return (tod->tod_alloc_tls_session(tod, tp, tls, direction));
}
void
tcp_offload_detach(struct tcpcb *tp)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
tod->tod_pcb_detach(tod, tp);
}
void
tcp_offload_pmtu_update(struct tcpcb *tp, tcp_seq seq, int mtu)
{
struct toedev *tod = tp->tod;
KASSERT(tod != NULL, ("%s: tp->tod is NULL, tp %p", __func__, tp));
INP_WLOCK_ASSERT(tptoinpcb(tp));
tod->tod_pmtu_update(tod, tp, seq, mtu);
}
diff --git a/sys/netinet/tcp_offload.h b/sys/netinet/tcp_offload.h
index 2524a56d0369..b397336ab487 100644
--- a/sys/netinet/tcp_offload.h
+++ b/sys/netinet/tcp_offload.h
@@ -1,53 +1,53 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2012 Chelsio Communications, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
#ifndef _NETINET_TCP_OFFLOAD_H_
#define _NETINET_TCP_OFFLOAD_H_
#ifndef _KERNEL
#error "no user-serviceable parts inside"
#endif
#include <netinet/tcp.h>
extern int registered_toedevs;
int tcp_offload_connect(struct socket *, struct sockaddr *);
void tcp_offload_listen_start(struct tcpcb *);
void tcp_offload_listen_stop(struct tcpcb *);
void tcp_offload_input(struct tcpcb *, struct mbuf *);
int tcp_offload_output(struct tcpcb *);
void tcp_offload_rcvd(struct tcpcb *);
void tcp_offload_ctloutput(struct tcpcb *, int, int);
-void tcp_offload_tcp_info(struct tcpcb *, struct tcp_info *);
+void tcp_offload_tcp_info(const struct tcpcb *, struct tcp_info *);
int tcp_offload_alloc_tls_session(struct tcpcb *, struct ktls_session *, int);
void tcp_offload_detach(struct tcpcb *);
void tcp_offload_pmtu_update(struct tcpcb *, tcp_seq, int);
#endif
diff --git a/sys/netinet/tcp_usrreq.c b/sys/netinet/tcp_usrreq.c
index 811dfbaa51bd..a6101bc422f7 100644
--- a/sys/netinet/tcp_usrreq.c
+++ b/sys/netinet/tcp_usrreq.c
@@ -1,3158 +1,3158 @@
/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 1982, 1986, 1988, 1993
* The Regents of the University of California.
* Copyright (c) 2006-2007 Robert N. M. Watson
* Copyright (c) 2010-2011 Juniper Networks, Inc.
* All rights reserved.
*
* Portions of this software were developed by Robert N. M. Watson under
* contract to Juniper Networks, Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
*/
#include <sys/cdefs.h>
#include "opt_ddb.h"
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_ipsec.h"
#include "opt_kern_tls.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/arb.h>
#include <sys/limits.h>
#include <sys/malloc.h>
#include <sys/refcount.h>
#include <sys/kernel.h>
#include <sys/ktls.h>
#include <sys/qmath.h>
#include <sys/sysctl.h>
#include <sys/mbuf.h>
#ifdef INET6
#include <sys/domain.h>
#endif /* INET6 */
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/protosw.h>
#include <sys/proc.h>
#include <sys/jail.h>
#include <sys/stats.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
#include <net/if.h>
#include <net/if_var.h>
#include <net/route.h>
#include <net/vnet.h>
#include <netinet/in.h>
#include <netinet/in_kdtrace.h>
#include <netinet/in_pcb.h>
#include <netinet/in_systm.h>
#include <netinet/in_var.h>
#include <netinet/ip.h>
#include <netinet/ip_var.h>
#ifdef INET6
#include <netinet/ip6.h>
#include <netinet6/in6_pcb.h>
#include <netinet6/ip6_var.h>
#include <netinet6/scope6_var.h>
#endif
#include <netinet/tcp.h>
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_seq.h>
#include <netinet/tcp_timer.h>
#include <netinet/tcp_var.h>
#include <netinet/tcp_log_buf.h>
#include <netinet/tcpip.h>
#include <netinet/cc/cc.h>
#include <netinet/tcp_fastopen.h>
#include <netinet/tcp_hpts.h>
#ifdef TCPPCAP
#include <netinet/tcp_pcap.h>
#endif
#ifdef TCP_OFFLOAD
#include <netinet/tcp_offload.h>
#endif
#include <netipsec/ipsec_support.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_extern.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
/*
* TCP protocol interface to socket abstraction.
*/
#ifdef INET
static int tcp_connect(struct tcpcb *, struct sockaddr_in *,
struct thread *td);
#endif /* INET */
#ifdef INET6
static int tcp6_connect(struct tcpcb *, struct sockaddr_in6 *,
struct thread *td);
#endif /* INET6 */
static void tcp_disconnect(struct tcpcb *);
static void tcp_usrclosed(struct tcpcb *);
-static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
+static void tcp_fill_info(const struct tcpcb *, struct tcp_info *);
static int tcp_pru_options_support(struct tcpcb *tp, int flags);
static void
tcp_bblog_pru(struct tcpcb *tp, uint32_t pru, int error)
{
struct tcp_log_buffer *lgb;
KASSERT(tp != NULL, ("tcp_bblog_pru: tp == NULL"));
INP_WLOCK_ASSERT(tptoinpcb(tp));
if (tcp_bblogging_on(tp)) {
lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_PRU, error,
0, NULL, false, NULL, NULL, 0, NULL);
} else {
lgb = NULL;
}
if (lgb != NULL) {
if (error >= 0) {
lgb->tlb_errno = (uint32_t)error;
}
lgb->tlb_flex1 = pru;
}
}
/*
* TCP attaches to socket via pru_attach(), reserving space,
* and an internet control block.
*/
static int
tcp_usr_attach(struct socket *so, int proto, struct thread *td)
{
struct inpcb *inp;
struct tcpcb *tp = NULL;
int error;
inp = sotoinpcb(so);
KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
if (error)
goto out;
so->so_rcv.sb_flags |= SB_AUTOSIZE;
so->so_snd.sb_flags |= SB_AUTOSIZE;
error = in_pcballoc(so, &V_tcbinfo);
if (error)
goto out;
inp = sotoinpcb(so);
tp = tcp_newtcpcb(inp);
if (tp == NULL) {
error = ENOBUFS;
in_pcbdetach(inp);
in_pcbfree(inp);
goto out;
}
tp->t_state = TCPS_CLOSED;
tcp_bblog_pru(tp, PRU_ATTACH, error);
INP_WUNLOCK(inp);
TCPSTATES_INC(TCPS_CLOSED);
out:
TCP_PROBE2(debug__user, tp, PRU_ATTACH);
return (error);
}
/*
* tcp_usr_detach is called when the socket layer loses its final reference
* to the socket, be it a file descriptor reference, a reference from TCP,
* etc. At this point, there is only one case in which we will keep around
* inpcb state: time wait.
*/
static void
tcp_usr_detach(struct socket *so)
{
struct inpcb *inp;
struct tcpcb *tp;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
INP_WLOCK(inp);
KASSERT(so->so_pcb == inp && inp->inp_socket == so,
("%s: socket %p inp %p mismatch", __func__, so, inp));
tp = intotcpcb(inp);
KASSERT(inp->inp_flags & INP_DROPPED ||
tp->t_state < TCPS_SYN_SENT,
("%s: inp %p not dropped or embryonic", __func__, inp));
tcp_discardcb(tp);
in_pcbdetach(inp);
in_pcbfree(inp);
}
#ifdef INET
/*
* Give the socket an address.
*/
static int
tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
struct sockaddr_in *sinp;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (EINVAL);
}
tp = intotcpcb(inp);
sinp = (struct sockaddr_in *)nam;
if (nam->sa_family != AF_INET) {
/*
* Preserve compatibility with old programs.
*/
if (nam->sa_family != AF_UNSPEC ||
nam->sa_len < offsetof(struct sockaddr_in, sin_zero) ||
sinp->sin_addr.s_addr != INADDR_ANY) {
error = EAFNOSUPPORT;
goto out;
}
nam->sa_family = AF_INET;
}
if (nam->sa_len != sizeof(*sinp)) {
error = EINVAL;
goto out;
}
/*
* Must check for multicast addresses and disallow binding
* to them.
*/
if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
error = EAFNOSUPPORT;
goto out;
}
INP_HASH_WLOCK(&V_tcbinfo);
error = in_pcbbind(inp, sinp, td->td_ucred);
INP_HASH_WUNLOCK(&V_tcbinfo);
out:
tcp_bblog_pru(tp, PRU_BIND, error);
TCP_PROBE2(debug__user, tp, PRU_BIND);
INP_WUNLOCK(inp);
return (error);
}
#endif /* INET */
#ifdef INET6
static int
tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
struct sockaddr_in6 *sin6;
u_char vflagsav;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (EINVAL);
}
tp = intotcpcb(inp);
vflagsav = inp->inp_vflag;
sin6 = (struct sockaddr_in6 *)nam;
if (nam->sa_family != AF_INET6) {
error = EAFNOSUPPORT;
goto out;
}
if (nam->sa_len != sizeof(*sin6)) {
error = EINVAL;
goto out;
}
/*
* Must check for multicast addresses and disallow binding
* to them.
*/
if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
error = EAFNOSUPPORT;
goto out;
}
INP_HASH_WLOCK(&V_tcbinfo);
inp->inp_vflag &= ~INP_IPV4;
inp->inp_vflag |= INP_IPV6;
#ifdef INET
if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
inp->inp_vflag |= INP_IPV4;
else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
struct sockaddr_in sin;
in6_sin6_2_sin(&sin, sin6);
if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
error = EAFNOSUPPORT;
INP_HASH_WUNLOCK(&V_tcbinfo);
goto out;
}
inp->inp_vflag |= INP_IPV4;
inp->inp_vflag &= ~INP_IPV6;
error = in_pcbbind(inp, &sin, td->td_ucred);
INP_HASH_WUNLOCK(&V_tcbinfo);
goto out;
}
}
#endif
error = in6_pcbbind(inp, sin6, td->td_ucred);
INP_HASH_WUNLOCK(&V_tcbinfo);
out:
if (error != 0)
inp->inp_vflag = vflagsav;
tcp_bblog_pru(tp, PRU_BIND, error);
TCP_PROBE2(debug__user, tp, PRU_BIND);
INP_WUNLOCK(inp);
return (error);
}
#endif /* INET6 */
#ifdef INET
/*
* Prepare to accept connections.
*/
static int
tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (EINVAL);
}
tp = intotcpcb(inp);
SOCK_LOCK(so);
error = solisten_proto_check(so);
if (error != 0) {
SOCK_UNLOCK(so);
goto out;
}
if (inp->inp_lport == 0) {
INP_HASH_WLOCK(&V_tcbinfo);
error = in_pcbbind(inp, NULL, td->td_ucred);
INP_HASH_WUNLOCK(&V_tcbinfo);
}
if (error == 0) {
tcp_state_change(tp, TCPS_LISTEN);
solisten_proto(so, backlog);
#ifdef TCP_OFFLOAD
if ((so->so_options & SO_NO_OFFLOAD) == 0)
tcp_offload_listen_start(tp);
#endif
} else {
solisten_proto_abort(so);
}
SOCK_UNLOCK(so);
if (IS_FASTOPEN(tp->t_flags))
tp->t_tfo_pending = tcp_fastopen_alloc_counter();
out:
tcp_bblog_pru(tp, PRU_LISTEN, error);
TCP_PROBE2(debug__user, tp, PRU_LISTEN);
INP_WUNLOCK(inp);
return (error);
}
#endif /* INET */
#ifdef INET6
static int
tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
u_char vflagsav;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (EINVAL);
}
tp = intotcpcb(inp);
vflagsav = inp->inp_vflag;
SOCK_LOCK(so);
error = solisten_proto_check(so);
if (error != 0) {
SOCK_UNLOCK(so);
goto out;
}
INP_HASH_WLOCK(&V_tcbinfo);
if (inp->inp_lport == 0) {
inp->inp_vflag &= ~INP_IPV4;
if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
inp->inp_vflag |= INP_IPV4;
error = in6_pcbbind(inp, NULL, td->td_ucred);
}
INP_HASH_WUNLOCK(&V_tcbinfo);
if (error == 0) {
tcp_state_change(tp, TCPS_LISTEN);
solisten_proto(so, backlog);
#ifdef TCP_OFFLOAD
if ((so->so_options & SO_NO_OFFLOAD) == 0)
tcp_offload_listen_start(tp);
#endif
} else {
solisten_proto_abort(so);
}
SOCK_UNLOCK(so);
if (IS_FASTOPEN(tp->t_flags))
tp->t_tfo_pending = tcp_fastopen_alloc_counter();
if (error != 0)
inp->inp_vflag = vflagsav;
out:
tcp_bblog_pru(tp, PRU_LISTEN, error);
TCP_PROBE2(debug__user, tp, PRU_LISTEN);
INP_WUNLOCK(inp);
return (error);
}
#endif /* INET6 */
#ifdef INET
/*
* Initiate connection to peer.
* Create a template for use in transmissions on this connection.
* Enter SYN_SENT state, and mark socket as connecting.
* Start keep-alive timer, and seed output sequence space.
* Send initial segment on connection.
*/
static int
tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
{
struct epoch_tracker et;
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
struct sockaddr_in *sinp;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNREFUSED);
}
tp = intotcpcb(inp);
sinp = (struct sockaddr_in *)nam;
if (nam->sa_family != AF_INET) {
error = EAFNOSUPPORT;
goto out;
}
if (nam->sa_len != sizeof (*sinp)) {
error = EINVAL;
goto out;
}
/*
* Must disallow TCP ``connections'' to multicast addresses.
*/
if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
error = EAFNOSUPPORT;
goto out;
}
if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
error = EACCES;
goto out;
}
if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
goto out;
if (SOLISTENING(so)) {
error = EOPNOTSUPP;
goto out;
}
NET_EPOCH_ENTER(et);
if ((error = tcp_connect(tp, sinp, td)) != 0)
goto out_in_epoch;
#ifdef TCP_OFFLOAD
if (registered_toedevs > 0 &&
(so->so_options & SO_NO_OFFLOAD) == 0 &&
(error = tcp_offload_connect(so, nam)) == 0)
goto out_in_epoch;
#endif
tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
error = tcp_output(tp);
KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
out_in_epoch:
NET_EPOCH_EXIT(et);
out:
tcp_bblog_pru(tp, PRU_CONNECT, error);
TCP_PROBE2(debug__user, tp, PRU_CONNECT);
INP_WUNLOCK(inp);
return (error);
}
#endif /* INET */
#ifdef INET6
static int
tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
{
struct epoch_tracker et;
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
struct sockaddr_in6 *sin6;
u_int8_t incflagsav;
u_char vflagsav;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNREFUSED);
}
tp = intotcpcb(inp);
vflagsav = inp->inp_vflag;
incflagsav = inp->inp_inc.inc_flags;
sin6 = (struct sockaddr_in6 *)nam;
if (nam->sa_family != AF_INET6) {
error = EAFNOSUPPORT;
goto out;
}
if (nam->sa_len != sizeof (*sin6)) {
error = EINVAL;
goto out;
}
/*
* Must disallow TCP ``connections'' to multicast addresses.
*/
if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
error = EAFNOSUPPORT;
goto out;
}
if (SOLISTENING(so)) {
error = EINVAL;
goto out;
}
#ifdef INET
/*
* XXXRW: Some confusion: V4/V6 flags relate to binding, and
* therefore probably require the hash lock, which isn't held here.
* Is this a significant problem?
*/
if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
struct sockaddr_in sin;
if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
error = EINVAL;
goto out;
}
if ((inp->inp_vflag & INP_IPV4) == 0) {
error = EAFNOSUPPORT;
goto out;
}
in6_sin6_2_sin(&sin, sin6);
if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
error = EAFNOSUPPORT;
goto out;
}
if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
error = EACCES;
goto out;
}
if ((error = prison_remote_ip4(td->td_ucred,
&sin.sin_addr)) != 0)
goto out;
inp->inp_vflag |= INP_IPV4;
inp->inp_vflag &= ~INP_IPV6;
NET_EPOCH_ENTER(et);
if ((error = tcp_connect(tp, &sin, td)) != 0)
goto out_in_epoch;
#ifdef TCP_OFFLOAD
if (registered_toedevs > 0 &&
(so->so_options & SO_NO_OFFLOAD) == 0 &&
(error = tcp_offload_connect(so, nam)) == 0)
goto out_in_epoch;
#endif
error = tcp_output(tp);
goto out_in_epoch;
} else {
if ((inp->inp_vflag & INP_IPV6) == 0) {
error = EAFNOSUPPORT;
goto out;
}
}
#endif
if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
goto out;
inp->inp_vflag &= ~INP_IPV4;
inp->inp_vflag |= INP_IPV6;
inp->inp_inc.inc_flags |= INC_ISIPV6;
NET_EPOCH_ENTER(et);
if ((error = tcp6_connect(tp, sin6, td)) != 0)
goto out_in_epoch;
#ifdef TCP_OFFLOAD
if (registered_toedevs > 0 &&
(so->so_options & SO_NO_OFFLOAD) == 0 &&
(error = tcp_offload_connect(so, nam)) == 0)
goto out_in_epoch;
#endif
tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
error = tcp_output(tp);
out_in_epoch:
NET_EPOCH_EXIT(et);
out:
KASSERT(error >= 0, ("TCP stack %s requested tcp_drop(%p) at connect()"
", error code %d", tp->t_fb->tfb_tcp_block_name, tp, -error));
/*
* If the implicit bind in the connect call fails, restore
* the flags we modified.
*/
if (error != 0 && inp->inp_lport == 0) {
inp->inp_vflag = vflagsav;
inp->inp_inc.inc_flags = incflagsav;
}
tcp_bblog_pru(tp, PRU_CONNECT, error);
TCP_PROBE2(debug__user, tp, PRU_CONNECT);
INP_WUNLOCK(inp);
return (error);
}
#endif /* INET6 */
/*
* Initiate disconnect from peer.
* If connection never passed embryonic stage, just drop;
* else if don't need to let data drain, then can just drop anyways,
* else have to begin TCP shutdown process: mark socket disconnecting,
* drain unread data, state switch to reflect user close, and
* send segment (e.g. FIN) to peer. Socket will be really disconnected
* when peer sends FIN and acks ours.
*
* SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
*/
static int
tcp_usr_disconnect(struct socket *so)
{
struct inpcb *inp;
struct tcpcb *tp = NULL;
struct epoch_tracker et;
int error = 0;
NET_EPOCH_ENTER(et);
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
NET_EPOCH_EXIT(et);
return (ECONNRESET);
}
tp = intotcpcb(inp);
if (tp->t_state == TCPS_TIME_WAIT)
goto out;
tcp_disconnect(tp);
out:
tcp_bblog_pru(tp, PRU_DISCONNECT, error);
TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
INP_WUNLOCK(inp);
NET_EPOCH_EXIT(et);
return (error);
}
#ifdef INET
/*
* Accept a connection. Essentially all the work is done at higher levels;
* just return the address of the peer, storing through addr.
*/
static int
tcp_usr_accept(struct socket *so, struct sockaddr **nam)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
struct in_addr addr;
in_port_t port = 0;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNABORTED);
}
tp = intotcpcb(inp);
if (so->so_state & SS_ISDISCONNECTED) {
error = ECONNABORTED;
goto out;
}
/*
* We inline in_getpeeraddr and COMMON_END here, so that we can
* copy the data of interest and defer the malloc until after we
* release the lock.
*/
port = inp->inp_fport;
addr = inp->inp_faddr;
out:
tcp_bblog_pru(tp, PRU_ACCEPT, error);
TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
INP_WUNLOCK(inp);
if (error == 0)
*nam = in_sockaddr(port, &addr);
return error;
}
#endif /* INET */
#ifdef INET6
static int
tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
{
struct inpcb *inp;
int error = 0;
struct tcpcb *tp;
struct in_addr addr;
struct in6_addr addr6;
struct epoch_tracker et;
in_port_t port = 0;
int v4 = 0;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
NET_EPOCH_ENTER(et); /* XXXMT Why is this needed? */
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
NET_EPOCH_EXIT(et);
return (ECONNABORTED);
}
tp = intotcpcb(inp);
if (so->so_state & SS_ISDISCONNECTED) {
error = ECONNABORTED;
goto out;
}
/*
* We inline in6_mapped_peeraddr and COMMON_END here, so that we can
* copy the data of interest and defer the malloc until after we
* release the lock.
*/
if (inp->inp_vflag & INP_IPV4) {
v4 = 1;
port = inp->inp_fport;
addr = inp->inp_faddr;
} else {
port = inp->inp_fport;
addr6 = inp->in6p_faddr;
}
out:
tcp_bblog_pru(tp, PRU_ACCEPT, error);
TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
INP_WUNLOCK(inp);
NET_EPOCH_EXIT(et);
if (error == 0) {
if (v4)
*nam = in6_v4mapsin6_sockaddr(port, &addr);
else
*nam = in6_sockaddr(port, &addr6);
}
return error;
}
#endif /* INET6 */
/*
* Mark the connection as being incapable of further output.
*/
static int
tcp_usr_shutdown(struct socket *so)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
struct epoch_tracker et;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
NET_EPOCH_ENTER(et);
socantsendmore(so);
tcp_usrclosed(tp);
if (!(inp->inp_flags & INP_DROPPED))
error = tcp_output_nodrop(tp);
tcp_bblog_pru(tp, PRU_SHUTDOWN, error);
TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
error = tcp_unlock_or_drop(tp, error);
NET_EPOCH_EXIT(et);
return (error);
}
/*
* After a receive, possibly send window update to peer.
*/
static int
tcp_usr_rcvd(struct socket *so, int flags)
{
struct epoch_tracker et;
struct inpcb *inp;
struct tcpcb *tp;
int outrv = 0, error = 0;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
NET_EPOCH_ENTER(et);
/*
* For passively-created TFO connections, don't attempt a window
* update while still in SYN_RECEIVED as this may trigger an early
* SYN|ACK. It is preferable to have the SYN|ACK be sent along with
* application response data, or failing that, when the DELACK timer
* expires.
*/
if (IS_FASTOPEN(tp->t_flags) &&
(tp->t_state == TCPS_SYN_RECEIVED))
goto out;
#ifdef TCP_OFFLOAD
if (tp->t_flags & TF_TOE)
tcp_offload_rcvd(tp);
else
#endif
outrv = tcp_output_nodrop(tp);
out:
tcp_bblog_pru(tp, PRU_RCVD, error);
TCP_PROBE2(debug__user, tp, PRU_RCVD);
(void) tcp_unlock_or_drop(tp, outrv);
NET_EPOCH_EXIT(et);
return (error);
}
/*
* Do a send by putting data in output queue and updating urgent
* marker if URG set. Possibly send more data. Unlike the other
* pru_*() routines, the mbuf chains are our responsibility. We
* must either enqueue them or free them. The other pru_* routines
* generally are caller-frees.
*/
static int
tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
struct sockaddr *nam, struct mbuf *control, struct thread *td)
{
struct epoch_tracker et;
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
#ifdef INET
#ifdef INET6
struct sockaddr_in sin;
#endif
struct sockaddr_in *sinp;
#endif
#ifdef INET6
struct sockaddr_in6 *sin6;
int isipv6;
#endif
u_int8_t incflagsav;
u_char vflagsav;
bool restoreflags;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
if (m != NULL && (flags & PRUS_NOTREADY) == 0)
m_freem(m);
INP_WUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
vflagsav = inp->inp_vflag;
incflagsav = inp->inp_inc.inc_flags;
restoreflags = false;
NET_EPOCH_ENTER(et);
if (control != NULL) {
/* TCP doesn't do control messages (rights, creds, etc) */
if (control->m_len > 0) {
m_freem(control);
error = EINVAL;
goto out;
}
m_freem(control); /* empty control, just free it */
}
if ((flags & PRUS_OOB) != 0 &&
(error = tcp_pru_options_support(tp, PRUS_OOB)) != 0)
goto out;
if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
if (tp->t_state == TCPS_LISTEN) {
error = EINVAL;
goto out;
}
switch (nam->sa_family) {
#ifdef INET
case AF_INET:
sinp = (struct sockaddr_in *)nam;
if (sinp->sin_len != sizeof(struct sockaddr_in)) {
error = EINVAL;
goto out;
}
if ((inp->inp_vflag & INP_IPV6) != 0) {
error = EAFNOSUPPORT;
goto out;
}
if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
error = EAFNOSUPPORT;
goto out;
}
if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
error = EACCES;
goto out;
}
if ((error = prison_remote_ip4(td->td_ucred,
&sinp->sin_addr)))
goto out;
#ifdef INET6
isipv6 = 0;
#endif
break;
#endif /* INET */
#ifdef INET6
case AF_INET6:
sin6 = (struct sockaddr_in6 *)nam;
if (sin6->sin6_len != sizeof(*sin6)) {
error = EINVAL;
goto out;
}
if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
error = EAFNOSUPPORT;
goto out;
}
if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
error = EAFNOSUPPORT;
goto out;
}
if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
#ifdef INET
if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
error = EINVAL;
goto out;
}
if ((inp->inp_vflag & INP_IPV4) == 0) {
error = EAFNOSUPPORT;
goto out;
}
restoreflags = true;
inp->inp_vflag &= ~INP_IPV6;
sinp = &sin;
in6_sin6_2_sin(sinp, sin6);
if (IN_MULTICAST(
ntohl(sinp->sin_addr.s_addr))) {
error = EAFNOSUPPORT;
goto out;
}
if ((error = prison_remote_ip4(td->td_ucred,
&sinp->sin_addr)))
goto out;
isipv6 = 0;
#else /* !INET */
error = EAFNOSUPPORT;
goto out;
#endif /* INET */
} else {
if ((inp->inp_vflag & INP_IPV6) == 0) {
error = EAFNOSUPPORT;
goto out;
}
restoreflags = true;
inp->inp_vflag &= ~INP_IPV4;
inp->inp_inc.inc_flags |= INC_ISIPV6;
if ((error = prison_remote_ip6(td->td_ucred,
&sin6->sin6_addr)))
goto out;
isipv6 = 1;
}
break;
#endif /* INET6 */
default:
error = EAFNOSUPPORT;
goto out;
}
}
if (!(flags & PRUS_OOB)) {
if (tp->t_acktime == 0)
tp->t_acktime = ticks;
sbappendstream(&so->so_snd, m, flags);
m = NULL;
if (nam && tp->t_state < TCPS_SYN_SENT) {
KASSERT(tp->t_state == TCPS_CLOSED,
("%s: tp %p is listening", __func__, tp));
/*
* Do implied connect if not yet connected,
* initialize window to default value, and
* initialize maxseg using peer's cached MSS.
*/
#ifdef INET6
if (isipv6)
error = tcp6_connect(tp, sin6, td);
#endif /* INET6 */
#if defined(INET6) && defined(INET)
else
#endif
#ifdef INET
error = tcp_connect(tp, sinp, td);
#endif
/*
* The bind operation in tcp_connect succeeded. We
* no longer want to restore the flags if later
* operations fail.
*/
if (error == 0 || inp->inp_lport != 0)
restoreflags = false;
if (error) {
/* m is freed if PRUS_NOTREADY is unset. */
sbflush(&so->so_snd);
goto out;
}
if (IS_FASTOPEN(tp->t_flags))
tcp_fastopen_connect(tp);
else {
tp->snd_wnd = TTCP_CLIENT_SND_WND;
tcp_mss(tp, -1);
}
}
if (flags & PRUS_EOF) {
/*
* Close the send side of the connection after
* the data is sent.
*/
socantsendmore(so);
tcp_usrclosed(tp);
}
if (TCPS_HAVEESTABLISHED(tp->t_state) &&
((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
(tp->t_fbyte_out == 0) &&
(so->so_snd.sb_ccc > 0)) {
tp->t_fbyte_out = ticks;
if (tp->t_fbyte_out == 0)
tp->t_fbyte_out = 1;
if (tp->t_fbyte_out && tp->t_fbyte_in)
tp->t_flags2 |= TF2_FBYTES_COMPLETE;
}
if (!(inp->inp_flags & INP_DROPPED) &&
!(flags & PRUS_NOTREADY)) {
if (flags & PRUS_MORETOCOME)
tp->t_flags |= TF_MORETOCOME;
error = tcp_output_nodrop(tp);
if (flags & PRUS_MORETOCOME)
tp->t_flags &= ~TF_MORETOCOME;
}
} else {
/*
* XXXRW: PRUS_EOF not implemented with PRUS_OOB?
*/
SOCKBUF_LOCK(&so->so_snd);
if (sbspace(&so->so_snd) < -512) {
SOCKBUF_UNLOCK(&so->so_snd);
error = ENOBUFS;
goto out;
}
/*
* According to RFC961 (Assigned Protocols),
* the urgent pointer points to the last octet
* of urgent data. We continue, however,
* to consider it to indicate the first octet
* of data past the urgent section.
* Otherwise, snd_up should be one lower.
*/
if (tp->t_acktime == 0)
tp->t_acktime = ticks;
sbappendstream_locked(&so->so_snd, m, flags);
SOCKBUF_UNLOCK(&so->so_snd);
m = NULL;
if (nam && tp->t_state < TCPS_SYN_SENT) {
/*
* Do implied connect if not yet connected,
* initialize window to default value, and
* initialize maxseg using peer's cached MSS.
*/
/*
* Not going to contemplate SYN|URG
*/
if (IS_FASTOPEN(tp->t_flags))
tp->t_flags &= ~TF_FASTOPEN;
#ifdef INET6
if (isipv6)
error = tcp6_connect(tp, sin6, td);
#endif /* INET6 */
#if defined(INET6) && defined(INET)
else
#endif
#ifdef INET
error = tcp_connect(tp, sinp, td);
#endif
/*
* The bind operation in tcp_connect succeeded. We
* no longer want to restore the flags if later
* operations fail.
*/
if (error == 0 || inp->inp_lport != 0)
restoreflags = false;
if (error != 0) {
/* m is freed if PRUS_NOTREADY is unset. */
sbflush(&so->so_snd);
goto out;
}
tp->snd_wnd = TTCP_CLIENT_SND_WND;
tcp_mss(tp, -1);
}
tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
if ((flags & PRUS_NOTREADY) == 0) {
tp->t_flags |= TF_FORCEDATA;
error = tcp_output_nodrop(tp);
tp->t_flags &= ~TF_FORCEDATA;
}
}
TCP_LOG_EVENT(tp, NULL,
&inp->inp_socket->so_rcv,
&inp->inp_socket->so_snd,
TCP_LOG_USERSEND, error,
0, NULL, false);
out:
/*
* In case of PRUS_NOTREADY, the caller or tcp_usr_ready() is
* responsible for freeing memory.
*/
if (m != NULL && (flags & PRUS_NOTREADY) == 0)
m_freem(m);
/*
* If the request was unsuccessful and we changed flags,
* restore the original flags.
*/
if (error != 0 && restoreflags) {
inp->inp_vflag = vflagsav;
inp->inp_inc.inc_flags = incflagsav;
}
tcp_bblog_pru(tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND), error);
TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
error = tcp_unlock_or_drop(tp, error);
NET_EPOCH_EXIT(et);
return (error);
}
static int
tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
{
struct epoch_tracker et;
struct inpcb *inp;
struct tcpcb *tp;
int error;
inp = sotoinpcb(so);
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
mb_free_notready(m, count);
return (ECONNRESET);
}
tp = intotcpcb(inp);
SOCKBUF_LOCK(&so->so_snd);
error = sbready(&so->so_snd, m, count);
SOCKBUF_UNLOCK(&so->so_snd);
if (error) {
INP_WUNLOCK(inp);
return (error);
}
NET_EPOCH_ENTER(et);
error = tcp_output_unlock(tp);
NET_EPOCH_EXIT(et);
return (error);
}
/*
* Abort the TCP. Drop the connection abruptly.
*/
static void
tcp_usr_abort(struct socket *so)
{
struct inpcb *inp;
struct tcpcb *tp;
struct epoch_tracker et;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
NET_EPOCH_ENTER(et);
INP_WLOCK(inp);
KASSERT(inp->inp_socket != NULL,
("tcp_usr_abort: inp_socket == NULL"));
/*
* If we still have full TCP state, and we're not dropped, drop.
*/
if (!(inp->inp_flags & INP_DROPPED)) {
tp = intotcpcb(inp);
tp = tcp_drop(tp, ECONNABORTED);
if (tp == NULL)
goto dropped;
tcp_bblog_pru(tp, PRU_ABORT, 0);
TCP_PROBE2(debug__user, tp, PRU_ABORT);
}
if (!(inp->inp_flags & INP_DROPPED)) {
soref(so);
inp->inp_flags |= INP_SOCKREF;
}
INP_WUNLOCK(inp);
dropped:
NET_EPOCH_EXIT(et);
}
/*
* TCP socket is closed. Start friendly disconnect.
*/
static void
tcp_usr_close(struct socket *so)
{
struct inpcb *inp;
struct tcpcb *tp;
struct epoch_tracker et;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
NET_EPOCH_ENTER(et);
INP_WLOCK(inp);
KASSERT(inp->inp_socket != NULL,
("tcp_usr_close: inp_socket == NULL"));
/*
* If we are still connected and we're not dropped, initiate
* a disconnect.
*/
if (!(inp->inp_flags & INP_DROPPED)) {
tp = intotcpcb(inp);
if (tp->t_state != TCPS_TIME_WAIT) {
tp->t_flags |= TF_CLOSED;
tcp_disconnect(tp);
tcp_bblog_pru(tp, PRU_CLOSE, 0);
TCP_PROBE2(debug__user, tp, PRU_CLOSE);
}
}
if (!(inp->inp_flags & INP_DROPPED)) {
soref(so);
inp->inp_flags |= INP_SOCKREF;
}
INP_WUNLOCK(inp);
NET_EPOCH_EXIT(et);
}
static int
tcp_pru_options_support(struct tcpcb *tp, int flags)
{
/*
* If the specific TCP stack has a pru_options
* specified then it does not always support
* all the PRU_XX options and we must ask it.
* If the function is not specified then all
* of the PRU_XX options are supported.
*/
int ret = 0;
if (tp->t_fb->tfb_pru_options) {
ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
}
return (ret);
}
/*
* Receive out-of-band data.
*/
static int
tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
{
int error = 0;
struct inpcb *inp;
struct tcpcb *tp;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
tp = intotcpcb(inp);
error = tcp_pru_options_support(tp, PRUS_OOB);
if (error) {
goto out;
}
if ((so->so_oobmark == 0 &&
(so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
so->so_options & SO_OOBINLINE ||
tp->t_oobflags & TCPOOB_HADDATA) {
error = EINVAL;
goto out;
}
if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
error = EWOULDBLOCK;
goto out;
}
m->m_len = 1;
*mtod(m, caddr_t) = tp->t_iobc;
if ((flags & MSG_PEEK) == 0)
tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
out:
tcp_bblog_pru(tp, PRU_RCVOOB, error);
TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
INP_WUNLOCK(inp);
return (error);
}
#ifdef INET
struct protosw tcp_protosw = {
.pr_type = SOCK_STREAM,
.pr_protocol = IPPROTO_TCP,
.pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL | PR_WANTRCVD |
PR_CAPATTACH,
.pr_ctloutput = tcp_ctloutput,
.pr_abort = tcp_usr_abort,
.pr_accept = tcp_usr_accept,
.pr_attach = tcp_usr_attach,
.pr_bind = tcp_usr_bind,
.pr_connect = tcp_usr_connect,
.pr_control = in_control,
.pr_detach = tcp_usr_detach,
.pr_disconnect = tcp_usr_disconnect,
.pr_listen = tcp_usr_listen,
.pr_peeraddr = in_getpeeraddr,
.pr_rcvd = tcp_usr_rcvd,
.pr_rcvoob = tcp_usr_rcvoob,
.pr_send = tcp_usr_send,
.pr_ready = tcp_usr_ready,
.pr_shutdown = tcp_usr_shutdown,
.pr_sockaddr = in_getsockaddr,
.pr_sosetlabel = in_pcbsosetlabel,
.pr_close = tcp_usr_close,
};
#endif /* INET */
#ifdef INET6
struct protosw tcp6_protosw = {
.pr_type = SOCK_STREAM,
.pr_protocol = IPPROTO_TCP,
.pr_flags = PR_CONNREQUIRED | PR_IMPLOPCL |PR_WANTRCVD |
PR_CAPATTACH,
.pr_ctloutput = tcp_ctloutput,
.pr_abort = tcp_usr_abort,
.pr_accept = tcp6_usr_accept,
.pr_attach = tcp_usr_attach,
.pr_bind = tcp6_usr_bind,
.pr_connect = tcp6_usr_connect,
.pr_control = in6_control,
.pr_detach = tcp_usr_detach,
.pr_disconnect = tcp_usr_disconnect,
.pr_listen = tcp6_usr_listen,
.pr_peeraddr = in6_mapped_peeraddr,
.pr_rcvd = tcp_usr_rcvd,
.pr_rcvoob = tcp_usr_rcvoob,
.pr_send = tcp_usr_send,
.pr_ready = tcp_usr_ready,
.pr_shutdown = tcp_usr_shutdown,
.pr_sockaddr = in6_mapped_sockaddr,
.pr_sosetlabel = in_pcbsosetlabel,
.pr_close = tcp_usr_close,
};
#endif /* INET6 */
#ifdef INET
/*
* Common subroutine to open a TCP connection to remote host specified
* by struct sockaddr_in. Call in_pcbconnect() to choose local host address
* and assign a local port number and install the inpcb into the hash.
* Initialize connection parameters and enter SYN-SENT state.
*/
static int
tcp_connect(struct tcpcb *tp, struct sockaddr_in *sin, struct thread *td)
{
struct inpcb *inp = tptoinpcb(tp);
struct socket *so = tptosocket(tp);
int error;
NET_EPOCH_ASSERT();
INP_WLOCK_ASSERT(inp);
if (__predict_false((so->so_state &
(SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING |
SS_ISDISCONNECTED)) != 0))
return (EISCONN);
INP_HASH_WLOCK(&V_tcbinfo);
error = in_pcbconnect(inp, sin, td->td_ucred, true);
INP_HASH_WUNLOCK(&V_tcbinfo);
if (error != 0)
return (error);
/*
* Compute window scaling to request:
* Scale to fit into sweet spot. See tcp_syncache.c.
* XXX: This should move to tcp_output().
*/
while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
(TCP_MAXWIN << tp->request_r_scale) < sb_max)
tp->request_r_scale++;
soisconnecting(so);
TCPSTAT_INC(tcps_connattempt);
tcp_state_change(tp, TCPS_SYN_SENT);
tp->iss = tcp_new_isn(&inp->inp_inc);
if (tp->t_flags & TF_REQ_TSTMP)
tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
tcp_sendseqinit(tp);
return (0);
}
#endif /* INET */
#ifdef INET6
static int
tcp6_connect(struct tcpcb *tp, struct sockaddr_in6 *sin6, struct thread *td)
{
struct inpcb *inp = tptoinpcb(tp);
struct socket *so = tptosocket(tp);
int error;
NET_EPOCH_ASSERT();
INP_WLOCK_ASSERT(inp);
if (__predict_false((so->so_state &
(SS_ISCONNECTING | SS_ISCONNECTED)) != 0))
return (EISCONN);
INP_HASH_WLOCK(&V_tcbinfo);
error = in6_pcbconnect(inp, sin6, td->td_ucred, true);
INP_HASH_WUNLOCK(&V_tcbinfo);
if (error != 0)
return (error);
/* Compute window scaling to request. */
while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
(TCP_MAXWIN << tp->request_r_scale) < sb_max)
tp->request_r_scale++;
soisconnecting(so);
TCPSTAT_INC(tcps_connattempt);
tcp_state_change(tp, TCPS_SYN_SENT);
tp->iss = tcp_new_isn(&inp->inp_inc);
if (tp->t_flags & TF_REQ_TSTMP)
tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
tcp_sendseqinit(tp);
return (0);
}
#endif /* INET6 */
/*
* Export TCP internal state information via a struct tcp_info, based on the
* Linux 2.6 API. Not ABI compatible as our constants are mapped differently
* (TCP state machine, etc). We export all information using FreeBSD-native
* constants -- for example, the numeric values for tcpi_state will differ
* from Linux.
*/
-static void
-tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
+void
+tcp_fill_info(const struct tcpcb *tp, struct tcp_info *ti)
{
- INP_WLOCK_ASSERT(tptoinpcb(tp));
+ INP_LOCK_ASSERT(tptoinpcb(tp));
bzero(ti, sizeof(*ti));
ti->tcpi_state = tp->t_state;
if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
if (tp->t_flags & TF_SACK_PERMIT)
ti->tcpi_options |= TCPI_OPT_SACK;
if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
ti->tcpi_options |= TCPI_OPT_WSCALE;
ti->tcpi_snd_wscale = tp->snd_scale;
ti->tcpi_rcv_wscale = tp->rcv_scale;
}
switch (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) {
case TF2_ECN_PERMIT:
ti->tcpi_options |= TCPI_OPT_ECN;
break;
case TF2_ACE_PERMIT:
/* FALLTHROUGH */
case TF2_ECN_PERMIT | TF2_ACE_PERMIT:
ti->tcpi_options |= TCPI_OPT_ACE;
break;
default:
break;
}
if (IS_FASTOPEN(tp->t_flags))
ti->tcpi_options |= TCPI_OPT_TFO;
ti->tcpi_rto = tp->t_rxtcur * tick;
ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
ti->tcpi_snd_cwnd = tp->snd_cwnd;
/*
* FreeBSD-specific extension fields for tcp_info.
*/
ti->tcpi_rcv_space = tp->rcv_wnd;
ti->tcpi_rcv_nxt = tp->rcv_nxt;
ti->tcpi_snd_wnd = tp->snd_wnd;
ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
ti->tcpi_snd_nxt = tp->snd_nxt;
ti->tcpi_snd_mss = tp->t_maxseg;
ti->tcpi_rcv_mss = tp->t_maxseg;
ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
ti->tcpi_snd_zerowin = tp->t_sndzerowin;
#ifdef TCP_OFFLOAD
if (tp->t_flags & TF_TOE) {
ti->tcpi_options |= TCPI_OPT_TOE;
tcp_offload_tcp_info(tp, ti);
}
#endif
/*
* AccECN related counters.
*/
if ((tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) ==
(TF2_ECN_PERMIT | TF2_ACE_PERMIT))
/*
* Internal counter starts at 5 for AccECN
* but 0 for RFC3168 ECN.
*/
ti->tcpi_delivered_ce = tp->t_scep - 5;
else
ti->tcpi_delivered_ce = tp->t_scep;
ti->tcpi_received_ce = tp->t_rcep;
}
/*
* tcp_ctloutput() must drop the inpcb lock before performing copyin on
* socket option arguments. When it re-acquires the lock after the copy, it
* has to revalidate that the connection is still valid for the socket
* option.
*/
#define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \
INP_WLOCK(inp); \
if (inp->inp_flags & INP_DROPPED) { \
INP_WUNLOCK(inp); \
cleanup; \
return (ECONNRESET); \
} \
tp = intotcpcb(inp); \
} while(0)
#define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
int
tcp_ctloutput_set(struct inpcb *inp, struct sockopt *sopt)
{
struct socket *so = inp->inp_socket;
struct tcpcb *tp = intotcpcb(inp);
int error = 0;
MPASS(sopt->sopt_dir == SOPT_SET);
INP_WLOCK_ASSERT(inp);
KASSERT((inp->inp_flags & INP_DROPPED) == 0,
("inp_flags == %x", inp->inp_flags));
KASSERT(so != NULL, ("inp_socket == NULL"));
if (sopt->sopt_level != IPPROTO_TCP) {
INP_WUNLOCK(inp);
#ifdef INET6
if (inp->inp_vflag & INP_IPV6PROTO)
error = ip6_ctloutput(so, sopt);
#endif
#if defined(INET6) && defined(INET)
else
#endif
#ifdef INET
error = ip_ctloutput(so, sopt);
#endif
/*
* When an IP-level socket option affects TCP, pass control
* down to stack tfb_tcp_ctloutput, otherwise return what
* IP level returned.
*/
switch (sopt->sopt_level) {
#ifdef INET6
case IPPROTO_IPV6:
if ((inp->inp_vflag & INP_IPV6PROTO) == 0)
return (error);
switch (sopt->sopt_name) {
case IPV6_TCLASS:
/* Notify tcp stacks that care (e.g. RACK). */
break;
case IPV6_USE_MIN_MTU:
/* Update t_maxseg accordingly. */
break;
default:
return (error);
}
break;
#endif
#ifdef INET
case IPPROTO_IP:
switch (sopt->sopt_name) {
case IP_TOS:
inp->inp_ip_tos &= ~IPTOS_ECN_MASK;
break;
case IP_TTL:
/* Notify tcp stacks that care (e.g. RACK). */
break;
default:
return (error);
}
break;
#endif
default:
return (error);
}
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
} else if (sopt->sopt_name == TCP_FUNCTION_BLK) {
/*
* Protect the TCP option TCP_FUNCTION_BLK so
* that a sub-function can *never* overwrite this.
*/
struct tcp_function_set fsn;
struct tcp_function_block *blk;
void *ptr = NULL;
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &fsn, sizeof fsn, sizeof fsn);
if (error)
return (error);
INP_WLOCK(inp);
tp = intotcpcb(inp);
blk = find_and_ref_tcp_functions(&fsn);
if (blk == NULL) {
INP_WUNLOCK(inp);
return (ENOENT);
}
if (tp->t_fb == blk) {
/* You already have this */
refcount_release(&blk->tfb_refcnt);
INP_WUNLOCK(inp);
return (0);
}
if (tp->t_state != TCPS_CLOSED) {
/*
* The user has advanced the state
* past the initial point, we may not
* be able to switch.
*/
if (blk->tfb_tcp_handoff_ok != NULL) {
/*
* Does the stack provide a
* query mechanism, if so it may
* still be possible?
*/
error = (*blk->tfb_tcp_handoff_ok)(tp);
} else
error = EINVAL;
if (error) {
refcount_release(&blk->tfb_refcnt);
INP_WUNLOCK(inp);
return(error);
}
}
if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
refcount_release(&blk->tfb_refcnt);
INP_WUNLOCK(inp);
return (ENOENT);
}
/*
* Ensure the new stack takes ownership with a
* clean slate on peak rate threshold.
*/
#ifdef TCPHPTS
/* Assure that we are not on any hpts */
tcp_hpts_remove(tp);
#endif
if (blk->tfb_tcp_fb_init) {
error = (*blk->tfb_tcp_fb_init)(tp, &ptr);
if (error) {
/*
* Release the ref count the lookup
* acquired.
*/
refcount_release(&blk->tfb_refcnt);
/*
* Now there is a chance that the
* init() function mucked with some
* things before it failed, such as
* hpts or inp_flags2 or timer granularity.
* It should not of, but lets give the old
* stack a chance to reset to a known good state.
*/
if (tp->t_fb->tfb_switch_failed) {
(*tp->t_fb->tfb_switch_failed)(tp);
}
goto err_out;
}
}
if (tp->t_fb->tfb_tcp_fb_fini) {
struct epoch_tracker et;
/*
* Tell the stack to cleanup with 0 i.e.
* the tcb is not going away.
*/
NET_EPOCH_ENTER(et);
(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
NET_EPOCH_EXIT(et);
}
/*
* Release the old refcnt, the
* lookup acquired a ref on the
* new one already.
*/
refcount_release(&tp->t_fb->tfb_refcnt);
/*
* Set in the new stack.
*/
tp->t_fb = blk;
tp->t_fb_ptr = ptr;
#ifdef TCP_OFFLOAD
if (tp->t_flags & TF_TOE) {
tcp_offload_ctloutput(tp, sopt->sopt_dir,
sopt->sopt_name);
}
#endif
err_out:
INP_WUNLOCK(inp);
return (error);
}
/* Pass in the INP locked, callee must unlock it. */
return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt));
}
static int
tcp_ctloutput_get(struct inpcb *inp, struct sockopt *sopt)
{
struct socket *so = inp->inp_socket;
struct tcpcb *tp = intotcpcb(inp);
int error = 0;
MPASS(sopt->sopt_dir == SOPT_GET);
INP_WLOCK_ASSERT(inp);
KASSERT((inp->inp_flags & INP_DROPPED) == 0,
("inp_flags == %x", inp->inp_flags));
KASSERT(so != NULL, ("inp_socket == NULL"));
if (sopt->sopt_level != IPPROTO_TCP) {
INP_WUNLOCK(inp);
#ifdef INET6
if (inp->inp_vflag & INP_IPV6PROTO)
error = ip6_ctloutput(so, sopt);
#endif /* INET6 */
#if defined(INET6) && defined(INET)
else
#endif
#ifdef INET
error = ip_ctloutput(so, sopt);
#endif
return (error);
}
if (((sopt->sopt_name == TCP_FUNCTION_BLK) ||
(sopt->sopt_name == TCP_FUNCTION_ALIAS))) {
struct tcp_function_set fsn;
if (sopt->sopt_name == TCP_FUNCTION_ALIAS) {
memset(&fsn, 0, sizeof(fsn));
find_tcp_function_alias(tp->t_fb, &fsn);
} else {
strncpy(fsn.function_set_name,
tp->t_fb->tfb_tcp_block_name,
TCP_FUNCTION_NAME_LEN_MAX);
fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
}
fsn.pcbcnt = tp->t_fb->tfb_refcnt;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &fsn, sizeof fsn);
return (error);
}
/* Pass in the INP locked, callee must unlock it. */
return (tp->t_fb->tfb_tcp_ctloutput(tp, sopt));
}
int
tcp_ctloutput(struct socket *so, struct sockopt *sopt)
{
struct inpcb *inp;
inp = sotoinpcb(so);
KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
return (ECONNRESET);
}
if (sopt->sopt_dir == SOPT_SET)
return (tcp_ctloutput_set(inp, sopt));
else if (sopt->sopt_dir == SOPT_GET)
return (tcp_ctloutput_get(inp, sopt));
else
panic("%s: sopt_dir $%d", __func__, sopt->sopt_dir);
}
/*
* If this assert becomes untrue, we need to change the size of the buf
* variable in tcp_default_ctloutput().
*/
#ifdef CTASSERT
CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
#endif
#ifdef KERN_TLS
static int
copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
{
struct tls_enable_v0 tls_v0;
int error;
if (sopt->sopt_valsize == sizeof(tls_v0)) {
error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
sizeof(tls_v0));
if (error)
return (error);
memset(tls, 0, sizeof(*tls));
tls->cipher_key = tls_v0.cipher_key;
tls->iv = tls_v0.iv;
tls->auth_key = tls_v0.auth_key;
tls->cipher_algorithm = tls_v0.cipher_algorithm;
tls->cipher_key_len = tls_v0.cipher_key_len;
tls->iv_len = tls_v0.iv_len;
tls->auth_algorithm = tls_v0.auth_algorithm;
tls->auth_key_len = tls_v0.auth_key_len;
tls->flags = tls_v0.flags;
tls->tls_vmajor = tls_v0.tls_vmajor;
tls->tls_vminor = tls_v0.tls_vminor;
return (0);
}
return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
}
#endif
extern struct cc_algo newreno_cc_algo;
static int
tcp_set_cc_mod(struct inpcb *inp, struct sockopt *sopt)
{
struct cc_algo *algo;
void *ptr = NULL;
struct tcpcb *tp;
struct cc_var cc_mem;
char buf[TCP_CA_NAME_MAX];
size_t mem_sz;
int error;
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
if (error)
return(error);
buf[sopt->sopt_valsize] = '\0';
CC_LIST_RLOCK();
STAILQ_FOREACH(algo, &cc_list, entries) {
if (strncmp(buf, algo->name,
TCP_CA_NAME_MAX) == 0) {
if (algo->flags & CC_MODULE_BEING_REMOVED) {
/* We can't "see" modules being unloaded */
continue;
}
break;
}
}
if (algo == NULL) {
CC_LIST_RUNLOCK();
return(ESRCH);
}
/*
* With a reference the algorithm cannot be removed
* so we hold a reference through the change process.
*/
cc_refer(algo);
CC_LIST_RUNLOCK();
if (algo->cb_init != NULL) {
/* We can now pre-get the memory for the CC */
mem_sz = (*algo->cc_data_sz)();
if (mem_sz == 0) {
goto no_mem_needed;
}
ptr = malloc(mem_sz, M_CC_MEM, M_WAITOK);
} else {
no_mem_needed:
mem_sz = 0;
ptr = NULL;
}
/*
* Make sure its all clean and zero and also get
* back the inplock.
*/
memset(&cc_mem, 0, sizeof(cc_mem));
INP_WLOCK(inp);
if (inp->inp_flags & INP_DROPPED) {
INP_WUNLOCK(inp);
if (ptr)
free(ptr, M_CC_MEM);
/* Release our temp reference */
CC_LIST_RLOCK();
cc_release(algo);
CC_LIST_RUNLOCK();
return (ECONNRESET);
}
tp = intotcpcb(inp);
if (ptr != NULL)
memset(ptr, 0, mem_sz);
cc_mem.ccvc.tcp = tp;
/*
* We once again hold a write lock over the tcb so it's
* safe to do these things without ordering concerns.
* Note here we init into stack memory.
*/
if (algo->cb_init != NULL)
error = algo->cb_init(&cc_mem, ptr);
else
error = 0;
/*
* The CC algorithms, when given their memory
* should not fail we could in theory have a
* KASSERT here.
*/
if (error == 0) {
/*
* Touchdown, lets go ahead and move the
* connection to the new CC module by
* copying in the cc_mem after we call
* the old ones cleanup (if any).
*/
if (CC_ALGO(tp)->cb_destroy != NULL)
CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
/* Detach the old CC from the tcpcb */
cc_detach(tp);
/* Copy in our temp memory that was inited */
memcpy(&tp->t_ccv, &cc_mem, sizeof(struct cc_var));
/* Now attach the new, which takes a reference */
cc_attach(tp, algo);
/* Ok now are we where we have gotten past any conn_init? */
if (TCPS_HAVEESTABLISHED(tp->t_state) && (CC_ALGO(tp)->conn_init != NULL)) {
/* Yep run the connection init for the new CC */
CC_ALGO(tp)->conn_init(&tp->t_ccv);
}
} else if (ptr)
free(ptr, M_CC_MEM);
INP_WUNLOCK(inp);
/* Now lets release our temp reference */
CC_LIST_RLOCK();
cc_release(algo);
CC_LIST_RUNLOCK();
return (error);
}
int
tcp_default_ctloutput(struct tcpcb *tp, struct sockopt *sopt)
{
struct inpcb *inp = tptoinpcb(tp);
int error, opt, optval;
u_int ui;
struct tcp_info ti;
#ifdef KERN_TLS
struct tls_enable tls;
struct socket *so = inp->inp_socket;
#endif
char *pbuf, buf[TCP_LOG_ID_LEN];
#ifdef STATS
struct statsblob *sbp;
#endif
size_t len;
INP_WLOCK_ASSERT(inp);
KASSERT((inp->inp_flags & INP_DROPPED) == 0,
("inp_flags == %x", inp->inp_flags));
KASSERT(inp->inp_socket != NULL, ("inp_socket == NULL"));
switch (sopt->sopt_level) {
#ifdef INET6
case IPPROTO_IPV6:
MPASS(inp->inp_vflag & INP_IPV6PROTO);
switch (sopt->sopt_name) {
case IPV6_USE_MIN_MTU:
tcp6_use_min_mtu(tp);
/* FALLTHROUGH */
}
INP_WUNLOCK(inp);
return (0);
#endif
#ifdef INET
case IPPROTO_IP:
INP_WUNLOCK(inp);
return (0);
#endif
}
/*
* For TCP_CCALGOOPT forward the control to CC module, for both
* SOPT_SET and SOPT_GET.
*/
switch (sopt->sopt_name) {
case TCP_CCALGOOPT:
INP_WUNLOCK(inp);
if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
return (EINVAL);
pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
sopt->sopt_valsize);
if (error) {
free(pbuf, M_TEMP);
return (error);
}
INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
if (CC_ALGO(tp)->ctl_output != NULL)
error = CC_ALGO(tp)->ctl_output(&tp->t_ccv, sopt, pbuf);
else
error = ENOENT;
INP_WUNLOCK(inp);
if (error == 0 && sopt->sopt_dir == SOPT_GET)
error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
free(pbuf, M_TEMP);
return (error);
}
switch (sopt->sopt_dir) {
case SOPT_SET:
switch (sopt->sopt_name) {
#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
case TCP_MD5SIG:
INP_WUNLOCK(inp);
if (!TCPMD5_ENABLED())
return (ENOPROTOOPT);
error = TCPMD5_PCBCTL(inp, sopt);
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
goto unlock_and_done;
#endif /* IPSEC */
case TCP_NODELAY:
case TCP_NOOPT:
case TCP_LRD:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
switch (sopt->sopt_name) {
case TCP_NODELAY:
opt = TF_NODELAY;
break;
case TCP_NOOPT:
opt = TF_NOOPT;
break;
case TCP_LRD:
opt = TF_LRD;
break;
default:
opt = 0; /* dead code to fool gcc */
break;
}
if (optval)
tp->t_flags |= opt;
else
tp->t_flags &= ~opt;
unlock_and_done:
#ifdef TCP_OFFLOAD
if (tp->t_flags & TF_TOE) {
tcp_offload_ctloutput(tp, sopt->sopt_dir,
sopt->sopt_name);
}
#endif
INP_WUNLOCK(inp);
break;
case TCP_NOPUSH:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
if (optval)
tp->t_flags |= TF_NOPUSH;
else if (tp->t_flags & TF_NOPUSH) {
tp->t_flags &= ~TF_NOPUSH;
if (TCPS_HAVEESTABLISHED(tp->t_state)) {
struct epoch_tracker et;
NET_EPOCH_ENTER(et);
error = tcp_output_nodrop(tp);
NET_EPOCH_EXIT(et);
}
}
goto unlock_and_done;
case TCP_REMOTE_UDP_ENCAPS_PORT:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
if ((optval < TCP_TUNNELING_PORT_MIN) ||
(optval > TCP_TUNNELING_PORT_MAX)) {
/* Its got to be in range */
return (EINVAL);
}
if ((V_tcp_udp_tunneling_port == 0) && (optval != 0)) {
/* You have to have enabled a UDP tunneling port first */
return (EINVAL);
}
INP_WLOCK_RECHECK(inp);
if (tp->t_state != TCPS_CLOSED) {
/* You can't change after you are connected */
error = EINVAL;
} else {
/* Ok we are all good set the port */
tp->t_port = htons(optval);
}
goto unlock_and_done;
case TCP_MAXSEG:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
if (optval > 0 && optval <= tp->t_maxseg &&
optval + 40 >= V_tcp_minmss)
tp->t_maxseg = optval;
else
error = EINVAL;
goto unlock_and_done;
case TCP_INFO:
INP_WUNLOCK(inp);
error = EINVAL;
break;
case TCP_STATS:
INP_WUNLOCK(inp);
#ifdef STATS
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
if (optval > 0)
sbp = stats_blob_alloc(
V_tcp_perconn_stats_dflt_tpl, 0);
else
sbp = NULL;
INP_WLOCK_RECHECK(inp);
if ((tp->t_stats != NULL && sbp == NULL) ||
(tp->t_stats == NULL && sbp != NULL)) {
struct statsblob *t = tp->t_stats;
tp->t_stats = sbp;
sbp = t;
}
INP_WUNLOCK(inp);
stats_blob_destroy(sbp);
#else
return (EOPNOTSUPP);
#endif /* !STATS */
break;
case TCP_CONGESTION:
error = tcp_set_cc_mod(inp, sopt);
break;
case TCP_REUSPORT_LB_NUMA:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof(optval),
sizeof(optval));
INP_WLOCK_RECHECK(inp);
if (!error)
error = in_pcblbgroup_numa(inp, optval);
INP_WUNLOCK(inp);
break;
#ifdef KERN_TLS
case TCP_TXTLS_ENABLE:
INP_WUNLOCK(inp);
error = copyin_tls_enable(sopt, &tls);
if (error)
break;
error = ktls_enable_tx(so, &tls);
break;
case TCP_TXTLS_MODE:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
error = ktls_set_tx_mode(so, ui);
INP_WUNLOCK(inp);
break;
case TCP_RXTLS_ENABLE:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &tls, sizeof(tls),
sizeof(tls));
if (error)
break;
error = ktls_enable_rx(so, &tls);
break;
#endif
case TCP_MAXUNACKTIME:
case TCP_KEEPIDLE:
case TCP_KEEPINTVL:
case TCP_KEEPINIT:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
if (error)
return (error);
if (ui > (UINT_MAX / hz)) {
error = EINVAL;
break;
}
ui *= hz;
INP_WLOCK_RECHECK(inp);
switch (sopt->sopt_name) {
case TCP_MAXUNACKTIME:
tp->t_maxunacktime = ui;
break;
case TCP_KEEPIDLE:
tp->t_keepidle = ui;
/*
* XXX: better check current remaining
* timeout and "merge" it with new value.
*/
if ((tp->t_state > TCPS_LISTEN) &&
(tp->t_state <= TCPS_CLOSING))
tcp_timer_activate(tp, TT_KEEP,
TP_KEEPIDLE(tp));
break;
case TCP_KEEPINTVL:
tp->t_keepintvl = ui;
if ((tp->t_state == TCPS_FIN_WAIT_2) &&
(TP_MAXIDLE(tp) > 0))
tcp_timer_activate(tp, TT_2MSL,
TP_MAXIDLE(tp));
break;
case TCP_KEEPINIT:
tp->t_keepinit = ui;
if (tp->t_state == TCPS_SYN_RECEIVED ||
tp->t_state == TCPS_SYN_SENT)
tcp_timer_activate(tp, TT_KEEP,
TP_KEEPINIT(tp));
break;
}
goto unlock_and_done;
case TCP_KEEPCNT:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
tp->t_keepcnt = ui;
if ((tp->t_state == TCPS_FIN_WAIT_2) &&
(TP_MAXIDLE(tp) > 0))
tcp_timer_activate(tp, TT_2MSL,
TP_MAXIDLE(tp));
goto unlock_and_done;
#ifdef TCPPCAP
case TCP_PCAP_OUT:
case TCP_PCAP_IN:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
if (optval >= 0)
tcp_pcap_set_sock_max(
(sopt->sopt_name == TCP_PCAP_OUT) ?
&(tp->t_outpkts) : &(tp->t_inpkts),
optval);
else
error = EINVAL;
goto unlock_and_done;
#endif
case TCP_FASTOPEN: {
struct tcp_fastopen tfo_optval;
INP_WUNLOCK(inp);
if (!V_tcp_fastopen_client_enable &&
!V_tcp_fastopen_server_enable)
return (EPERM);
error = sooptcopyin(sopt, &tfo_optval,
sizeof(tfo_optval), sizeof(int));
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
if ((tp->t_state != TCPS_CLOSED) &&
(tp->t_state != TCPS_LISTEN)) {
error = EINVAL;
goto unlock_and_done;
}
if (tfo_optval.enable) {
if (tp->t_state == TCPS_LISTEN) {
if (!V_tcp_fastopen_server_enable) {
error = EPERM;
goto unlock_and_done;
}
if (tp->t_tfo_pending == NULL)
tp->t_tfo_pending =
tcp_fastopen_alloc_counter();
} else {
/*
* If a pre-shared key was provided,
* stash it in the client cookie
* field of the tcpcb for use during
* connect.
*/
if (sopt->sopt_valsize ==
sizeof(tfo_optval)) {
memcpy(tp->t_tfo_cookie.client,
tfo_optval.psk,
TCP_FASTOPEN_PSK_LEN);
tp->t_tfo_client_cookie_len =
TCP_FASTOPEN_PSK_LEN;
}
}
tp->t_flags |= TF_FASTOPEN;
} else
tp->t_flags &= ~TF_FASTOPEN;
goto unlock_and_done;
}
#ifdef TCP_BLACKBOX
case TCP_LOG:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, &optval, sizeof optval,
sizeof optval);
if (error)
return (error);
INP_WLOCK_RECHECK(inp);
error = tcp_log_state_change(tp, optval);
goto unlock_and_done;
case TCP_LOGBUF:
INP_WUNLOCK(inp);
error = EINVAL;
break;
case TCP_LOGID:
INP_WUNLOCK(inp);
error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
if (error)
break;
buf[sopt->sopt_valsize] = '\0';
INP_WLOCK_RECHECK(inp);
error = tcp_log_set_id(tp, buf);
/* tcp_log_set_id() unlocks the INP. */
break;
case TCP_LOGDUMP:
case TCP_LOGDUMPID:
INP_WUNLOCK(inp);
error =
sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
if (error)
break;
buf[sopt->sopt_valsize] = '\0';
INP_WLOCK_RECHECK(inp);
if (sopt->sopt_name == TCP_LOGDUMP) {
error = tcp_log_dump_tp_logbuf(tp, buf,
M_WAITOK, true);
INP_WUNLOCK(inp);
} else {
tcp_log_dump_tp_bucket_logbufs(tp, buf);
/*
* tcp_log_dump_tp_bucket_logbufs() drops the
* INP lock.
*/
}
break;
#endif
default:
INP_WUNLOCK(inp);
error = ENOPROTOOPT;
break;
}
break;
case SOPT_GET:
tp = intotcpcb(inp);
switch (sopt->sopt_name) {
#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
case TCP_MD5SIG:
INP_WUNLOCK(inp);
if (!TCPMD5_ENABLED())
return (ENOPROTOOPT);
error = TCPMD5_PCBCTL(inp, sopt);
break;
#endif
case TCP_NODELAY:
optval = tp->t_flags & TF_NODELAY;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
case TCP_MAXSEG:
optval = tp->t_maxseg;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
case TCP_REMOTE_UDP_ENCAPS_PORT:
optval = ntohs(tp->t_port);
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
case TCP_NOOPT:
optval = tp->t_flags & TF_NOOPT;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
case TCP_NOPUSH:
optval = tp->t_flags & TF_NOPUSH;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
case TCP_INFO:
tcp_fill_info(tp, &ti);
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &ti, sizeof ti);
break;
case TCP_STATS:
{
#ifdef STATS
int nheld;
TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
error = 0;
socklen_t outsbsz = sopt->sopt_valsize;
if (tp->t_stats == NULL)
error = ENOENT;
else if (outsbsz >= tp->t_stats->cursz)
outsbsz = tp->t_stats->cursz;
else if (outsbsz >= sizeof(struct statsblob))
outsbsz = sizeof(struct statsblob);
else
error = EINVAL;
INP_WUNLOCK(inp);
if (error)
break;
sbp = sopt->sopt_val;
nheld = atop(round_page(((vm_offset_t)sbp) +
(vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
vm_page_t ma[nheld];
if (vm_fault_quick_hold_pages(
&curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
nheld) < 0) {
error = EFAULT;
break;
}
if ((error = copyin_nofault(&(sbp->flags), &sbflags,
SIZEOF_MEMBER(struct statsblob, flags))))
goto unhold;
INP_WLOCK_RECHECK(inp);
error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
sbflags | SB_CLONE_USRDSTNOFAULT);
INP_WUNLOCK(inp);
sopt->sopt_valsize = outsbsz;
unhold:
vm_page_unhold_pages(ma, nheld);
#else
INP_WUNLOCK(inp);
error = EOPNOTSUPP;
#endif /* !STATS */
break;
}
case TCP_CONGESTION:
len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, buf, len + 1);
break;
case TCP_MAXUNACKTIME:
case TCP_KEEPIDLE:
case TCP_KEEPINTVL:
case TCP_KEEPINIT:
case TCP_KEEPCNT:
switch (sopt->sopt_name) {
case TCP_MAXUNACKTIME:
ui = TP_MAXUNACKTIME(tp) / hz;
break;
case TCP_KEEPIDLE:
ui = TP_KEEPIDLE(tp) / hz;
break;
case TCP_KEEPINTVL:
ui = TP_KEEPINTVL(tp) / hz;
break;
case TCP_KEEPINIT:
ui = TP_KEEPINIT(tp) / hz;
break;
case TCP_KEEPCNT:
ui = TP_KEEPCNT(tp);
break;
}
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &ui, sizeof(ui));
break;
#ifdef TCPPCAP
case TCP_PCAP_OUT:
case TCP_PCAP_IN:
optval = tcp_pcap_get_sock_max(
(sopt->sopt_name == TCP_PCAP_OUT) ?
&(tp->t_outpkts) : &(tp->t_inpkts));
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
#endif
case TCP_FASTOPEN:
optval = tp->t_flags & TF_FASTOPEN;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
#ifdef TCP_BLACKBOX
case TCP_LOG:
optval = tcp_get_bblog_state(tp);
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof(optval));
break;
case TCP_LOGBUF:
/* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
error = tcp_log_getlogbuf(sopt, tp);
break;
case TCP_LOGID:
len = tcp_log_get_id(tp, buf);
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, buf, len + 1);
break;
case TCP_LOGDUMP:
case TCP_LOGDUMPID:
INP_WUNLOCK(inp);
error = EINVAL;
break;
#endif
#ifdef KERN_TLS
case TCP_TXTLS_MODE:
error = ktls_get_tx_mode(so, &optval);
INP_WUNLOCK(inp);
if (error == 0)
error = sooptcopyout(sopt, &optval,
sizeof(optval));
break;
case TCP_RXTLS_MODE:
error = ktls_get_rx_mode(so, &optval);
INP_WUNLOCK(inp);
if (error == 0)
error = sooptcopyout(sopt, &optval,
sizeof(optval));
break;
#endif
case TCP_LRD:
optval = tp->t_flags & TF_LRD;
INP_WUNLOCK(inp);
error = sooptcopyout(sopt, &optval, sizeof optval);
break;
default:
INP_WUNLOCK(inp);
error = ENOPROTOOPT;
break;
}
break;
}
return (error);
}
#undef INP_WLOCK_RECHECK
#undef INP_WLOCK_RECHECK_CLEANUP
/*
* Initiate (or continue) disconnect.
* If embryonic state, just send reset (once).
* If in ``let data drain'' option and linger null, just drop.
* Otherwise (hard), mark socket disconnecting and drop
* current input data; switch states based on user close, and
* send segment to peer (with FIN).
*/
static void
tcp_disconnect(struct tcpcb *tp)
{
struct inpcb *inp = tptoinpcb(tp);
struct socket *so = tptosocket(tp);
NET_EPOCH_ASSERT();
INP_WLOCK_ASSERT(inp);
/*
* Neither tcp_close() nor tcp_drop() should return NULL, as the
* socket is still open.
*/
if (tp->t_state < TCPS_ESTABLISHED &&
!(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
tp = tcp_close(tp);
KASSERT(tp != NULL,
("tcp_disconnect: tcp_close() returned NULL"));
} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
tp = tcp_drop(tp, 0);
KASSERT(tp != NULL,
("tcp_disconnect: tcp_drop() returned NULL"));
} else {
soisdisconnecting(so);
sbflush(&so->so_rcv);
tcp_usrclosed(tp);
if (!(inp->inp_flags & INP_DROPPED))
/* Ignore stack's drop request, we already at it. */
(void)tcp_output_nodrop(tp);
}
}
/*
* User issued close, and wish to trail through shutdown states:
* if never received SYN, just forget it. If got a SYN from peer,
* but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
* If already got a FIN from peer, then almost done; go to LAST_ACK
* state. In all other cases, have already sent FIN to peer (e.g.
* after PRU_SHUTDOWN), and just have to play tedious game waiting
* for peer to send FIN or not respond to keep-alives, etc.
* We can let the user exit from the close as soon as the FIN is acked.
*/
static void
tcp_usrclosed(struct tcpcb *tp)
{
NET_EPOCH_ASSERT();
INP_WLOCK_ASSERT(tptoinpcb(tp));
switch (tp->t_state) {
case TCPS_LISTEN:
#ifdef TCP_OFFLOAD
tcp_offload_listen_stop(tp);
#endif
tcp_state_change(tp, TCPS_CLOSED);
/* FALLTHROUGH */
case TCPS_CLOSED:
tp = tcp_close(tp);
/*
* tcp_close() should never return NULL here as the socket is
* still open.
*/
KASSERT(tp != NULL,
("tcp_usrclosed: tcp_close() returned NULL"));
break;
case TCPS_SYN_SENT:
case TCPS_SYN_RECEIVED:
tp->t_flags |= TF_NEEDFIN;
break;
case TCPS_ESTABLISHED:
tcp_state_change(tp, TCPS_FIN_WAIT_1);
break;
case TCPS_CLOSE_WAIT:
tcp_state_change(tp, TCPS_LAST_ACK);
break;
}
if (tp->t_acktime == 0)
tp->t_acktime = ticks;
if (tp->t_state >= TCPS_FIN_WAIT_2) {
soisdisconnected(tptosocket(tp));
/* Prevent the connection hanging in FIN_WAIT_2 forever. */
if (tp->t_state == TCPS_FIN_WAIT_2) {
int timeout;
timeout = (tcp_fast_finwait2_recycle) ?
tcp_finwait2_timeout : TP_MAXIDLE(tp);
tcp_timer_activate(tp, TT_2MSL, timeout);
}
}
}
#ifdef DDB
static void
db_print_indent(int indent)
{
int i;
for (i = 0; i < indent; i++)
db_printf(" ");
}
static void
db_print_tstate(int t_state)
{
switch (t_state) {
case TCPS_CLOSED:
db_printf("TCPS_CLOSED");
return;
case TCPS_LISTEN:
db_printf("TCPS_LISTEN");
return;
case TCPS_SYN_SENT:
db_printf("TCPS_SYN_SENT");
return;
case TCPS_SYN_RECEIVED:
db_printf("TCPS_SYN_RECEIVED");
return;
case TCPS_ESTABLISHED:
db_printf("TCPS_ESTABLISHED");
return;
case TCPS_CLOSE_WAIT:
db_printf("TCPS_CLOSE_WAIT");
return;
case TCPS_FIN_WAIT_1:
db_printf("TCPS_FIN_WAIT_1");
return;
case TCPS_CLOSING:
db_printf("TCPS_CLOSING");
return;
case TCPS_LAST_ACK:
db_printf("TCPS_LAST_ACK");
return;
case TCPS_FIN_WAIT_2:
db_printf("TCPS_FIN_WAIT_2");
return;
case TCPS_TIME_WAIT:
db_printf("TCPS_TIME_WAIT");
return;
default:
db_printf("unknown");
return;
}
}
static void
db_print_tflags(u_int t_flags)
{
int comma;
comma = 0;
if (t_flags & TF_ACKNOW) {
db_printf("%sTF_ACKNOW", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_DELACK) {
db_printf("%sTF_DELACK", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_NODELAY) {
db_printf("%sTF_NODELAY", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_NOOPT) {
db_printf("%sTF_NOOPT", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_SENTFIN) {
db_printf("%sTF_SENTFIN", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_REQ_SCALE) {
db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_RCVD_SCALE) {
db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_REQ_TSTMP) {
db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_RCVD_TSTMP) {
db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_SACK_PERMIT) {
db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_NEEDSYN) {
db_printf("%sTF_NEEDSYN", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_NEEDFIN) {
db_printf("%sTF_NEEDFIN", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_NOPUSH) {
db_printf("%sTF_NOPUSH", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_PREVVALID) {
db_printf("%sTF_PREVVALID", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_MORETOCOME) {
db_printf("%sTF_MORETOCOME", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_SONOTCONN) {
db_printf("%sTF_SONOTCONN", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_LASTIDLE) {
db_printf("%sTF_LASTIDLE", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_RXWIN0SENT) {
db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_FASTRECOVERY) {
db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_CONGRECOVERY) {
db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_WASFRECOVERY) {
db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_WASCRECOVERY) {
db_printf("%sTF_WASCRECOVERY", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_SIGNATURE) {
db_printf("%sTF_SIGNATURE", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_FORCEDATA) {
db_printf("%sTF_FORCEDATA", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_TSO) {
db_printf("%sTF_TSO", comma ? ", " : "");
comma = 1;
}
if (t_flags & TF_FASTOPEN) {
db_printf("%sTF_FASTOPEN", comma ? ", " : "");
comma = 1;
}
}
static void
db_print_tflags2(u_int t_flags2)
{
int comma;
comma = 0;
if (t_flags2 & TF2_PLPMTU_BLACKHOLE) {
db_printf("%sTF2_PLPMTU_BLACKHOLE", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_PLPMTU_PMTUD) {
db_printf("%sTF2_PLPMTU_PMTUD", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_PLPMTU_MAXSEGSNT) {
db_printf("%sTF2_PLPMTU_MAXSEGSNT", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_LOG_AUTO) {
db_printf("%sTF2_LOG_AUTO", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_DROP_AF_DATA) {
db_printf("%sTF2_DROP_AF_DATA", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_ECN_PERMIT) {
db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_ECN_SND_CWR) {
db_printf("%sTF2_ECN_SND_CWR", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_ECN_SND_ECE) {
db_printf("%sTF2_ECN_SND_ECE", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_ACE_PERMIT) {
db_printf("%sTF2_ACE_PERMIT", comma ? ", " : "");
comma = 1;
}
if (t_flags2 & TF2_FBYTES_COMPLETE) {
db_printf("%sTF2_FBYTES_COMPLETE", comma ? ", " : "");
comma = 1;
}
}
static void
db_print_toobflags(char t_oobflags)
{
int comma;
comma = 0;
if (t_oobflags & TCPOOB_HAVEDATA) {
db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
comma = 1;
}
if (t_oobflags & TCPOOB_HADDATA) {
db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
comma = 1;
}
}
static void
db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
{
db_print_indent(indent);
db_printf("%s at %p\n", name, tp);
indent += 2;
db_print_indent(indent);
db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
db_print_indent(indent);
db_printf("t_callout: %p t_timers: %p\n",
&tp->t_callout, &tp->t_timers);
db_print_indent(indent);
db_printf("t_state: %d (", tp->t_state);
db_print_tstate(tp->t_state);
db_printf(")\n");
db_print_indent(indent);
db_printf("t_flags: 0x%x (", tp->t_flags);
db_print_tflags(tp->t_flags);
db_printf(")\n");
db_print_indent(indent);
db_printf("t_flags2: 0x%x (", tp->t_flags2);
db_print_tflags2(tp->t_flags2);
db_printf(")\n");
db_print_indent(indent);
db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: 0x%08x\n",
tp->snd_una, tp->snd_max, tp->snd_nxt);
db_print_indent(indent);
db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
tp->snd_up, tp->snd_wl1, tp->snd_wl2);
db_print_indent(indent);
db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
tp->iss, tp->irs, tp->rcv_nxt);
db_print_indent(indent);
db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n",
tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
db_print_indent(indent);
db_printf("snd_wnd: %u snd_cwnd: %u\n",
tp->snd_wnd, tp->snd_cwnd);
db_print_indent(indent);
db_printf("snd_ssthresh: %u snd_recover: "
"0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
db_print_indent(indent);
db_printf("t_rcvtime: %u t_startime: %u\n",
tp->t_rcvtime, tp->t_starttime);
db_print_indent(indent);
db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
tp->t_rtttime, tp->t_rtseq);
db_print_indent(indent);
db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
db_print_indent(indent);
db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u\n",
tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin);
db_print_indent(indent);
db_printf("t_rttupdated: %u max_sndwnd: %u t_softerror: %d\n",
tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
db_print_indent(indent);
db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
db_print_toobflags(tp->t_oobflags);
db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
db_print_indent(indent);
db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
db_print_indent(indent);
db_printf("ts_recent: %u ts_recent_age: %u\n",
tp->ts_recent, tp->ts_recent_age);
db_print_indent(indent);
db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
"%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
db_print_indent(indent);
db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x "
"t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
tp->snd_recover_prev, tp->t_badrxtwin);
db_print_indent(indent);
db_printf("snd_numholes: %d snd_holes first: %p\n",
tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
db_print_indent(indent);
db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n",
tp->snd_fack, tp->rcv_numsacks);
/* Skip sackblks, sackhint. */
db_print_indent(indent);
db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
}
DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
{
struct tcpcb *tp;
if (!have_addr) {
db_printf("usage: show tcpcb <addr>\n");
return;
}
tp = (struct tcpcb *)addr;
db_print_tcpcb(tp, "tcpcb", 0);
}
#endif
diff --git a/sys/netinet/toecore.c b/sys/netinet/toecore.c
index 5f5863431dfe..76aadad9a3b9 100644
--- a/sys/netinet/toecore.c
+++ b/sys/netinet/toecore.c
@@ -1,604 +1,604 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2012 Chelsio Communications, Inc.
* All rights reserved.
* Written by: Navdeep Parhar <np@FreeBSD.org>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_inet6.h"
#include <sys/param.h>
#include <sys/eventhandler.h>
#include <sys/kernel.h>
#include <sys/systm.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/types.h>
#include <sys/sockopt.h>
#include <sys/sysctl.h>
#include <sys/socket.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_private.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <net/if_llatbl.h>
#include <net/route.h>
#include <netinet/if_ether.h>
#include <netinet/in.h>
#include <netinet/in_pcb.h>
#include <netinet/in_var.h>
#include <netinet6/in6_var.h>
#include <netinet6/in6_pcb.h>
#include <netinet6/nd6.h>
#define TCPSTATES
#include <netinet/tcp.h>
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_timer.h>
#include <netinet/tcp_var.h>
#include <netinet/tcp_syncache.h>
#include <netinet/tcp_offload.h>
#include <netinet/toecore.h>
static struct mtx toedev_lock;
static TAILQ_HEAD(, toedev) toedev_list;
static eventhandler_tag listen_start_eh;
static eventhandler_tag listen_stop_eh;
static eventhandler_tag lle_event_eh;
static int
toedev_connect(struct toedev *tod __unused, struct socket *so __unused,
struct nhop_object *nh __unused, struct sockaddr *nam __unused)
{
return (ENOTSUP);
}
static int
toedev_listen_start(struct toedev *tod __unused, struct tcpcb *tp __unused)
{
return (ENOTSUP);
}
static int
toedev_listen_stop(struct toedev *tod __unused, struct tcpcb *tp __unused)
{
return (ENOTSUP);
}
static void
toedev_input(struct toedev *tod __unused, struct tcpcb *tp __unused,
struct mbuf *m)
{
m_freem(m);
return;
}
static void
toedev_rcvd(struct toedev *tod __unused, struct tcpcb *tp __unused)
{
return;
}
static int
toedev_output(struct toedev *tod __unused, struct tcpcb *tp __unused)
{
return (ENOTSUP);
}
static void
toedev_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp __unused)
{
return;
}
static void
toedev_l2_update(struct toedev *tod __unused, struct ifnet *ifp __unused,
struct sockaddr *sa __unused, uint8_t *lladdr __unused,
uint16_t vtag __unused)
{
return;
}
static void
toedev_route_redirect(struct toedev *tod __unused, struct ifnet *ifp __unused,
struct nhop_object *nh0 __unused, struct nhop_object *nh1 __unused)
{
return;
}
static void
toedev_syncache_added(struct toedev *tod __unused, void *ctx __unused)
{
return;
}
static void
toedev_syncache_removed(struct toedev *tod __unused, void *ctx __unused)
{
return;
}
static int
toedev_syncache_respond(struct toedev *tod __unused, void *ctx __unused,
struct mbuf *m)
{
m_freem(m);
return (0);
}
static void
toedev_offload_socket(struct toedev *tod __unused, void *ctx __unused,
struct socket *so __unused)
{
return;
}
static void
toedev_ctloutput(struct toedev *tod __unused, struct tcpcb *tp __unused,
int sopt_dir __unused, int sopt_name __unused)
{
return;
}
static void
-toedev_tcp_info(struct toedev *tod __unused, struct tcpcb *tp __unused,
+toedev_tcp_info(struct toedev *tod __unused, const struct tcpcb *tp __unused,
struct tcp_info *ti __unused)
{
return;
}
static int
toedev_alloc_tls_session(struct toedev *tod __unused, struct tcpcb *tp __unused,
struct ktls_session *tls __unused, int direction __unused)
{
return (EINVAL);
}
static void
toedev_pmtu_update(struct toedev *tod __unused, struct tcpcb *tp __unused,
tcp_seq seq __unused, int mtu __unused)
{
return;
}
/*
* Inform one or more TOE devices about a listening socket.
*/
static void
toe_listen_start(struct inpcb *inp, void *arg)
{
struct toedev *t, *tod;
struct tcpcb *tp;
INP_WLOCK_ASSERT(inp);
KASSERT(inp->inp_pcbinfo == &V_tcbinfo,
("%s: inp is not a TCP inp", __func__));
if (inp->inp_flags & INP_DROPPED)
return;
tp = intotcpcb(inp);
if (tp->t_state != TCPS_LISTEN)
return;
t = arg;
mtx_lock(&toedev_lock);
TAILQ_FOREACH(tod, &toedev_list, link) {
if (t == NULL || t == tod)
tod->tod_listen_start(tod, tp);
}
mtx_unlock(&toedev_lock);
}
static void
toe_listen_start_event(void *arg __unused, struct tcpcb *tp)
{
struct inpcb *inp = tptoinpcb(tp);
INP_WLOCK_ASSERT(inp);
KASSERT(tp->t_state == TCPS_LISTEN,
("%s: t_state %s", __func__, tcpstates[tp->t_state]));
toe_listen_start(inp, NULL);
}
static void
toe_listen_stop_event(void *arg __unused, struct tcpcb *tp)
{
struct toedev *tod;
#ifdef INVARIANTS
struct inpcb *inp = tptoinpcb(tp);
#endif
INP_WLOCK_ASSERT(inp);
KASSERT(tp->t_state == TCPS_LISTEN,
("%s: t_state %s", __func__, tcpstates[tp->t_state]));
mtx_lock(&toedev_lock);
TAILQ_FOREACH(tod, &toedev_list, link)
tod->tod_listen_stop(tod, tp);
mtx_unlock(&toedev_lock);
}
/*
* Fill up a freshly allocated toedev struct with reasonable defaults.
*/
void
init_toedev(struct toedev *tod)
{
tod->tod_softc = NULL;
/*
* Provide no-op defaults so that the kernel can call any toedev
* function without having to check whether the TOE driver supplied one
* or not.
*/
tod->tod_connect = toedev_connect;
tod->tod_listen_start = toedev_listen_start;
tod->tod_listen_stop = toedev_listen_stop;
tod->tod_input = toedev_input;
tod->tod_rcvd = toedev_rcvd;
tod->tod_output = toedev_output;
tod->tod_send_rst = toedev_output;
tod->tod_send_fin = toedev_output;
tod->tod_pcb_detach = toedev_pcb_detach;
tod->tod_l2_update = toedev_l2_update;
tod->tod_route_redirect = toedev_route_redirect;
tod->tod_syncache_added = toedev_syncache_added;
tod->tod_syncache_removed = toedev_syncache_removed;
tod->tod_syncache_respond = toedev_syncache_respond;
tod->tod_offload_socket = toedev_offload_socket;
tod->tod_ctloutput = toedev_ctloutput;
tod->tod_tcp_info = toedev_tcp_info;
tod->tod_alloc_tls_session = toedev_alloc_tls_session;
tod->tod_pmtu_update = toedev_pmtu_update;
}
/*
* Register an active TOE device with the system. This allows it to receive
* notifications from the kernel.
*/
int
register_toedev(struct toedev *tod)
{
struct toedev *t;
mtx_lock(&toedev_lock);
TAILQ_FOREACH(t, &toedev_list, link) {
if (t == tod) {
mtx_unlock(&toedev_lock);
return (EEXIST);
}
}
TAILQ_INSERT_TAIL(&toedev_list, tod, link);
registered_toedevs++;
mtx_unlock(&toedev_lock);
inp_apply_all(&V_tcbinfo, toe_listen_start, tod);
return (0);
}
/*
* Remove the TOE device from the global list of active TOE devices. It is the
* caller's responsibility to ensure that the TOE device is quiesced prior to
* this call.
*/
int
unregister_toedev(struct toedev *tod)
{
struct toedev *t, *t2;
int rc = ENODEV;
mtx_lock(&toedev_lock);
TAILQ_FOREACH_SAFE(t, &toedev_list, link, t2) {
if (t == tod) {
TAILQ_REMOVE(&toedev_list, tod, link);
registered_toedevs--;
rc = 0;
break;
}
}
KASSERT(registered_toedevs >= 0,
("%s: registered_toedevs (%d) < 0", __func__, registered_toedevs));
mtx_unlock(&toedev_lock);
return (rc);
}
void
toe_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th,
struct inpcb *inp, void *tod, void *todctx, uint8_t iptos)
{
INP_RLOCK_ASSERT(inp);
(void )syncache_add(inc, to, th, inp, inp->inp_socket, NULL, tod,
todctx, iptos, htons(0));
}
int
toe_syncache_expand(struct in_conninfo *inc, struct tcpopt *to,
struct tcphdr *th, struct socket **lsop)
{
NET_EPOCH_ASSERT();
return (syncache_expand(inc, to, th, lsop, NULL, htons(0)));
}
/*
* General purpose check to see if a 4-tuple is in use by the kernel. If a TCP
* header (presumably for an incoming SYN) is also provided, an existing 4-tuple
* in TIME_WAIT may be assassinated freeing it up for re-use.
*
* Note that the TCP header must have been run through tcp_fields_to_host() or
* equivalent.
*/
int
toe_4tuple_check(struct in_conninfo *inc, struct tcphdr *th, struct ifnet *ifp)
{
struct inpcb *inp;
struct tcpcb *tp;
if (inc->inc_flags & INC_ISIPV6) {
inp = in6_pcblookup(&V_tcbinfo, &inc->inc6_faddr,
inc->inc_fport, &inc->inc6_laddr, inc->inc_lport,
INPLOOKUP_RLOCKPCB, ifp);
} else {
inp = in_pcblookup(&V_tcbinfo, inc->inc_faddr, inc->inc_fport,
inc->inc_laddr, inc->inc_lport, INPLOOKUP_RLOCKPCB, ifp);
}
if (inp != NULL) {
INP_RLOCK_ASSERT(inp);
tp = intotcpcb(inp);
if (tp->t_state == TCPS_TIME_WAIT && th != NULL) {
if (!tcp_twcheck(inp, NULL, th, NULL, 0))
return (EADDRINUSE);
} else {
INP_RUNLOCK(inp);
return (EADDRINUSE);
}
}
return (0);
}
static void
toe_lle_event(void *arg __unused, struct llentry *lle, int evt)
{
struct toedev *tod;
struct ifnet *ifp;
struct sockaddr *sa;
uint8_t *lladdr;
uint16_t vid, pcp;
int family;
struct sockaddr_in6 sin6;
LLE_WLOCK_ASSERT(lle);
ifp = lltable_get_ifp(lle->lle_tbl);
family = lltable_get_af(lle->lle_tbl);
if (family != AF_INET && family != AF_INET6)
return;
/*
* Not interested if the interface's TOE capability is not enabled.
*/
if ((family == AF_INET && !(ifp->if_capenable & IFCAP_TOE4)) ||
(family == AF_INET6 && !(ifp->if_capenable & IFCAP_TOE6)))
return;
tod = TOEDEV(ifp);
if (tod == NULL)
return;
sa = (struct sockaddr *)&sin6;
lltable_fill_sa_entry(lle, sa);
vid = 0xfff;
pcp = 0;
if (evt != LLENTRY_RESOLVED) {
/*
* LLENTRY_TIMEDOUT, LLENTRY_DELETED, LLENTRY_EXPIRED all mean
* this entry is going to be deleted.
*/
lladdr = NULL;
} else {
KASSERT(lle->la_flags & LLE_VALID,
("%s: %p resolved but not valid?", __func__, lle));
lladdr = (uint8_t *)lle->ll_addr;
VLAN_TAG(ifp, &vid);
VLAN_PCP(ifp, &pcp);
}
tod->tod_l2_update(tod, ifp, sa, lladdr, EVL_MAKETAG(vid, pcp, 0));
}
/*
* Returns 0 or EWOULDBLOCK on success (any other value is an error). 0 means
* lladdr and vtag are valid on return, EWOULDBLOCK means the TOE driver's
* tod_l2_update will be called later, when the entry is resolved or times out.
*/
int
toe_l2_resolve(struct toedev *tod, struct ifnet *ifp, struct sockaddr *sa,
uint8_t *lladdr, uint16_t *vtag)
{
int rc;
uint16_t vid, pcp;
switch (sa->sa_family) {
#ifdef INET
case AF_INET:
rc = arpresolve(ifp, 0, NULL, sa, lladdr, NULL, NULL);
break;
#endif
#ifdef INET6
case AF_INET6:
rc = nd6_resolve(ifp, LLE_SF(AF_INET6, 0), NULL, sa, lladdr,
NULL, NULL);
break;
#endif
default:
return (EPROTONOSUPPORT);
}
if (rc == 0) {
vid = 0xfff;
pcp = 0;
if (ifp->if_type == IFT_L2VLAN) {
VLAN_TAG(ifp, &vid);
VLAN_PCP(ifp, &pcp);
} else if (ifp->if_pcp != IFNET_PCP_NONE) {
vid = 0;
pcp = ifp->if_pcp;
}
*vtag = EVL_MAKETAG(vid, pcp, 0);
}
return (rc);
}
void
toe_connect_failed(struct toedev *tod, struct inpcb *inp, int err)
{
NET_EPOCH_ASSERT();
INP_WLOCK_ASSERT(inp);
if (!(inp->inp_flags & INP_DROPPED)) {
struct tcpcb *tp = intotcpcb(inp);
KASSERT(tp->t_flags & TF_TOE,
("%s: tp %p not offloaded.", __func__, tp));
if (err == EAGAIN) {
/*
* Temporary failure during offload, take this PCB back.
* Detach from the TOE driver and do the rest of what
* TCP's pru_connect would have done if the connection
* wasn't offloaded.
*/
tod->tod_pcb_detach(tod, tp);
KASSERT(!(tp->t_flags & TF_TOE),
("%s: tp %p still offloaded.", __func__, tp));
tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
if (tcp_output(tp) < 0)
INP_WLOCK(inp); /* re-acquire */
} else {
tp = tcp_drop(tp, err);
if (tp == NULL)
INP_WLOCK(inp); /* re-acquire */
}
}
INP_WLOCK_ASSERT(inp);
}
static int
toecore_load(void)
{
mtx_init(&toedev_lock, "toedev lock", NULL, MTX_DEF);
TAILQ_INIT(&toedev_list);
listen_start_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_start,
toe_listen_start_event, NULL, EVENTHANDLER_PRI_ANY);
listen_stop_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_stop,
toe_listen_stop_event, NULL, EVENTHANDLER_PRI_ANY);
lle_event_eh = EVENTHANDLER_REGISTER(lle_event, toe_lle_event, NULL,
EVENTHANDLER_PRI_ANY);
return (0);
}
static int
toecore_unload(void)
{
mtx_lock(&toedev_lock);
if (!TAILQ_EMPTY(&toedev_list)) {
mtx_unlock(&toedev_lock);
return (EBUSY);
}
EVENTHANDLER_DEREGISTER(tcp_offload_listen_start, listen_start_eh);
EVENTHANDLER_DEREGISTER(tcp_offload_listen_stop, listen_stop_eh);
EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
mtx_unlock(&toedev_lock);
mtx_destroy(&toedev_lock);
return (0);
}
static int
toecore_mod_handler(module_t mod, int cmd, void *arg)
{
if (cmd == MOD_LOAD)
return (toecore_load());
if (cmd == MOD_UNLOAD)
return (toecore_unload());
return (EOPNOTSUPP);
}
static moduledata_t mod_data= {
"toecore",
toecore_mod_handler,
0
};
MODULE_VERSION(toecore, 1);
DECLARE_MODULE(toecore, mod_data, SI_SUB_EXEC, SI_ORDER_ANY);
diff --git a/sys/netinet/toecore.h b/sys/netinet/toecore.h
index 746d21c138f0..a8e5afd6b50a 100644
--- a/sys/netinet/toecore.h
+++ b/sys/netinet/toecore.h
@@ -1,146 +1,146 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2012 Chelsio Communications, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#ifndef _NETINET_TOE_H_
#define _NETINET_TOE_H_
#ifndef _KERNEL
#error "no user-serviceable parts inside"
#endif
#include <netinet/tcp.h>
#include <sys/_eventhandler.h>
struct tcpopt;
struct tcphdr;
struct in_conninfo;
struct tcp_info;
struct nhop_object;
struct ktls_session;
struct toedev {
TAILQ_ENTRY(toedev) link; /* glue for toedev_list */
void *tod_softc; /* TOE driver private data */
/*
* Active open. If a failure occurs, it is reported back by the driver
* via toe_connect_failed.
*/
int (*tod_connect)(struct toedev *, struct socket *, struct nhop_object *,
struct sockaddr *);
/* Passive open. */
int (*tod_listen_start)(struct toedev *, struct tcpcb *);
int (*tod_listen_stop)(struct toedev *, struct tcpcb *);
/*
* The kernel uses this routine to pass on any frame it receives for an
* offloaded connection to the TOE driver. This is an unusual event.
*/
void (*tod_input)(struct toedev *, struct tcpcb *, struct mbuf *);
/*
* This is called by the kernel during pru_rcvd for an offloaded TCP
* connection and provides an opportunity for the TOE driver to manage
* its rx window and credits.
*/
void (*tod_rcvd)(struct toedev *, struct tcpcb *);
/*
* Transmit routine. The kernel calls this to have the TOE driver
* evaluate whether there is data to be transmitted, and transmit it.
*/
int (*tod_output)(struct toedev *, struct tcpcb *);
/* Immediate teardown: send RST to peer. */
int (*tod_send_rst)(struct toedev *, struct tcpcb *);
/* Initiate orderly disconnect by sending FIN to the peer. */
int (*tod_send_fin)(struct toedev *, struct tcpcb *);
/* Called to indicate that the kernel is done with this TCP PCB. */
void (*tod_pcb_detach)(struct toedev *, struct tcpcb *);
/*
* The kernel calls this once it has information about an L2 entry that
* the TOE driver enquired about previously (via toe_l2_resolve).
*/
void (*tod_l2_update)(struct toedev *, struct ifnet *,
struct sockaddr *, uint8_t *, uint16_t);
/* XXX. Route has been redirected. */
void (*tod_route_redirect)(struct toedev *, struct ifnet *,
struct nhop_object *, struct nhop_object *);
/* Syncache interaction. */
void (*tod_syncache_added)(struct toedev *, void *);
void (*tod_syncache_removed)(struct toedev *, void *);
int (*tod_syncache_respond)(struct toedev *, void *, struct mbuf *);
void (*tod_offload_socket)(struct toedev *, void *, struct socket *);
/* TCP socket option */
void (*tod_ctloutput)(struct toedev *, struct tcpcb *, int, int);
/* Update software state */
- void (*tod_tcp_info)(struct toedev *, struct tcpcb *,
+ void (*tod_tcp_info)(struct toedev *, const struct tcpcb *,
struct tcp_info *);
/* Create a TLS session */
int (*tod_alloc_tls_session)(struct toedev *, struct tcpcb *,
struct ktls_session *, int);
/* ICMP fragmentation-needed received, adjust PMTU. */
void (*tod_pmtu_update)(struct toedev *, struct tcpcb *, tcp_seq, int);
};
typedef void (*tcp_offload_listen_start_fn)(void *, struct tcpcb *);
typedef void (*tcp_offload_listen_stop_fn)(void *, struct tcpcb *);
EVENTHANDLER_DECLARE(tcp_offload_listen_start, tcp_offload_listen_start_fn);
EVENTHANDLER_DECLARE(tcp_offload_listen_stop, tcp_offload_listen_stop_fn);
void init_toedev(struct toedev *);
int register_toedev(struct toedev *);
int unregister_toedev(struct toedev *);
/*
* General interface for looking up L2 information for an IP address. If an
* answer is not available right away then the TOE driver's tod_l2_update will
* be called later.
*/
int toe_l2_resolve(struct toedev *, struct ifnet *, struct sockaddr *,
uint8_t *, uint16_t *);
void toe_connect_failed(struct toedev *, struct inpcb *, int);
void toe_syncache_add(struct in_conninfo *, struct tcpopt *, struct tcphdr *,
struct inpcb *, void *, void *, uint8_t);
int toe_syncache_expand(struct in_conninfo *, struct tcpopt *, struct tcphdr *,
struct socket **);
int toe_4tuple_check(struct in_conninfo *, struct tcphdr *, struct ifnet *);
#endif