static int rping_create_qp(struct rping_cb *cb)
{
struct ibv_qp_init_attr init_attr;
int ret;
memset(&init_attr, 0, sizeof(init_attr));
init_attr.cap.max_send_wr = RPING_SQ_DEPTH;
init_attr.cap.max_recv_wr = 2;
init_attr.cap.max_recv_sge = 1;
init_attr.cap.max_send_sge = 1;
init_attr.qp_type = IBV_QPT_RC;
init_attr.send_cq = cb->cq;
init_attr.recv_cq = cb->cq;
if (cb->server) {
ret = rdma_create_qp(cb->child_cm_id, cb->pd, &init_attr);
if (!ret)
cb->qp = cb->child_cm_id->qp;
} else {
ret = rdma_create_qp(cb->cm_id, cb->pd, &init_attr);
if (!ret)
cb->qp = cb->cm_id->qp;
}
return ret;
}
int on_addr_resolved(struct rdma_cm_id *id)
{
struct ibv_qp_init_attr qp_attr;
struct connection *conn;
printf("address resolved.\n");
build_context(id->verbs);
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id, s_ctx->pd, &qp_attr));
id->context = conn = (struct connection *)malloc(sizeof(struct connection));
conn->id = id;
conn->qp = id->qp;
conn->num_completions = 0;
register_memory(conn);
post_receives(conn);
TEST_NZ(rdma_resolve_route(id, TIMEOUT_IN_MS));
return 0;
}
static int init_node(struct cmatest_node *node)
{
struct ibv_qp_init_attr init_qp_attr;
int cqe, ret;
int i;
struct ibv_cq **cqs[] = {&node->cq[SEND_CQ_INDEX],
&node->cq[RECV_CQ_INDEX]};
node->pd = ibv_alloc_pd(node->cma_id->verbs);
if (!node->pd) {
ret = -ENOMEM;
printf("cmatose: unable to allocate PD\n");
goto out;
}
cqe = message_count ? message_count : 1;
for (i = 0; i < sizeof(cqs)/sizeof(cqs[0]); i++) {
if (set_ts) {
struct ibv_exp_cq_init_attr cq_init_attr;
memset(&cq_init_attr, 0, sizeof(cq_init_attr));
cq_init_attr.flags = IBV_EXP_CQ_TIMESTAMP;
cq_init_attr.comp_mask = IBV_EXP_CQ_INIT_ATTR_FLAGS;
*cqs[i] = (struct ibv_cq *)ibv_exp_create_cq(
node->cma_id->verbs, cqe, node,
NULL, 0, &cq_init_attr);
} else {
*cqs[i] = ibv_create_cq(node->cma_id->verbs, cqe, node,
0, 0);
}
}
if (!node->cq[SEND_CQ_INDEX] || !node->cq[RECV_CQ_INDEX]) {
ret = -ENOMEM;
printf("cmatose: unable to create CQ\n");
goto out;
}
memset(&init_qp_attr, 0, sizeof init_qp_attr);
init_qp_attr.cap.max_send_wr = cqe;
init_qp_attr.cap.max_recv_wr = cqe;
init_qp_attr.cap.max_send_sge = 1;
init_qp_attr.cap.max_recv_sge = 1;
init_qp_attr.qp_context = node;
init_qp_attr.sq_sig_all = 1;
init_qp_attr.qp_type = IBV_QPT_RC;
init_qp_attr.send_cq = node->cq[SEND_CQ_INDEX];
init_qp_attr.recv_cq = node->cq[RECV_CQ_INDEX];
ret = rdma_create_qp(node->cma_id, node->pd, &init_qp_attr);
if (ret) {
perror("cmatose: unable to create QP");
goto out;
}
ret = create_message(node);
if (ret) {
printf("cmatose: failed to create messages: %d\n", ret);
goto out;
}
out:
return ret;
}
void build_connection(struct rdma_cm_id *id)
{
struct connection *conn;
struct ibv_qp_init_attr qp_attr;
//init semaphores
sem_init(&read_ops, 0, 0);
sem_init(&done_ops, 0, 0);
sem_init(&write_ops, 0, 1);
build_context(id->verbs);
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id, s_ctx->pd, &qp_attr));
id->context = conn = (struct connection *)malloc(sizeof(struct connection));
conn->id = id;
conn->qp = id->qp;
conn->send_state = SS_INIT;
conn->recv_state = RS_INIT;
conn->connected = 0;
register_memory(conn);
post_receives(conn);
}
void build_connection(struct rdma_cm_id *id)
{
rdma_conn_t *conn;
struct ibv_qp_init_attr qp_attr;
build_context(id->verbs);
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id, s_ctx->pd, &qp_attr));
conn = malloc(sizeof(rdma_conn_t));
id->context = conn;
rdma_conn = conn;
conn->id = id;
conn->qp = id->qp;
conn->send_state = SS_INIT;
conn->recv_state = RS_INIT;
conn->connected = 0;
register_memory(conn);
post_receives(conn);
}
void RDMAServerSocket::accept(client_t client_id) const {
ibv_qp_init_attr qp_attr = {};
qp_attr.qp_type = IBV_QPT_RC;
qp_attr.cap.max_send_wr = 256;
qp_attr.cap.max_recv_wr = 0;
qp_attr.cap.max_send_sge = 1;
qp_attr.cap.max_recv_sge = 0;
qp_attr.cap.max_inline_data = 72;
qp_attr.recv_cq = cq;
qp_attr.send_cq = cq;
qp_attr.srq = id->srq;
qp_attr.sq_sig_all = 1;
check_zero(rdma_create_qp(client_id.get(), NULL, &qp_attr));
check_zero(rdma_accept(client_id.get(), nullptr));
clients([client_id = std::move(client_id)](auto && clients) mutable {
auto pos = std::lower_bound(std::begin(clients), std::end(clients),
client_id->qp->qp_num,
[](const auto &client, const qp_t &qp_num) {
return client->qp->qp_num < qp_num;
});
clients.insert(pos, std::move(client_id));
});
}
static int isert_conn_qp_create(struct isert_connection *isert_conn)
{
struct rdma_cm_id *cm_id = isert_conn->cm_id;
struct isert_device *isert_dev = isert_conn->isert_dev;
struct ib_qp_init_attr qp_attr;
int err;
int cq_idx;
int max_wr = ISER_MAX_WCE;
TRACE_ENTRY();
cq_idx = isert_get_cq_idx(isert_dev);
memset(&qp_attr, 0, sizeof(qp_attr));
qp_attr.event_handler = isert_async_evt_handler;
qp_attr.qp_context = isert_conn;
qp_attr.send_cq = isert_dev->cq_desc[cq_idx].cq;
qp_attr.recv_cq = isert_dev->cq_desc[cq_idx].cq;
isert_conn->cq_desc = &isert_dev->cq_desc[cq_idx];
qp_attr.cap.max_send_sge = isert_conn->max_sge;
qp_attr.cap.max_recv_sge = 3;
qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
qp_attr.qp_type = IB_QPT_RC;
do {
if (max_wr < ISER_MIN_SQ_SIZE) {
pr_err("Failed to create qp, not enough memory\n");
goto fail_create_qp;
}
qp_attr.cap.max_send_wr = max_wr;
qp_attr.cap.max_recv_wr = max_wr;
err = rdma_create_qp(cm_id, isert_dev->pd, &qp_attr);
if (err && err != -ENOMEM) {
pr_err("Failed to create qp, err:%d\n", err);
goto fail_create_qp;
}
max_wr /= 2;
} while (err == -ENOMEM);
isert_conn->qp = cm_id->qp;
pr_info("iser created cm_id:%p qp:0x%X\n", cm_id, cm_id->qp->qp_num);
out:
TRACE_EXIT_RES(err);
return err;
fail_create_qp:
mutex_lock(&dev_list_mutex);
isert_dev->cq_qps[cq_idx]--;
mutex_unlock(&dev_list_mutex);
goto out;
}
void Connector::build_conn(struct rdma_cm_id* id_)
{
struct ibv_qp_init_attr qp_attr;
build_context(id_->verbs);
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id_, s_ctx_->pd_, &qp_attr) )
}
static ssize_t
fi_ibv_rdm_process_addr_resolved(struct rdma_cm_id *id,
struct fi_ibv_rdm_ep *ep)
{
ssize_t ret = FI_SUCCESS;
struct ibv_qp_init_attr qp_attr;
struct fi_ibv_rdm_tagged_conn *conn = id->context;
VERBS_INFO(FI_LOG_AV, "ADDR_RESOLVED conn %p, addr %s:%u\n",
conn, inet_ntoa(conn->addr.sin_addr),
ntohs(conn->addr.sin_port));
assert(id->verbs == ep->domain->verbs);
do {
fi_ibv_rdm_tagged_init_qp_attributes(&qp_attr, ep);
if (rdma_create_qp(id, ep->domain->pd, &qp_attr)) {
VERBS_INFO_ERRNO(FI_LOG_AV,
"rdma_create_qp failed\n", errno);
return -errno;
}
if (conn->cm_role == FI_VERBS_CM_PASSIVE) {
break;
}
conn->qp[0] = id->qp;
assert(conn->id[0] == id);
if (conn->cm_role == FI_VERBS_CM_SELF) {
break;
}
ret = fi_ibv_rdm_prepare_conn_memory(ep, conn);
if (ret != FI_SUCCESS) {
goto err;
}
ret = fi_ibv_rdm_repost_receives(conn, ep, ep->rq_wr_depth);
if (ret < 0) {
VERBS_INFO(FI_LOG_AV, "repost receives failed\n");
goto err;
} else {
ret = FI_SUCCESS;
}
} while (0);
if (rdma_resolve_route(id, FI_IBV_RDM_CM_RESOLVEADDR_TIMEOUT)) {
VERBS_INFO(FI_LOG_AV, "rdma_resolve_route failed\n");
ret = -FI_EHOSTUNREACH;
goto err;
}
return ret;
err:
rdma_destroy_qp(id);
return ret;
}
/**
* @param[in] ni
* @param[in] conn
* @param[in] event
*
* @return status
*
* conn is locked
*/
static int accept_connection_request(ni_t *ni, conn_t *conn,
struct rdma_cm_event *event)
{
struct rdma_conn_param conn_param;
struct ibv_qp_init_attr init_attr;
struct cm_priv_accept priv;
conn->state = CONN_STATE_CONNECTING;
memset(&init_attr, 0, sizeof(init_attr));
init_attr.qp_type = IBV_QPT_RC;
init_attr.cap.max_send_wr = ni->iface->cap.max_send_wr;
init_attr.send_cq = ni->rdma.cq;
init_attr.recv_cq = ni->rdma.cq;
init_attr.srq = ni->rdma.srq;
init_attr.cap.max_send_sge = ni->iface->cap.max_send_sge;
if (rdma_create_qp(event->id, ni->iface->pd, &init_attr)) {
conn->state = CONN_STATE_DISCONNECTED;
pthread_cond_broadcast(&conn->move_wait);
return PTL_FAIL;
}
/* If we were already trying to connect ourselves, cancel it. */
if (conn->rdma.cm_id != NULL) {
assert(conn->rdma.cm_id->context == conn);
conn->rdma.cm_id->context = NULL;
}
event->id->context = conn;
conn->rdma.cm_id = event->id;
memset(&conn_param, 0, sizeof conn_param);
conn_param.responder_resources = 1;
conn_param.initiator_depth = 1;
conn_param.retry_count = 7;
conn_param.rnr_retry_count = 7;
if (ni->options & PTL_NI_LOGICAL) {
conn_param.private_data = &priv;
conn_param.private_data_len = sizeof(priv);
}
if (rdma_accept(event->id, &conn_param)) {
rdma_destroy_qp(event->id);
conn->rdma.cm_id = NULL;
conn->state = CONN_STATE_DISCONNECTED;
pthread_cond_broadcast(&conn->move_wait);
return PTL_FAIL;
}
return PTL_OK;
}
/*
* Connect unconnected endpoint.
*/
int
rpcrdma_ep_connect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
{
struct rdma_cm_id *id, *old;
int rc = 0;
int retry_count = 0;
if (ep->rep_connected != 0) {
struct rpcrdma_xprt *xprt;
retry:
dprintk("RPC: %s: reconnecting...\n", __func__);
rpcrdma_ep_disconnect(ep, ia);
rpcrdma_flush_cqs(ep);
xprt = container_of(ia, struct rpcrdma_xprt, rx_ia);
ia->ri_ops->ro_reset(xprt);
id = rpcrdma_create_id(xprt, ia,
(struct sockaddr *)&xprt->rx_data.addr);
if (IS_ERR(id)) {
rc = -EHOSTUNREACH;
goto out;
}
/* TEMP TEMP TEMP - fail if new device:
* Deregister/remarshal *all* requests!
* Close and recreate adapter, pd, etc!
* Re-determine all attributes still sane!
* More stuff I haven't thought of!
* Rrrgh!
*/
if (ia->ri_id->device != id->device) {
printk("RPC: %s: can't reconnect on "
"different device!\n", __func__);
rdma_destroy_id(id);
rc = -ENETUNREACH;
goto out;
}
/* END TEMP */
rc = rdma_create_qp(id, ia->ri_pd, &ep->rep_attr);
if (rc) {
dprintk("RPC: %s: rdma_create_qp failed %i\n",
__func__, rc);
rdma_destroy_id(id);
rc = -ENETUNREACH;
goto out;
}
write_lock(&ia->ri_qplock);
old = ia->ri_id;
ia->ri_id = id;
write_unlock(&ia->ri_qplock);
rdma_destroy_qp(old);
rdma_destroy_id(old);
} else {
void IBConnection::create_qp(struct ibv_pd* pd, struct ibv_cq* cq)
{
struct ibv_qp_init_attr qp_attr;
memset(&qp_attr, 0, sizeof qp_attr);
qp_attr.cap = qp_cap_;
qp_attr.send_cq = cq;
qp_attr.recv_cq = cq;
qp_attr.qp_type = IBV_QPT_RC;
int err = rdma_create_qp(cm_id_, pd, &qp_attr);
if (err)
throw InfinibandException("creation of QP failed");
}
void build_connection(struct rdma_cm_id *id){
struct ibv_qp_init_attr qp_attr;
struct connection *conn;
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id, s_ctx->pd, &qp_attr));
id->context = conn = (struct connection *)malloc(sizeof(struct connection));
conn->id = id;
conn->qp = id->qp;
}
static int init_node(struct cmatest_node *node)
{
struct ibv_qp_init_attr init_qp_attr;
int cqe, ret;
node->pd = ibv_alloc_pd(node->cma_id->verbs);
if (!node->pd) {
ret = -ENOMEM;
printf("cmatose: unable to allocate PD\n");
goto out;
}
cqe = message_count ? message_count : 1;
node->cq[SEND_CQ_INDEX] = ibv_create_cq(node->cma_id->verbs, cqe, node, NULL, 0);
node->cq[RECV_CQ_INDEX] = ibv_create_cq(node->cma_id->verbs, cqe, node, NULL, 0);
if (!node->cq[SEND_CQ_INDEX] || !node->cq[RECV_CQ_INDEX]) {
ret = -ENOMEM;
printf("cmatose: unable to create CQ\n");
goto out;
}
memset(&init_qp_attr, 0, sizeof init_qp_attr);
init_qp_attr.cap.max_send_wr = cqe;
init_qp_attr.cap.max_recv_wr = cqe;
init_qp_attr.cap.max_send_sge = 1;
init_qp_attr.cap.max_recv_sge = 1;
init_qp_attr.qp_context = node;
init_qp_attr.sq_sig_all = 1;
init_qp_attr.qp_type = IBV_QPT_RC;
init_qp_attr.send_cq = node->cq[SEND_CQ_INDEX];
init_qp_attr.recv_cq = node->cq[RECV_CQ_INDEX];
ret = rdma_create_qp(node->cma_id, node->pd, &init_qp_attr);
if (ret) {
perror("cmatose: unable to create QP");
goto out;
}
ret = create_message(node);
if (ret) {
printf("cmatose: failed to create messages: %d\n", ret);
goto out;
}
out:
return ret;
}
/**
* Accept an RC connection request to self.
*
* called while holding connect->mutex
* only used for physical NIs
*
* @param[in] ni
* @param[in] conn
* @param[in] event
*
* @return status
*/
static int accept_connection_self(ni_t *ni, conn_t *conn,
struct rdma_cm_event *event)
{
struct rdma_conn_param conn_param;
struct ibv_qp_init_attr init_attr;
conn->state = CONN_STATE_CONNECTING;
memset(&init_attr, 0, sizeof(init_attr));
init_attr.qp_type = IBV_QPT_RC;
init_attr.send_cq = ni->rdma.cq;
init_attr.recv_cq = ni->rdma.cq;
init_attr.srq = ni->rdma.srq;
init_attr.cap.max_send_wr = ni->iface->cap.max_send_wr;
init_attr.cap.max_send_sge = ni->iface->cap.max_send_sge;
if (rdma_create_qp(event->id, ni->iface->pd, &init_attr)) {
conn->state = CONN_STATE_DISCONNECTED;
pthread_cond_broadcast(&conn->move_wait);
return PTL_FAIL;
}
ni->rdma.self_cm_id = event->id;
/* The lower 2 bits (on 32 bits hosts), or 3 bits (on 64 bits
* hosts) of a pointer is always 0. Use it to store the type of
* context. 0=conn; 1=NI. */
event->id->context = (void *)((uintptr_t) ni | 1);
memset(&conn_param, 0, sizeof conn_param);
conn_param.responder_resources = 1;
conn_param.initiator_depth = 1;
conn_param.rnr_retry_count = 7;
if (rdma_accept(event->id, &conn_param)) {
rdma_destroy_qp(event->id);
conn->state = CONN_STATE_DISCONNECTED;
pthread_cond_broadcast(&conn->move_wait);
return PTL_FAIL;
}
return PTL_OK;
}
static int init_node(struct cmatest_node *node)
{
struct ibv_qp_init_attr init_qp_attr;
int cqe, ret;
node->pd = ibv_alloc_pd(node->cma_id->verbs);
if (!node->pd) {
ret = -ENOMEM;
printf("rxe_send_mc: unable to allocate PD\n");
goto out;
}
cqe = message_buffer ? message_buffer * 2 : 2;
node->cq = ibv_create_cq(node->cma_id->verbs, cqe, node, 0, 0);
if (!node->cq) {
ret = -ENOMEM;
printf("rxe_send_mc: unable to create CQ\n");
goto out;
}
memset(&init_qp_attr, 0, sizeof init_qp_attr);
init_qp_attr.cap.max_send_wr = message_buffer ? message_buffer : 1;
init_qp_attr.cap.max_recv_wr = message_buffer ? message_buffer : 1;
init_qp_attr.cap.max_send_sge = 1;
init_qp_attr.cap.max_recv_sge = 1;
init_qp_attr.qp_context = node;
init_qp_attr.sq_sig_all = 1; //singal all
init_qp_attr.qp_type = IBV_QPT_UD;
init_qp_attr.send_cq = node->cq;
init_qp_attr.recv_cq = node->cq;
ret = rdma_create_qp(node->cma_id, node->pd, &init_qp_attr);
if (ret) {
perror("rxe_send_mc: unable to create QP");
goto out;
}
ret = create_message(node);
if (ret) {
printf("rxe_send_mc: failed to create messages: %d\n", ret);
goto out;
}
out:
return ret;
}
void IBConnection::on_addr_resolved(struct ibv_pd* pd, struct ibv_cq* cq)
{
L_(debug) << "address resolved";
struct ibv_qp_init_attr qp_attr;
memset(&qp_attr, 0, sizeof qp_attr);
qp_attr.cap = qp_cap_;
qp_attr.send_cq = cq;
qp_attr.recv_cq = cq;
qp_attr.qp_type = IBV_QPT_RC;
int err = rdma_create_qp(cm_id_, pd, &qp_attr);
if (err)
throw InfinibandException("creation of QP failed");
err = rdma_resolve_route(cm_id_, RESOLVE_TIMEOUT_MS);
if (err)
throw InfinibandException("rdma_resolve_route failed");
setup(pd);
}
void build_connection(struct rdma_cm_id *id)
{
IbvConnection *conn;
struct ibv_qp_init_attr qp_attr;
id->context = conn = (IbvConnection *)malloc(sizeof(IbvConnection));
build_verbs(conn, id->verbs);
build_qp_attr(conn, &qp_attr);
TEST_NZ(rdma_create_qp(id, conn->pd, &qp_attr));
conn->id = id;
conn->qp = id->qp;
conn->connected = 0;
register_memory(conn);
post_receives(conn);
}
void build_connection(struct rdma_cm_id *id){
struct ibv_qp_init_attr qp_attr;
struct connection *conn;
struct timeval start, end, dt;
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id, s_ctx->pd, &qp_attr));
id->context = conn = (struct connection *)malloc(sizeof(struct connection));
conn->id = id;
conn->qp = id->qp;
gettimeofday(&start, NULL);
register_memory(conn);
gettimeofday(&end, NULL);
timersub(&end, &start, &dt);
long usec = dt.tv_usec + 1000000 * dt.tv_sec;
printf("[Register] takes %ld micro_secs.\n", usec);
}
int on_connect_request(struct rdma_cm_id *id)
{
struct ibv_qp_init_attr qp_attr;
struct rdma_conn_param cm_params;
struct connection *conn;
printf("received connection request.\n");
build_context(id->verbs);
build_qp_attr(&qp_attr);
TEST_NZ(rdma_create_qp(id, s_ctx->pd, &qp_attr));
id->context = conn = (struct connection *)malloc(sizeof(struct connection));
conn->qp = id->qp;
register_memory(conn);
post_receives(conn);
memset(&cm_params, 0, sizeof(cm_params));
TEST_NZ(rdma_accept(id, &cm_params));
return 0;
}
}
/* END TEMP */
rc = rdma_create_qp(id, ia->ri_pd, &ep->rep_attr);
if (rc) {
dprintk("RPC: %s: rdma_create_qp failed %i\n",
__func__, rc);
rdma_destroy_id(id);
rc = -ENETUNREACH;
goto out;
}
rdma_destroy_qp(ia->ri_id);
rdma_destroy_id(ia->ri_id);
ia->ri_id = id;
} else {
dprintk("RPC: %s: connecting...\n", __func__);
rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &ep->rep_attr);
if (rc) {
dprintk("RPC: %s: rdma_create_qp failed %i\n",
__func__, rc);
/* do not update ep->rep_connected */
return -ENETUNREACH;
}
}
ep->rep_connected = 0;
rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma);
if (rc) {
dprintk("RPC: %s: rdma_connect() failed with %i\n",
__func__, rc);
goto out;
/**
* iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
*
* returns 0 on success, -1 on failure
*/
static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
{
struct iser_device *device;
struct ib_qp_init_attr init_attr;
int ret = -ENOMEM;
struct ib_fmr_pool_param params;
BUG_ON(ib_conn->device == NULL);
device = ib_conn->device;
ib_conn->login_buf = kmalloc(ISER_RX_LOGIN_SIZE, GFP_KERNEL);
if (!ib_conn->login_buf) {
goto alloc_err;
ret = -ENOMEM;
}
ib_conn->login_dma = ib_dma_map_single(ib_conn->device->ib_device,
(void *)ib_conn->login_buf, ISER_RX_LOGIN_SIZE,
DMA_FROM_DEVICE);
ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
(sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
GFP_KERNEL);
if (!ib_conn->page_vec) {
ret = -ENOMEM;
goto alloc_err;
}
ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
params.page_shift = SHIFT_4K;
/* when the first/last SG element are not start/end *
* page aligned, the map whould be of N+1 pages */
params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
/* make the pool size twice the max number of SCSI commands *
* the ML is expected to queue, watermark for unmap at 50% */
params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
params.cache = 0;
params.flush_function = NULL;
params.access = (IB_ACCESS_LOCAL_WRITE |
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_REMOTE_READ);
ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
if (IS_ERR(ib_conn->fmr_pool)) {
ret = PTR_ERR(ib_conn->fmr_pool);
goto fmr_pool_err;
}
memset(&init_attr, 0, sizeof init_attr);
init_attr.event_handler = iser_qp_event_callback;
init_attr.qp_context = (void *)ib_conn;
init_attr.send_cq = device->tx_cq;
init_attr.recv_cq = device->rx_cq;
init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
init_attr.cap.max_send_sge = 2;
init_attr.cap.max_recv_sge = 1;
init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
init_attr.qp_type = IB_QPT_RC;
ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
if (ret)
goto qp_err;
ib_conn->qp = ib_conn->cma_id->qp;
iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
ib_conn, ib_conn->cma_id,
ib_conn->fmr_pool, ib_conn->cma_id->qp);
return ret;
qp_err:
(void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
fmr_pool_err:
kfree(ib_conn->page_vec);
kfree(ib_conn->login_buf);
alloc_err:
iser_err("unable to alloc mem or create resource, err %d\n", ret);
return ret;
}
/*
* This needs to be very careful to not leave IS_ERR pointers around for
* cleanup to trip over.
*/
static int rds_iw_setup_qp(struct rds_connection *conn)
{
struct rds_iw_connection *ic = conn->c_transport_data;
struct ib_device *dev = ic->i_cm_id->device;
struct ib_qp_init_attr attr;
struct rds_iw_device *rds_iwdev;
int ret;
/* rds_iw_add_one creates a rds_iw_device object per IB device,
* and allocates a protection domain, memory range and MR pool
* for each. If that fails for any reason, it will not register
* the rds_iwdev at all.
*/
rds_iwdev = ib_get_client_data(dev, &rds_iw_client);
if (!rds_iwdev) {
if (printk_ratelimit())
printk(KERN_NOTICE "RDS/IW: No client_data for device %s\n",
dev->name);
return -EOPNOTSUPP;
}
/* Protection domain and memory range */
ic->i_pd = rds_iwdev->pd;
ic->i_mr = rds_iwdev->mr;
ret = rds_iw_init_qp_attrs(&attr, rds_iwdev,
&ic->i_send_ring, rds_iw_send_cq_comp_handler,
&ic->i_recv_ring, rds_iw_recv_cq_comp_handler,
conn);
if (ret < 0)
goto out;
ic->i_send_cq = attr.send_cq;
ic->i_recv_cq = attr.recv_cq;
/*
* XXX this can fail if max_*_wr is too large? Are we supposed
* to back off until we get a value that the hardware can support?
*/
ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
if (ret) {
rdsdebug("rdma_create_qp failed: %d\n", ret);
goto out;
}
ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
ic->i_send_ring.w_nr *
sizeof(struct rds_header),
&ic->i_send_hdrs_dma, GFP_KERNEL);
if (!ic->i_send_hdrs) {
ret = -ENOMEM;
rdsdebug("ib_dma_alloc_coherent send failed\n");
goto out;
}
ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
ic->i_recv_ring.w_nr *
sizeof(struct rds_header),
&ic->i_recv_hdrs_dma, GFP_KERNEL);
if (!ic->i_recv_hdrs) {
ret = -ENOMEM;
rdsdebug("ib_dma_alloc_coherent recv failed\n");
goto out;
}
ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
&ic->i_ack_dma, GFP_KERNEL);
if (!ic->i_ack) {
ret = -ENOMEM;
rdsdebug("ib_dma_alloc_coherent ack failed\n");
goto out;
}
ic->i_sends = vmalloc(ic->i_send_ring.w_nr * sizeof(struct rds_iw_send_work));
if (!ic->i_sends) {
ret = -ENOMEM;
rdsdebug("send allocation failed\n");
goto out;
}
rds_iw_send_init_ring(ic);
ic->i_recvs = vmalloc(ic->i_recv_ring.w_nr * sizeof(struct rds_iw_recv_work));
if (!ic->i_recvs) {
ret = -ENOMEM;
rdsdebug("recv allocation failed\n");
goto out;
}
rds_iw_recv_init_ring(ic);
rds_iw_recv_init_ack(ic);
/* Post receive buffers - as a side effect, this will update
* the posted credit count. */
rds_iw_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 1);
rdsdebug("conn %p pd %p mr %p cq %p %p\n", conn, ic->i_pd, ic->i_mr,
ic->i_send_cq, ic->i_recv_cq);
out:
return ret;
}
static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
{
struct iser_device *device;
struct ib_qp_init_attr init_attr;
int req_err, resp_err, ret = -ENOMEM;
struct ib_fmr_pool_param params;
BUG_ON(ib_conn->device == NULL);
device = ib_conn->device;
ib_conn->login_buf = kmalloc(ISCSI_DEF_MAX_RECV_SEG_LEN +
ISER_RX_LOGIN_SIZE, GFP_KERNEL);
if (!ib_conn->login_buf)
goto out_err;
ib_conn->login_req_buf = ib_conn->login_buf;
ib_conn->login_resp_buf = ib_conn->login_buf + ISCSI_DEF_MAX_RECV_SEG_LEN;
ib_conn->login_req_dma = ib_dma_map_single(ib_conn->device->ib_device,
(void *)ib_conn->login_req_buf,
ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
ib_conn->login_resp_dma = ib_dma_map_single(ib_conn->device->ib_device,
(void *)ib_conn->login_resp_buf,
ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
req_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_req_dma);
resp_err = ib_dma_mapping_error(device->ib_device, ib_conn->login_resp_dma);
if (req_err || resp_err) {
if (req_err)
ib_conn->login_req_dma = 0;
if (resp_err)
ib_conn->login_resp_dma = 0;
goto out_err;
}
ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
(sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
GFP_KERNEL);
if (!ib_conn->page_vec)
goto out_err;
ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
params.page_shift = SHIFT_4K;
params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
params.cache = 0;
params.flush_function = NULL;
params.access = (IB_ACCESS_LOCAL_WRITE |
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_REMOTE_READ);
ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
if (IS_ERR(ib_conn->fmr_pool)) {
ret = PTR_ERR(ib_conn->fmr_pool);
ib_conn->fmr_pool = NULL;
goto out_err;
}
memset(&init_attr, 0, sizeof init_attr);
init_attr.event_handler = iser_qp_event_callback;
init_attr.qp_context = (void *)ib_conn;
init_attr.send_cq = device->tx_cq;
init_attr.recv_cq = device->rx_cq;
init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
init_attr.cap.max_send_sge = 2;
init_attr.cap.max_recv_sge = 1;
init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
init_attr.qp_type = IB_QPT_RC;
ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
if (ret)
goto out_err;
ib_conn->qp = ib_conn->cma_id->qp;
iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
ib_conn, ib_conn->cma_id,
ib_conn->fmr_pool, ib_conn->cma_id->qp);
return ret;
out_err:
iser_err("unable to alloc mem or create resource, err %d\n", ret);
return ret;
}
static int ibw_setup_cq_qp(struct ibw_conn *conn)
{
struct ibw_ctx_priv *pctx = talloc_get_type(conn->ctx->internal, struct ibw_ctx_priv);
struct ibw_conn_priv *pconn = talloc_get_type(conn->internal, struct ibw_conn_priv);
struct ibv_qp_init_attr init_attr;
struct ibv_qp_attr attr;
int rc;
DEBUG(DEBUG_DEBUG, ("ibw_setup_cq_qp(cmid: %p)\n", pconn->cm_id));
/* init verbs */
pconn->verbs_channel = ibv_create_comp_channel(pconn->cm_id->verbs);
if (!pconn->verbs_channel) {
sprintf(ibw_lasterr, "ibv_create_comp_channel failed %d\n", errno);
return -1;
}
DEBUG(DEBUG_DEBUG, ("created channel %p\n", pconn->verbs_channel));
pconn->verbs_channel_event = tevent_add_fd(pctx->ectx, NULL, /* not pconn or conn */
pconn->verbs_channel->fd, TEVENT_FD_READ, ibw_event_handler_verbs, conn);
pconn->pd = ibv_alloc_pd(pconn->cm_id->verbs);
if (!pconn->pd) {
sprintf(ibw_lasterr, "ibv_alloc_pd failed %d\n", errno);
return -1;
}
DEBUG(DEBUG_DEBUG, ("created pd %p\n", pconn->pd));
/* init mr */
if (ibw_init_memory(conn))
return -1;
/* init cq */
pconn->cq = ibv_create_cq(pconn->cm_id->verbs,
pctx->opts.max_recv_wr + pctx->opts.max_send_wr,
conn, pconn->verbs_channel, 0);
if (pconn->cq==NULL) {
sprintf(ibw_lasterr, "ibv_create_cq failed\n");
return -1;
}
rc = ibv_req_notify_cq(pconn->cq, 0);
if (rc) {
sprintf(ibw_lasterr, "ibv_req_notify_cq failed with %d\n", rc);
return rc;
}
/* init qp */
memset(&init_attr, 0, sizeof(init_attr));
init_attr.cap.max_send_wr = pctx->opts.max_send_wr;
init_attr.cap.max_recv_wr = pctx->opts.max_recv_wr;
init_attr.cap.max_recv_sge = 1;
init_attr.cap.max_send_sge = 1;
init_attr.qp_type = IBV_QPT_RC;
init_attr.send_cq = pconn->cq;
init_attr.recv_cq = pconn->cq;
rc = rdma_create_qp(pconn->cm_id, pconn->pd, &init_attr);
if (rc) {
sprintf(ibw_lasterr, "rdma_create_qp failed with %d\n", rc);
return rc;
}
/* elase result is in pconn->cm_id->qp */
rc = ibv_query_qp(pconn->cm_id->qp, &attr, IBV_QP_PATH_MTU, &init_attr);
if (rc) {
sprintf(ibw_lasterr, "ibv_query_qp failed with %d\n", rc);
return rc;
}
return ibw_fill_cq(conn);
}
static int p9_rdma_bind_privport(struct p9_trans_rdma *rdma)
{
struct sockaddr_in cl = {
.sin_family = AF_INET,
.sin_addr.s_addr = htonl(INADDR_ANY),
};
int port, err = -EINVAL;
for (port = P9_DEF_MAX_RESVPORT; port >= P9_DEF_MIN_RESVPORT; port--) {
cl.sin_port = htons((ushort)port);
err = rdma_bind_addr(rdma->cm_id, (struct sockaddr *)&cl);
if (err != -EADDRINUSE)
break;
}
return err;
}
/**
* trans_create_rdma - Transport method for creating atransport instance
* @client: client instance
* @addr: IP address string
* @args: Mount options string
*/
static int
rdma_create_trans(struct p9_client *client, const char *addr, char *args)
{
int err;
struct p9_rdma_opts opts;
struct p9_trans_rdma *rdma;
struct rdma_conn_param conn_param;
struct ib_qp_init_attr qp_attr;
struct ib_device_attr devattr;
struct ib_cq_init_attr cq_attr = {};
/* Parse the transport specific mount options */
err = parse_opts(args, &opts);
if (err < 0)
return err;
/* Create and initialize the RDMA transport structure */
rdma = alloc_rdma(&opts);
if (!rdma)
return -ENOMEM;
/* Create the RDMA CM ID */
rdma->cm_id = rdma_create_id(p9_cm_event_handler, client, RDMA_PS_TCP,
IB_QPT_RC);
if (IS_ERR(rdma->cm_id))
goto error;
/* Associate the client with the transport */
client->trans = rdma;
/* Bind to a privileged port if we need to */
if (opts.privport) {
err = p9_rdma_bind_privport(rdma);
if (err < 0) {
pr_err("%s (%d): problem binding to privport: %d\n",
__func__, task_pid_nr(current), -err);
goto error;
}
}
/* Resolve the server's address */
rdma->addr.sin_family = AF_INET;
rdma->addr.sin_addr.s_addr = in_aton(addr);
rdma->addr.sin_port = htons(opts.port);
err = rdma_resolve_addr(rdma->cm_id, NULL,
(struct sockaddr *)&rdma->addr,
rdma->timeout);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
goto error;
/* Resolve the route to the server */
err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
goto error;
/* Query the device attributes */
err = ib_query_device(rdma->cm_id->device, &devattr);
if (err)
goto error;
/* Create the Completion Queue */
cq_attr.cqe = opts.sq_depth + opts.rq_depth + 1;
rdma->cq = ib_create_cq(rdma->cm_id->device, cq_comp_handler,
cq_event_handler, client,
&cq_attr);
if (IS_ERR(rdma->cq))
goto error;
ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
/* Create the Protection Domain */
rdma->pd = ib_alloc_pd(rdma->cm_id->device);
if (IS_ERR(rdma->pd))
goto error;
/* Cache the DMA lkey in the transport */
rdma->dma_mr = NULL;
if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
rdma->lkey = rdma->cm_id->device->local_dma_lkey;
else {
rdma->dma_mr = ib_get_dma_mr(rdma->pd, IB_ACCESS_LOCAL_WRITE);
if (IS_ERR(rdma->dma_mr))
goto error;
rdma->lkey = rdma->dma_mr->lkey;
}
/* Create the Queue Pair */
memset(&qp_attr, 0, sizeof qp_attr);
qp_attr.event_handler = qp_event_handler;
qp_attr.qp_context = client;
qp_attr.cap.max_send_wr = opts.sq_depth;
qp_attr.cap.max_recv_wr = opts.rq_depth;
qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
qp_attr.qp_type = IB_QPT_RC;
qp_attr.send_cq = rdma->cq;
qp_attr.recv_cq = rdma->cq;
err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
if (err)
goto error;
rdma->qp = rdma->cm_id->qp;
/* Request a connection */
memset(&conn_param, 0, sizeof(conn_param));
conn_param.private_data = NULL;
conn_param.private_data_len = 0;
conn_param.responder_resources = P9_RDMA_IRD;
conn_param.initiator_depth = P9_RDMA_ORD;
err = rdma_connect(rdma->cm_id, &conn_param);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_CONNECTED))
goto error;
client->status = Connected;
return 0;
error:
rdma_destroy_trans(rdma);
return -ENOTCONN;
}
static int sdp_init_qp(struct sock *sk, struct rdma_cm_id *id)
{
struct ib_qp_init_attr qp_init_attr = {
.event_handler = sdp_qp_event_handler,
.cap.max_send_wr = SDP_TX_SIZE,
.cap.max_recv_wr = sdp_rx_size,
.cap.max_inline_data = sdp_inline_thresh,
.sq_sig_type = IB_SIGNAL_REQ_WR,
.qp_type = IB_QPT_RC,
};
struct ib_device *device = id->device;
int rc;
sdp_dbg(sk, "%s\n", __func__);
sdp_sk(sk)->max_sge = sdp_get_max_dev_sge(device);
sdp_dbg(sk, "Max sges: %d\n", sdp_sk(sk)->max_sge);
qp_init_attr.cap.max_send_sge = MIN(sdp_sk(sk)->max_sge, SDP_MAX_SEND_SGES);
sdp_dbg(sk, "Setting max send sge to: %d\n", qp_init_attr.cap.max_send_sge);
qp_init_attr.cap.max_recv_sge = MIN(sdp_sk(sk)->max_sge, SDP_MAX_RECV_SGES);
sdp_dbg(sk, "Setting max recv sge to: %d\n", qp_init_attr.cap.max_recv_sge);
sdp_sk(sk)->sdp_dev = ib_get_client_data(device, &sdp_client);
if (!sdp_sk(sk)->sdp_dev) {
sdp_warn(sk, "SDP not available on device %s\n", device->name);
rc = -ENODEV;
goto err_rx;
}
rc = sdp_rx_ring_create(sdp_sk(sk), device);
if (rc)
goto err_rx;
rc = sdp_tx_ring_create(sdp_sk(sk), device);
if (rc)
goto err_tx;
qp_init_attr.recv_cq = sdp_sk(sk)->rx_ring.cq;
qp_init_attr.send_cq = sdp_sk(sk)->tx_ring.cq;
rc = rdma_create_qp(id, sdp_sk(sk)->sdp_dev->pd, &qp_init_attr);
if (rc) {
sdp_warn(sk, "Unable to create QP: %d.\n", rc);
goto err_qp;
}
sdp_sk(sk)->qp = id->qp;
sdp_sk(sk)->ib_device = device;
sdp_sk(sk)->qp_active = 1;
sdp_sk(sk)->context.device = device;
sdp_sk(sk)->inline_thresh = qp_init_attr.cap.max_inline_data;
sdp_dbg(sk, "%s done\n", __func__);
return 0;
err_qp:
sdp_tx_ring_destroy(sdp_sk(sk));
err_tx:
sdp_rx_ring_destroy(sdp_sk(sk));
err_rx:
return rc;
}
static int sdp_get_max_send_frags(u32 buf_size)
{
return MIN(
/* +1 to conpensate on not aligned buffers */
(PAGE_ALIGN(buf_size) >> PAGE_SHIFT) + 1,
SDP_MAX_SEND_SGES - 1);
}
static int sdp_connect_handler(struct sock *sk, struct rdma_cm_id *id,
struct rdma_cm_event *event)
{
struct sockaddr_in *dst_addr;
struct sock *child;
const struct sdp_hh *h;
int rc = 0;
sdp_dbg(sk, "%s %p -> %p\n", __func__, sdp_sk(sk)->id, id);
h = event->param.conn.private_data;
SDP_DUMP_PACKET(sk, "RX", NULL, &h->bsdh);
if (h->ipv_cap & HH_IPV_MASK & ~(HH_IPV4 | HH_IPV6)) {
sdp_warn(sk, "Bad IPV field in SDP Hello header: 0x%x\n",
h->ipv_cap & HH_IPV_MASK);
return -EINVAL;
}
if (!h->max_adverts)
return -EINVAL;
#if (LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0))
child = sk_clone(sk, GFP_KERNEL);
#else
child = sk_clone_lock(sk, GFP_KERNEL);
#endif
if (!child)
return -ENOMEM;
sdp_init_sock(child);
dst_addr = (struct sockaddr_in *)&id->route.addr.dst_addr;
sdp_inet_dport(child) = dst_addr->sin_port;
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
if (inet6_sk(sk)) {
struct ipv6_pinfo *newnp;
newnp = inet_sk(child)->pinet6 = sdp_inet6_sk_generic(child);
memcpy(newnp, inet6_sk(sk), sizeof(struct ipv6_pinfo));
if ((h->ipv_cap & HH_IPV_MASK) == HH_IPV4) {
/* V6 mapped */
sdp_inet_daddr(child) = dst_addr->sin_addr.s_addr;
ipv6_addr_set(&child->sk_v6_daddr, 0, 0, htonl(0x0000FFFF),
h->src_addr.ip4.addr);
ipv6_addr_set(&child->sk_v6_rcv_saddr, 0, 0, htonl(0x0000FFFF),
h->dst_addr.ip4.addr);
ipv6_addr_copy(&child->sk_v6_rcv_saddr, &child->sk_v6_daddr);
} else if ((h->ipv_cap & HH_IPV_MASK) == HH_IPV6) {
struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *)dst_addr;
struct sockaddr_in6 *src_addr6 =
(struct sockaddr_in6 *)&id->route.addr.src_addr;
ipv6_addr_copy(&child->sk_v6_daddr, &dst_addr6->sin6_addr);
ipv6_addr_copy(&child->sk_v6_rcv_saddr, &src_addr6->sin6_addr);
ipv6_addr_copy(&newnp->saddr, &src_addr6->sin6_addr);
} else {
sdp_warn(child, "Bad IPV field: 0x%x\n", h->ipv_cap & HH_IPV_MASK);
}
sdp_inet_daddr(child) = sdp_inet_saddr(child) =
sdp_inet_rcv_saddr(child) = LOOPBACK4_IPV6;
} else
#endif
{
sdp_inet_daddr(child) = dst_addr->sin_addr.s_addr;
}
#ifdef SDP_SOCK_HISTORY
sdp_ssk_hist_rename(sk);
#endif
__sock_put(child, SOCK_REF_CLONE);
down_read(&device_removal_lock);
rc = sdp_init_qp(child, id);
if (rc) {
bh_unlock_sock(child);
up_read(&device_removal_lock);
sdp_sk(child)->destructed_already = 1;
#ifdef SDP_SOCK_HISTORY
sdp_ssk_hist_close(child);
#endif
sk_free(child);
return rc;
}
sdp_sk(child)->max_bufs = ntohs(h->bsdh.bufs);
atomic_set(&sdp_sk(child)->tx_ring.credits, sdp_sk(child)->max_bufs);
sdp_sk(child)->min_bufs = tx_credits(sdp_sk(child)) / 4;
sdp_sk(child)->xmit_size_goal = ntohl(h->localrcvsz) -
sizeof(struct sdp_bsdh);
sdp_sk(child)->send_frags = sdp_get_max_send_frags(sdp_sk(child)->xmit_size_goal);
sdp_init_buffers(sdp_sk(child), rcvbuf_initial_size);
id->context = child;
sdp_sk(child)->id = id;
list_add_tail(&sdp_sk(child)->backlog_queue,
&sdp_sk(sk)->backlog_queue);
sdp_sk(child)->parent = sk;
bh_unlock_sock(child);
sdp_add_sock(sdp_sk(child));
up_read(&device_removal_lock);
sdp_exch_state(child, TCPF_LISTEN | TCPF_CLOSE, TCP_SYN_RECV);
/* child->sk_write_space(child); */
/* child->sk_data_ready(child, 0); */
sk->sk_data_ready(sk);
return 0;
}
static int sdp_response_handler(struct sock *sk, struct rdma_cm_id *id,
struct rdma_cm_event *event)
{
const struct sdp_hah *h;
struct sockaddr_in *dst_addr;
sdp_dbg(sk, "%s\n", __func__);
sdp_exch_state(sk, TCPF_SYN_SENT, TCP_ESTABLISHED);
sdp_set_default_moderation(sdp_sk(sk));
if (sock_flag(sk, SOCK_KEEPOPEN))
sdp_start_keepalive_timer(sk);
if (sock_flag(sk, SOCK_DEAD))
return 0;
h = event->param.conn.private_data;
SDP_DUMP_PACKET(sk, "RX", NULL, &h->bsdh);
sdp_sk(sk)->max_bufs = ntohs(h->bsdh.bufs);
atomic_set(&sdp_sk(sk)->tx_ring.credits, sdp_sk(sk)->max_bufs);
sdp_sk(sk)->min_bufs = tx_credits(sdp_sk(sk)) / 4;
sdp_sk(sk)->xmit_size_goal =
ntohl(h->actrcvsz) - sizeof(struct sdp_bsdh);
sdp_sk(sk)->send_frags = sdp_get_max_send_frags(sdp_sk(sk)->xmit_size_goal);
sdp_sk(sk)->xmit_size_goal = MIN(sdp_sk(sk)->xmit_size_goal,
sdp_sk(sk)->send_frags * PAGE_SIZE);
sdp_sk(sk)->poll_cq = 1;
sk->sk_state_change(sk);
sk_wake_async(sk, 0, POLL_OUT);
dst_addr = (struct sockaddr_in *)&id->route.addr.dst_addr;
sdp_inet_dport(sk) = dst_addr->sin_port;
sdp_inet_daddr(sk) = dst_addr->sin_addr.s_addr;
#ifdef SDP_SOCK_HISTORY
sdp_ssk_hist_rename(sk);
#endif
return 0;
}
static int sdp_connected_handler(struct sock *sk)
{
struct sock *parent;
sdp_dbg(sk, "%s\n", __func__);
parent = sdp_sk(sk)->parent;
BUG_ON(!parent);
sdp_exch_state(sk, TCPF_SYN_RECV, TCP_ESTABLISHED);
#ifdef SDP_SOCK_HISTORY
sdp_ssk_hist_rename(sk);
#endif
sdp_set_default_moderation(sdp_sk(sk));
if (sock_flag(sk, SOCK_KEEPOPEN))
sdp_start_keepalive_timer(sk);
if (sock_flag(sk, SOCK_DEAD))
return 0;
lock_sock(parent);
if (!sdp_sk(parent)->id) { /* TODO: look at SOCK_DEAD? */
sdp_dbg(sk, "parent is going away.\n");
goto done;
}
sk_acceptq_added(parent);
sdp_dbg(parent, "%s child connection established\n", __func__);
list_del_init(&sdp_sk(sk)->backlog_queue);
list_add_tail(&sdp_sk(sk)->accept_queue,
&sdp_sk(parent)->accept_queue);
parent->sk_state_change(parent);
sk_wake_async(parent, 0, POLL_OUT);
done:
release_sock(parent);
return 0;
}
static int sdp_disconnected_handler(struct sock *sk)
{
struct sdp_sock *ssk = sdp_sk(sk);
sdp_dbg(sk, "%s\n", __func__);
if (ssk->tx_ring.cq)
if (sdp_xmit_poll(ssk, 1))
sdp_post_sends(ssk, 0);
if (sk->sk_state == TCP_SYN_RECV) {
sdp_connected_handler(sk);
if (rcv_nxt(ssk))
return 0;
}
return -ECONNRESET;
}
int sdp_cma_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
{
struct rdma_conn_param conn_param;
struct sock *parent = NULL;
struct sock *child = NULL;
struct sock *sk;
struct sdp_hah hah;
struct sdp_hh hh;
int rc = 0, rc2;
sk = id->context;
if (!sk) {
sdp_dbg(NULL, "cm_id is being torn down, event %s\n",
rdma_cm_event_str(event->event));
return event->event == RDMA_CM_EVENT_CONNECT_REQUEST ?
-EINVAL : 0;
}
sdp_add_to_history(sk, rdma_cm_event_str(event->event));
lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
sdp_dbg(sk, "event: %s\n", rdma_cm_event_str(event->event));
if (!sdp_sk(sk)->id) {
sdp_dbg(sk, "socket is being torn down\n");
rc = event->event == RDMA_CM_EVENT_CONNECT_REQUEST ?
-EINVAL : 0;
release_sock(sk);
return rc;
}
switch (event->event) {
case RDMA_CM_EVENT_ADDR_RESOLVED:
if (sdp_link_layer_ib_only &&
rdma_node_get_transport(id->device->node_type) ==
RDMA_TRANSPORT_IB &&
rdma_port_get_link_layer(id->device, id->port_num) !=
IB_LINK_LAYER_INFINIBAND) {
sdp_dbg(sk, "Link layer is: %d. Only IB link layer "
"is allowed\n",
rdma_port_get_link_layer(id->device,
id->port_num));
rc = -ENETUNREACH;
break;
}
rc = rdma_resolve_route(id, SDP_ROUTE_TIMEOUT);
break;
case RDMA_CM_EVENT_ADDR_ERROR:
rc = -ENETUNREACH;
break;
case RDMA_CM_EVENT_ROUTE_RESOLVED:
rc = sdp_init_qp(sk, id);
if (rc)
break;
memset(&hh, 0, sizeof hh);
hh.bsdh.mid = SDP_MID_HELLO;
hh.bsdh.len = htonl(sizeof(struct sdp_hh));
hh.max_adverts = 1;
hh.majv_minv = SDP_MAJV_MINV;
sdp_init_buffers(sdp_sk(sk), rcvbuf_initial_size);
hh.bsdh.bufs = htons(rx_ring_posted(sdp_sk(sk)));
atomic_set(&sdp_sk(sk)->remote_credits,
rx_ring_posted(sdp_sk(sk)));
hh.localrcvsz = hh.desremrcvsz = htonl(sdp_sk(sk)->recv_frags *
PAGE_SIZE + sizeof(struct sdp_bsdh));
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
if (inet6_sk(sk)) {
struct sockaddr *src_addr = (struct sockaddr *)&id->route.addr.src_addr;
struct sockaddr_in *addr4 = (struct sockaddr_in *)src_addr;
struct sockaddr_in6 *addr6 = (struct sockaddr_in6 *)src_addr;
if (src_addr->sa_family == AF_INET) {
/* IPv4 over IPv6 */
ipv6_addr_set(&sk->sk_v6_rcv_saddr, 0, 0, htonl(0xFFFF),
addr4->sin_addr.s_addr);
} else {
sk->sk_v6_rcv_saddr = addr6->sin6_addr;
}
inet6_sk(sk)->saddr = sk->sk_v6_rcv_saddr;
}
else
#endif
{
sdp_inet_saddr(sk) = sdp_inet_rcv_saddr(sk) =
((struct sockaddr_in *)&id->route.addr.src_addr)->sin_addr.s_addr;
}
memset(&conn_param, 0, sizeof conn_param);
conn_param.private_data_len = sizeof hh;
conn_param.private_data = &hh;
conn_param.responder_resources = 4 /* TODO */;
conn_param.initiator_depth = 4 /* TODO */;
conn_param.retry_count = sdp_retry_count;
SDP_DUMP_PACKET(sk, "TX", NULL, &hh.bsdh);
if (sdp_apm_enable) {
rc = rdma_enable_apm(id, RDMA_ALT_PATH_BEST);
if (rc)
sdp_warn(sk, "APM couldn't be enabled for active side: %d\n", rc);
}
rc = rdma_connect(id, &conn_param);
break;
case RDMA_CM_EVENT_ALT_ROUTE_RESOLVED:
sdp_dbg(sk, "alt route was resolved slid=%d, dlid=%d\n",
id->route.path_rec[1].slid, id->route.path_rec[1].dlid);
break;
case RDMA_CM_EVENT_ALT_PATH_LOADED:
sdp_dbg(sk, "alt route path loaded\n");
break;
case RDMA_CM_EVENT_ALT_ROUTE_ERROR:
sdp_warn(sk, "alt route resolve error\n");
break;
case RDMA_CM_EVENT_ROUTE_ERROR:
rc = -ETIMEDOUT;
break;
case RDMA_CM_EVENT_CONNECT_REQUEST:
rc = sdp_connect_handler(sk, id, event);
if (rc) {
sdp_dbg(sk, "Destroying qp\n");
rdma_reject(id, NULL, 0);
break;
}
child = id->context;
atomic_set(&sdp_sk(child)->remote_credits,
rx_ring_posted(sdp_sk(child)));
memset(&hah, 0, sizeof hah);
hah.bsdh.mid = SDP_MID_HELLO_ACK;
hah.bsdh.bufs = htons(rx_ring_posted(sdp_sk(child)));
hah.bsdh.len = htonl(sizeof(struct sdp_hah));
hah.majv_minv = SDP_MAJV_MINV;
hah.ext_max_adverts = 1; /* Doesn't seem to be mandated by spec,
but just in case */
hah.actrcvsz = htonl(sdp_sk(child)->recv_frags * PAGE_SIZE +
sizeof(struct sdp_bsdh));
memset(&conn_param, 0, sizeof conn_param);
conn_param.private_data_len = sizeof hah;
conn_param.private_data = &hah;
conn_param.responder_resources = 4 /* TODO */;
conn_param.initiator_depth = 4 /* TODO */;
conn_param.retry_count = sdp_retry_count;
SDP_DUMP_PACKET(sk, "TX", NULL, &hah.bsdh);
rc = rdma_accept(id, &conn_param);
if (rc) {
sdp_sk(child)->id = NULL;
id->qp = NULL;
id->context = NULL;
parent = sdp_sk(child)->parent; /* TODO: hold ? */
} else if (sdp_apm_enable) {
rc2 = rdma_enable_apm(id, RDMA_ALT_PATH_BEST);
if (rc2)
sdp_warn(sk, "APM couldn't be enabled for passive side: %d\n", rc2);
}
break;
case RDMA_CM_EVENT_CONNECT_RESPONSE:
rc = sdp_response_handler(sk, id, event);
if (rc) {
sdp_dbg(sk, "Destroying qp\n");
rdma_reject(id, NULL, 0);
} else {
rc = rdma_accept(id, NULL);
if (!rc && sdp_apm_enable) {
rc2 = rdma_enable_apm(id, RDMA_ALT_PATH_BEST);
if (rc2)
sdp_warn(sk, "APM couldn't be enabled for passive side:%d \n", rc2);
}
}
break;
case RDMA_CM_EVENT_CONNECT_ERROR:
rc = -ETIMEDOUT;
break;
case RDMA_CM_EVENT_UNREACHABLE:
rc = -ENETUNREACH;
break;
case RDMA_CM_EVENT_REJECTED:
rc = -ECONNREFUSED;
break;
case RDMA_CM_EVENT_ESTABLISHED:
sdp_inet_saddr(sk) = sdp_inet_rcv_saddr(sk) =
((struct sockaddr_in *)&id->route.addr.src_addr)->sin_addr.s_addr;
rc = sdp_connected_handler(sk);
break;
case RDMA_CM_EVENT_DISCONNECTED: /* This means DREQ/DREP received */
if (sk->sk_state == TCP_LAST_ACK) {
sdp_cancel_dreq_wait_timeout(sdp_sk(sk));
sdp_exch_state(sk, TCPF_LAST_ACK, TCP_TIME_WAIT);
sdp_dbg(sk, "%s: waiting for Infiniband tear down\n",
__func__);
}
sdp_sk(sk)->qp_active = 0;
rdma_disconnect(id);
if (sk->sk_state != TCP_TIME_WAIT) {
if (sk->sk_state == TCP_CLOSE_WAIT) {
sdp_dbg(sk, "IB teardown while in "
"TCP_CLOSE_WAIT taking reference to "
"let close() finish the work\n");
sock_hold(sk, SOCK_REF_CMA);
sdp_start_cma_timewait_timeout(sdp_sk(sk),
SDP_CMA_TIMEWAIT_TIMEOUT);
}
sdp_set_error(sk, -EPIPE);
rc = sdp_disconnected_handler(sk);
}
break;
case RDMA_CM_EVENT_TIMEWAIT_EXIT:
rc = sdp_disconnected_handler(sk);
break;
case RDMA_CM_EVENT_DEVICE_REMOVAL:
rc = -ENETRESET;
break;
case RDMA_CM_EVENT_ADDR_CHANGE:
sdp_dbg(sk, "Got Address change event\n");
rc = 0;
break;
default:
printk(KERN_ERR "SDP: Unexpected CMA event: %d\n",
event->event);
rc = -ECONNABORTED;
break;
}
sdp_dbg(sk, "event: %s handled\n", rdma_cm_event_str(event->event));
if (rc && sdp_sk(sk)->id == id) {
child = sk;
sdp_sk(sk)->id = NULL;
id->qp = NULL;
id->context = NULL;
parent = sdp_sk(sk)->parent;
sdp_reset_sk(sk, rc);
}
release_sock(sk);
sdp_dbg(sk, "event: %s done. status %d\n",
rdma_cm_event_str(event->event), rc);
if (parent) {
lock_sock(parent);
if (!sdp_sk(parent)->id) { /* TODO: look at SOCK_DEAD? */
sdp_dbg(sk, "parent is going away.\n");
child = NULL;
goto done;
}
if (!list_empty(&sdp_sk(child)->backlog_queue))
list_del_init(&sdp_sk(child)->backlog_queue);
else
child = NULL;
done:
release_sock(parent);
if (child)
sdp_common_release(child);
}
return rc;
}
static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
{
struct iser_device *device;
struct ib_qp_init_attr init_attr;
int ret;
struct ib_fmr_pool_param params;
BUG_ON(ib_conn->device == NULL);
device = ib_conn->device;
ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
(sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
GFP_KERNEL);
if (!ib_conn->page_vec) {
ret = -ENOMEM;
goto alloc_err;
}
ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
params.page_shift = SHIFT_4K;
params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
params.cache = 0;
params.flush_function = NULL;
params.access = (IB_ACCESS_LOCAL_WRITE |
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_REMOTE_READ);
ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
if (IS_ERR(ib_conn->fmr_pool)) {
ret = PTR_ERR(ib_conn->fmr_pool);
goto fmr_pool_err;
}
memset(&init_attr, 0, sizeof init_attr);
init_attr.event_handler = iser_qp_event_callback;
init_attr.qp_context = (void *)ib_conn;
init_attr.send_cq = device->cq;
init_attr.recv_cq = device->cq;
init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
init_attr.cap.max_send_sge = MAX_REGD_BUF_VECTOR_LEN;
init_attr.cap.max_recv_sge = 2;
init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
init_attr.qp_type = IB_QPT_RC;
ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
if (ret)
goto qp_err;
ib_conn->qp = ib_conn->cma_id->qp;
iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
ib_conn, ib_conn->cma_id,
ib_conn->fmr_pool, ib_conn->cma_id->qp);
return ret;
qp_err:
(void)ib_destroy_fmr_pool(ib_conn->fmr_pool);
fmr_pool_err:
kfree(ib_conn->page_vec);
alloc_err:
iser_err("unable to alloc mem or create resource, err %d\n", ret);
return ret;
}
/**
* trans_create_rdma - Transport method for creating atransport instance
* @client: client instance
* @addr: IP address string
* @args: Mount options string
*/
static int
rdma_create_trans(struct p9_client *client, const char *addr, char *args)
{
int err;
struct p9_rdma_opts opts;
struct p9_trans_rdma *rdma;
struct rdma_conn_param conn_param;
struct ib_qp_init_attr qp_attr;
struct ib_device_attr devattr;
/* Parse the transport specific mount options */
err = parse_opts(args, &opts);
if (err < 0)
return err;
/* Create and initialize the RDMA transport structure */
rdma = alloc_rdma(&opts);
if (!rdma)
return -ENOMEM;
/* Create the RDMA CM ID */
rdma->cm_id = rdma_create_id(p9_cm_event_handler, client, RDMA_PS_TCP);
if (IS_ERR(rdma->cm_id))
goto error;
/* Associate the client with the transport */
client->trans = rdma;
/* Resolve the server's address */
rdma->addr.sin_family = AF_INET;
rdma->addr.sin_addr.s_addr = in_aton(addr);
rdma->addr.sin_port = htons(opts.port);
err = rdma_resolve_addr(rdma->cm_id, NULL,
(struct sockaddr *)&rdma->addr,
rdma->timeout);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
goto error;
/* Resolve the route to the server */
err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
goto error;
/* Query the device attributes */
err = ib_query_device(rdma->cm_id->device, &devattr);
if (err)
goto error;
/* Create the Completion Queue */
rdma->cq = ib_create_cq(rdma->cm_id->device, cq_comp_handler,
cq_event_handler, client,
opts.sq_depth + opts.rq_depth + 1, 0);
if (IS_ERR(rdma->cq))
goto error;
ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
/* Create the Protection Domain */
rdma->pd = ib_alloc_pd(rdma->cm_id->device);
if (IS_ERR(rdma->pd))
goto error;
/* Cache the DMA lkey in the transport */
rdma->dma_mr = NULL;
if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
rdma->lkey = rdma->cm_id->device->local_dma_lkey;
else {
rdma->dma_mr = ib_get_dma_mr(rdma->pd, IB_ACCESS_LOCAL_WRITE);
if (IS_ERR(rdma->dma_mr))
goto error;
rdma->lkey = rdma->dma_mr->lkey;
}
/* Create the Queue Pair */
memset(&qp_attr, 0, sizeof qp_attr);
qp_attr.event_handler = qp_event_handler;
qp_attr.qp_context = client;
qp_attr.cap.max_send_wr = opts.sq_depth;
qp_attr.cap.max_recv_wr = opts.rq_depth;
qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
qp_attr.qp_type = IB_QPT_RC;
qp_attr.send_cq = rdma->cq;
qp_attr.recv_cq = rdma->cq;
err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
if (err)
goto error;
rdma->qp = rdma->cm_id->qp;
/* Request a connection */
memset(&conn_param, 0, sizeof(conn_param));
conn_param.private_data = NULL;
conn_param.private_data_len = 0;
conn_param.responder_resources = P9_RDMA_IRD;
conn_param.initiator_depth = P9_RDMA_ORD;
err = rdma_connect(rdma->cm_id, &conn_param);
if (err)
goto error;
err = wait_for_completion_interruptible(&rdma->cm_done);
if (err || (rdma->state != P9_RDMA_CONNECTED))
goto error;
client->status = Connected;
return 0;
error:
rdma_destroy_trans(rdma);
return -ENOTCONN;
}
int main(int argc, char *argv[]) {
struct pdata rep_pdata;
struct rdma_event_channel *cm_channel;
struct rdma_cm_id *listen_id;
struct rdma_cm_id *cm_id;
struct rdma_cm_event *event;
struct rdma_conn_param conn_param = { };
struct ibv_pd *pd;
struct ibv_comp_channel *comp_chan;
struct ibv_cq *cq;
struct ibv_cq *evt_cq;
struct ibv_mr *mr;
struct ibv_qp_init_attr qp_attr = { };
struct ibv_sge sge;
struct ibv_send_wr send_wr = { };
struct ibv_send_wr *bad_send_wr;
struct ibv_recv_wr recv_wr = { };
struct ibv_recv_wr *bad_recv_wr;
struct ibv_wc wc;
void *cq_context;
struct sockaddr_in sin;
uint32_t *buf;
int err;
/* Set up RDMA CM structures */
cm_channel = rdma_create_event_channel();
if (!cm_channel)
return 1;
err = rdma_create_id(cm_channel, &listen_id, NULL, RDMA_PS_TCP);
if (err)
return err;
sin.sin_family = AF_INET;
sin.sin_port = htons(20079);
sin.sin_addr.s_addr = INADDR_ANY;
/* Bind to local port and listen for connection request */
err = rdma_bind_addr(listen_id, (struct sockaddr *) &sin);
if (err)
return 1;
err = rdma_listen(listen_id, 1);
if (err)
return 1;
err = rdma_get_cm_event(cm_channel, &event);
if (err)
return err;
printf("after get_cm_event\n");
if (event->event != RDMA_CM_EVENT_CONNECT_REQUEST)
return 1;
cm_id = event->id;
rdma_ack_cm_event(event);
/* Create verbs objects now that we know which device to use */
pd = ibv_alloc_pd(cm_id->verbs);
if (!pd)
return 1;
comp_chan = ibv_create_comp_channel(cm_id->verbs);
if (!comp_chan)
return 1;
cq = ibv_create_cq(cm_id->verbs, 2, NULL, comp_chan, 0);
if (!cq)
return 1;
if (ibv_req_notify_cq(cq, 0))
return 1;
buf = calloc(2, sizeof(uint32_t));
if (!buf)
return 1;
mr = ibv_reg_mr(pd, buf, 2 * sizeof(uint32_t),
IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_READ
| IBV_ACCESS_REMOTE_WRITE);
if (!mr)
return 1;
qp_attr.cap.max_send_wr = 1;
qp_attr.cap.max_send_sge = 1;
qp_attr.cap.max_recv_wr = 1;
qp_attr.cap.max_recv_sge = 1;
qp_attr.send_cq = cq;
qp_attr.recv_cq = cq;
qp_attr.qp_type = IBV_QPT_RC;
err = rdma_create_qp(cm_id, pd, &qp_attr);
if (err)
return err;
/* Post receive before accepting connection */
sge.addr = (uintptr_t) buf + sizeof(uint32_t);
sge.length = sizeof(uint32_t);
sge.lkey = mr->lkey;
recv_wr.sg_list = &sge;
recv_wr.num_sge = 1;
if (ibv_post_recv(cm_id->qp, &recv_wr, &bad_recv_wr))
return 1;
rep_pdata.buf_va = htonll((uintptr_t) buf);
rep_pdata.buf_rkey = htonl(mr->rkey);
conn_param.responder_resources = 1;
conn_param.private_data = &rep_pdata;
conn_param.private_data_len = sizeof rep_pdata;
/* Accept connection */
printf("before accept\n");
err = rdma_accept(cm_id, &conn_param);
if (err)
return 1;
printf("after accept\n");
err = rdma_get_cm_event(cm_channel, &event);
if (err)
return err;
if (event->event != RDMA_CM_EVENT_ESTABLISHED)
return 1;
rdma_ack_cm_event(event);
/* Wait for receive completion */
if (ibv_get_cq_event(comp_chan, &evt_cq, &cq_context))
return 1;
if (ibv_req_notify_cq(cq, 0))
return 1;
if (ibv_poll_cq(cq, 1, &wc) < 1)
return 1;
if (wc.status != IBV_WC_SUCCESS)
return 1;
/* Add two integers and send reply back */
buf[0] = htonl(ntohl(buf[0]) + ntohl(buf[1]));
sge.addr = (uintptr_t) buf;
sge.length = sizeof(uint32_t);
sge.lkey = mr->lkey;
send_wr.opcode = IBV_WR_SEND;
send_wr.send_flags = IBV_SEND_SIGNALED;
send_wr.sg_list = &sge;
send_wr.num_sge = 1;
if (ibv_post_send(cm_id->qp, &send_wr, &bad_send_wr))
return 1;
/* Wait for send completion */
if (ibv_get_cq_event(comp_chan, &evt_cq, &cq_context))
return 1;
if (ibv_poll_cq(cq, 1, &wc) < 1)
return 1;
if (wc.status != IBV_WC_SUCCESS)
return 1;
printf("before ack cq 2\n");
ibv_ack_cq_events(cq, 2);
return 0;
}
