*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lnet/klnds/o2iblnd/o2iblnd_cb.c
*
{
struct kib_tx *tx;
- while (!list_empty(txlist)) {
- tx = list_entry(txlist->next, struct kib_tx, tx_list);
-
+ while ((tx = list_first_entry_or_null(txlist,
+ struct kib_tx,
+ tx_list)) != NULL) {
list_del(&tx->tx_list);
/* complete now */
tx->tx_waiting = 0;
tx->tx_status = status;
- tx->tx_hstatus = hstatus;
+ if (hstatus != LNET_MSG_STATUS_OK)
+ tx->tx_hstatus = hstatus;
kiblnd_tx_done(tx);
}
}
* own this rx (and rx::rx_conn) anymore, LU-5678.
*/
kiblnd_conn_addref(conn);
+#ifdef HAVE_IB_POST_SEND_RECV_CONST
+ rc = ib_post_recv(conn->ibc_cmid->qp, &rx->rx_wrq,
+ (const struct ib_recv_wr **)&bad_wrq);
+#else
rc = ib_post_recv(conn->ibc_cmid->qp, &rx->rx_wrq, &bad_wrq);
+#endif
if (unlikely(rc != 0)) {
CERROR("Can't post rx for %s: %d, bad_wrq: %p\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid), rc, bad_wrq);
static struct kib_tx *
kiblnd_find_waiting_tx_locked(struct kib_conn *conn, int txtype, u64 cookie)
{
- struct list_head *tmp;
-
- list_for_each(tmp, &conn->ibc_active_txs) {
- struct kib_tx *tx = list_entry(tmp, struct kib_tx, tx_list);
+ struct kib_tx *tx;
+ list_for_each_entry(tx, &conn->ibc_active_txs, tx_list) {
LASSERT(!tx->tx_queued);
LASSERT(tx->tx_sending != 0 || tx->tx_waiting);
int rc;
int err = -EIO;
- LASSERT (net != NULL);
- LASSERT (rx->rx_nob < 0); /* was posted */
- rx->rx_nob = 0; /* isn't now */
+ LASSERT(net);
+ LASSERT(rx->rx_nob < 0); /* was posted */
+ rx->rx_nob = 0; /* isn't now */
- if (conn->ibc_state > IBLND_CONN_ESTABLISHED)
- goto ignore;
+ if (conn->ibc_state > IBLND_CONN_ESTABLISHED)
+ goto ignore;
- if (status != IB_WC_SUCCESS) {
- CNETERR("Rx from %s failed: %d\n",
- libcfs_nid2str(conn->ibc_peer->ibp_nid), status);
- goto failed;
- }
+ if (status != IB_WC_SUCCESS) {
+ CNETERR("Rx from %s failed: %d\n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid), status);
+ goto failed;
+ }
- LASSERT (nob >= 0);
- rx->rx_nob = nob;
+ LASSERT(nob >= 0);
+ rx->rx_nob = nob;
- rc = kiblnd_unpack_msg(msg, rx->rx_nob);
- if (rc != 0) {
- CERROR ("Error %d unpacking rx from %s\n",
- rc, libcfs_nid2str(conn->ibc_peer->ibp_nid));
- goto failed;
- }
+ rc = kiblnd_unpack_msg(msg, rx->rx_nob);
+ if (rc != 0) {
+ CERROR("Error %d unpacking rx from %s\n",
+ rc, libcfs_nid2str(conn->ibc_peer->ibp_nid));
+ goto failed;
+ }
- if (msg->ibm_srcnid != conn->ibc_peer->ibp_nid ||
- msg->ibm_dstnid != ni->ni_nid ||
- msg->ibm_srcstamp != conn->ibc_incarnation ||
- msg->ibm_dststamp != net->ibn_incarnation) {
- CERROR ("Stale rx from %s\n",
- libcfs_nid2str(conn->ibc_peer->ibp_nid));
- err = -ESTALE;
- goto failed;
- }
+ if (msg->ibm_srcnid != conn->ibc_peer->ibp_nid ||
+ msg->ibm_dstnid != lnet_nid_to_nid4(&ni->ni_nid) ||
+ msg->ibm_srcstamp != conn->ibc_incarnation ||
+ msg->ibm_dststamp != net->ibn_incarnation) {
+ CERROR("Stale rx from %s\n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid));
+ err = -ESTALE;
+ goto failed;
+ }
- /* set time last known alive */
- kiblnd_peer_alive(conn->ibc_peer);
+ /* set time last known alive */
+ kiblnd_peer_alive(conn->ibc_peer);
- /* racing with connection establishment/teardown! */
+ /* racing with connection establishment/teardown! */
- if (conn->ibc_state < IBLND_CONN_ESTABLISHED) {
+ if (conn->ibc_state < IBLND_CONN_ESTABLISHED) {
rwlock_t *g_lock = &kiblnd_data.kib_global_lock;
unsigned long flags;
return;
}
write_unlock_irqrestore(g_lock, flags);
- }
- kiblnd_handle_rx(rx);
- return;
+ }
+ kiblnd_handle_rx(rx);
+ return;
- failed:
- CDEBUG(D_NET, "rx %p conn %p\n", rx, conn);
- kiblnd_close_conn(conn, err);
- ignore:
- kiblnd_drop_rx(rx); /* Don't re-post rx. */
+failed:
+ CDEBUG(D_NET, "rx %p conn %p\n", rx, conn);
+ kiblnd_close_conn(conn, err);
+ignore:
+ kiblnd_drop_rx(rx); /* Don't re-post rx. */
}
static int
return -EPROTONOSUPPORT;
}
+#ifdef HAVE_FMR_POOL_API
/*
* FMR does not support gaps but the tx has gaps then
* we should make sure that the number of fragments we'll be sending
return -EFBIG;
}
}
+#endif
fps = net->ibn_fmr_ps[cpt];
rc = kiblnd_fmr_pool_map(fps, tx, rd, nob, 0, &tx->tx_fmr);
if (rc != 0) {
- CERROR("Can't map %u pages: %d\n", nob, rc);
+ CERROR("Can't map %u bytes (%u/%u)s: %d\n", nob,
+ tx->tx_nfrags, rd->rd_nfrags, rc);
return rc;
}
* for FastReg or FMR with no gaps we can accumulate all
* the fragments in one FastReg or FMR fragment.
*/
- if (((dev->ibd_dev_caps & IBLND_DEV_CAPS_FMR_ENABLED) && !tx->tx_gaps) ||
+ if (
+#ifdef HAVE_FMR_POOL_API
+ ((dev->ibd_dev_caps & IBLND_DEV_CAPS_FMR_ENABLED)
+ && !tx->tx_gaps) ||
+#endif
(dev->ibd_dev_caps & IBLND_DEV_CAPS_FASTREG_ENABLED)) {
/* FMR requires zero based address */
+#ifdef HAVE_FMR_POOL_API
if (dev->ibd_dev_caps & IBLND_DEV_CAPS_FMR_ENABLED)
rd->rd_frags[0].rf_addr &= ~hdev->ibh_page_mask;
+#endif
rd->rd_frags[0].rf_nob = nob;
rd->rd_nfrags = 1;
} else {
static void
kiblnd_unmap_tx(struct kib_tx *tx)
{
- if (tx->tx_fmr.fmr_pfmr || tx->tx_fmr.fmr_frd)
+ if (
+#ifdef HAVE_FMR_POOL_API
+ tx->tx_fmr.fmr_pfmr ||
+#endif
+ tx->tx_fmr.fmr_frd)
kiblnd_fmr_pool_unmap(&tx->tx_fmr, tx->tx_status);
if (tx->tx_nfrags != 0) {
- kiblnd_dma_unmap_sg(tx->tx_pool->tpo_hdev->ibh_ibdev,
+ kiblnd_dma_unmap_sg(tx->tx_pool->tpo_hdev,
tx->tx_frags, tx->tx_nfrags, tx->tx_dmadir);
tx->tx_nfrags = 0;
}
* dead in the water and fail the operation.
*/
if (tunables->lnd_map_on_demand &&
- (net->ibn_dev->ibd_dev_caps & IBLND_DEV_CAPS_FASTREG_ENABLED ||
- net->ibn_dev->ibd_dev_caps & IBLND_DEV_CAPS_FMR_ENABLED))
+ (net->ibn_dev->ibd_dev_caps & IBLND_DEV_CAPS_FASTREG_ENABLED
+#ifdef HAVE_FMR_POOL_API
+ || net->ibn_dev->ibd_dev_caps & IBLND_DEV_CAPS_FMR_ENABLED
+#endif
+ ))
return NULL;
/*
tx->tx_dmadir = (rd != tx->tx_rd) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
tx->tx_nfrags = nfrags;
- rd->rd_nfrags = kiblnd_dma_map_sg(hdev->ibh_ibdev, tx->tx_frags,
+ rd->rd_nfrags = kiblnd_dma_map_sg(hdev, tx->tx_frags,
tx->tx_nfrags, tx->tx_dmadir);
for (i = 0, nob = 0; i < rd->rd_nfrags; i++) {
return -EINVAL;
}
-static int kiblnd_setup_rd_iov(struct lnet_ni *ni, struct kib_tx *tx,
- struct kib_rdma_desc *rd, unsigned int niov,
- struct kvec *iov, int offset, int nob)
-{
- struct kib_net *net = ni->ni_data;
- struct page *page;
- struct scatterlist *sg;
- unsigned long vaddr;
- int fragnob;
- int page_offset;
- unsigned int max_niov;
-
- LASSERT (nob > 0);
- LASSERT (niov > 0);
- LASSERT (net != NULL);
-
- while (offset >= iov->iov_len) {
- offset -= iov->iov_len;
- niov--;
- iov++;
- LASSERT (niov > 0);
- }
-
- max_niov = niov;
-
- sg = tx->tx_frags;
- do {
- LASSERT(niov > 0);
-
- vaddr = ((unsigned long)iov->iov_base) + offset;
- page_offset = vaddr & (PAGE_SIZE - 1);
- page = lnet_kvaddr_to_page(vaddr);
- if (page == NULL) {
- CERROR("Can't find page\n");
- return -EFAULT;
- }
-
- fragnob = min((int)(iov->iov_len - offset), nob);
- fragnob = min(fragnob, (int)PAGE_SIZE - page_offset);
-
- /*
- * We're allowed to start at a non-aligned page offset in
- * the first fragment and end at a non-aligned page offset
- * in the last fragment.
- */
- if ((fragnob < (int)PAGE_SIZE - page_offset) &&
- (niov < max_niov) && nob > fragnob) {
- CDEBUG(D_NET, "fragnob %d < available page %d: with"
- " remaining %d iovs with %d nob left\n",
- fragnob, (int)PAGE_SIZE - page_offset, niov,
- nob);
- tx->tx_gaps = true;
- }
-
- sg_set_page(sg, page, fragnob, page_offset);
- sg = sg_next(sg);
- if (!sg) {
- CERROR("lacking enough sg entries to map tx\n");
- return -EFAULT;
- }
-
- if (offset + fragnob < iov->iov_len) {
- offset += fragnob;
- } else {
- offset = 0;
- iov++;
- niov--;
- }
- nob -= fragnob;
- } while (nob > 0);
-
- return kiblnd_map_tx(ni, tx, rd, sg - tx->tx_frags);
-}
-
static int kiblnd_setup_rd_kiov(struct lnet_ni *ni, struct kib_tx *tx,
struct kib_rdma_desc *rd, int nkiov,
- lnet_kiov_t *kiov, int offset, int nob)
+ struct bio_vec *kiov, int offset, int nob)
{
struct kib_net *net = ni->ni_data;
struct scatterlist *sg;
- int fragnob;
- int max_nkiov;
+ int fragnob;
+ int max_nkiov;
+ int sg_count = 0;
CDEBUG(D_NET, "niov %d offset %d nob %d\n", nkiov, offset, nob);
LASSERT(nkiov > 0);
LASSERT(net != NULL);
- while (offset >= kiov->kiov_len) {
- offset -= kiov->kiov_len;
+ while (offset >= kiov->bv_len) {
+ offset -= kiov->bv_len;
nkiov--;
kiov++;
LASSERT(nkiov > 0);
do {
LASSERT(nkiov > 0);
- fragnob = min((int)(kiov->kiov_len - offset), nob);
+ if (!sg) {
+ CERROR("lacking enough sg entries to map tx\n");
+ return -EFAULT;
+ }
+ sg_count++;
+
+ fragnob = min((int)(kiov->bv_len - offset), nob);
/*
* We're allowed to start at a non-aligned page offset in
* the first fragment and end at a non-aligned page offset
* in the last fragment.
*/
- if ((fragnob < (int)(kiov->kiov_len - offset)) &&
+ if ((fragnob < (int)(kiov->bv_len - offset)) &&
nkiov < max_nkiov && nob > fragnob) {
CDEBUG(D_NET, "fragnob %d < available page %d: with"
" remaining %d kiovs with %d nob left\n",
- fragnob, (int)(kiov->kiov_len - offset),
+ fragnob, (int)(kiov->bv_len - offset),
nkiov, nob);
tx->tx_gaps = true;
}
- sg_set_page(sg, kiov->kiov_page, fragnob,
- kiov->kiov_offset + offset);
+ sg_set_page(sg, kiov->bv_page, fragnob,
+ kiov->bv_offset + offset);
sg = sg_next(sg);
- if (!sg) {
- CERROR("lacking enough sg entries to map tx\n");
- return -EFAULT;
- }
offset = 0;
kiov++;
nob -= fragnob;
} while (nob > 0);
- return kiblnd_map_tx(ni, tx, rd, sg - tx->tx_frags);
+ return kiblnd_map_tx(ni, tx, rd, sg_count);
}
static int
{
struct kib_msg *msg = tx->tx_msg;
struct kib_peer_ni *peer_ni = conn->ibc_peer;
+ struct lnet_ni *ni = peer_ni->ibp_ni;
+ struct kib_fast_reg_descriptor *frd = tx->tx_fmr.fmr_frd;
int ver = conn->ibc_version;
int rc;
int done;
LASSERT(conn->ibc_credits <= conn->ibc_queue_depth);
if (conn->ibc_nsends_posted ==
- conn->ibc_queue_depth) {
+ kiblnd_concurrent_sends(ver, ni)) {
/* tx completions outstanding... */
CDEBUG(D_NET, "%s: posted enough\n",
libcfs_nid2str(peer_ni->ibp_nid));
/* close_conn will launch failover */
rc = -ENETDOWN;
} else {
- struct kib_fast_reg_descriptor *frd = tx->tx_fmr.fmr_frd;
struct ib_send_wr *bad = &tx->tx_wrq[tx->tx_nwrq - 1].wr;
struct ib_send_wr *wr = &tx->tx_wrq[0].wr;
- if (frd != NULL) {
+ if (frd != NULL && !frd->frd_posted) {
if (!frd->frd_valid) {
wr = &frd->frd_inv_wr.wr;
wr->next = &frd->frd_fastreg_wr.wr;
libcfs_nid2str(conn->ibc_peer->ibp_nid));
bad = NULL;
- rc = ib_post_send(conn->ibc_cmid->qp, wr, &bad);
+ if (lnet_send_error_simulation(tx->tx_lntmsg[0], &tx->tx_hstatus))
+ rc = -EINVAL;
+ else
+#ifdef HAVE_IB_POST_SEND_RECV_CONST
+ rc = ib_post_send(conn->ibc_cmid->qp, wr,
+ (const struct ib_send_wr **)&bad);
+#else
+ rc = ib_post_send(conn->ibc_cmid->qp, wr, &bad);
+#endif
}
conn->ibc_last_send = ktime_get();
- if (rc == 0)
- return 0;
+ if (rc == 0) {
+ if (frd != NULL)
+ frd->frd_posted = true;
+ return 0;
+ }
/* NB credits are transferred in the actual
* message, which can only be the last work item */
}
LASSERT(conn->ibc_nsends_posted <=
- conn->ibc_queue_depth);
+ kiblnd_concurrent_sends(ver, ni));
LASSERT (!IBLND_OOB_CAPABLE(ver) ||
conn->ibc_noops_posted <= IBLND_OOB_MSGS(ver));
LASSERT (conn->ibc_reserved_credits >= 0);
while (conn->ibc_reserved_credits > 0 &&
- !list_empty(&conn->ibc_tx_queue_rsrvd)) {
- tx = list_entry(conn->ibc_tx_queue_rsrvd.next,
- struct kib_tx, tx_list);
- list_del(&tx->tx_list);
- list_add_tail(&tx->tx_list, &conn->ibc_tx_queue);
+ (tx = list_first_entry_or_null(&conn->ibc_tx_queue_rsrvd,
+ struct kib_tx, tx_list)) != NULL) {
+ list_move_tail(&tx->tx_list, &conn->ibc_tx_queue);
conn->ibc_reserved_credits--;
}
if (!list_empty(&conn->ibc_tx_queue_nocred)) {
credit = 0;
- tx = list_entry(conn->ibc_tx_queue_nocred.next,
- struct kib_tx, tx_list);
+ tx = list_first_entry(&conn->ibc_tx_queue_nocred,
+ struct kib_tx, tx_list);
} else if (!list_empty(&conn->ibc_tx_noops)) {
LASSERT (!IBLND_OOB_CAPABLE(ver));
credit = 1;
- tx = list_entry(conn->ibc_tx_noops.next,
- struct kib_tx, tx_list);
+ tx = list_first_entry(&conn->ibc_tx_noops,
+ struct kib_tx, tx_list);
} else if (!list_empty(&conn->ibc_tx_queue)) {
credit = 1;
- tx = list_entry(conn->ibc_tx_queue.next,
- struct kib_tx, tx_list);
+ tx = list_first_entry(&conn->ibc_tx_queue,
+ struct kib_tx, tx_list);
} else
break;
static void
kiblnd_tx_complete(struct kib_tx *tx, int status)
{
- int failed = (status != IB_WC_SUCCESS);
+ int failed = (status != IB_WC_SUCCESS);
struct kib_conn *conn = tx->tx_conn;
- int idle;
+ int idle;
- LASSERT (tx->tx_sending > 0);
+ if (tx->tx_sending <= 0) {
+ CERROR("Received an event on a freed tx: %p status %d\n",
+ tx, tx->tx_status);
+ return;
+ }
- if (failed) {
- if (conn->ibc_state == IBLND_CONN_ESTABLISHED)
+ if (failed) {
+ if (conn->ibc_state == IBLND_CONN_ESTABLISHED)
CNETERR("Tx -> %s cookie %#llx"
- " sending %d waiting %d: failed %d\n",
- libcfs_nid2str(conn->ibc_peer->ibp_nid),
- tx->tx_cookie, tx->tx_sending, tx->tx_waiting,
- status);
+ " sending %d waiting %d: failed %d\n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid),
+ tx->tx_cookie, tx->tx_sending, tx->tx_waiting,
+ status);
- kiblnd_close_conn(conn, -EIO);
- } else {
- kiblnd_peer_alive(conn->ibc_peer);
- }
+ kiblnd_close_conn(conn, -EIO);
+ } else {
+ kiblnd_peer_alive(conn->ibc_peer);
+ }
spin_lock(&conn->ibc_lock);
#endif
LASSERT(tx->tx_nwrq >= 0);
- LASSERT(tx->tx_nwrq < IBLND_MAX_RDMA_FRAGS + 1);
+ LASSERT(tx->tx_nwrq <= IBLND_MAX_RDMA_FRAGS);
LASSERT(nob <= IBLND_MSG_SIZE);
#ifdef HAVE_IB_GET_DMA_MR
LASSERT(mr != NULL);
break;
}
- sge_nob = MIN(MIN(kiblnd_rd_frag_size(srcrd, srcidx),
- kiblnd_rd_frag_size(dstrd, dstidx)), resid);
+ sge_nob = min3(kiblnd_rd_frag_size(srcrd, srcidx),
+ kiblnd_rd_frag_size(dstrd, dstidx),
+ resid);
sge = &tx->tx_sge[tx->tx_nsge];
sge->addr = kiblnd_rd_frag_addr(srcrd, srcidx);
LASSERT(!tx->tx_queued); /* not queued for sending already */
LASSERT(conn->ibc_state >= IBLND_CONN_ESTABLISHED);
- timeout_ns = lnet_get_lnd_timeout() * NSEC_PER_SEC;
+ if (conn->ibc_state >= IBLND_CONN_DISCONNECTED) {
+ tx->tx_status = -ECONNABORTED;
+ tx->tx_waiting = 0;
+ if (tx->tx_conn != NULL) {
+ /* PUT_DONE first attached to conn as a PUT_REQ */
+ LASSERT(tx->tx_conn == conn);
+ LASSERT(tx->tx_msg->ibm_type == IBLND_MSG_PUT_DONE);
+ tx->tx_conn = NULL;
+ kiblnd_conn_decref(conn);
+ }
+ list_add(&tx->tx_list, &conn->ibc_zombie_txs);
+
+ return;
+ }
+
+ timeout_ns = kiblnd_timeout() * NSEC_PER_SEC;
tx->tx_queued = 1;
tx->tx_deadline = ktime_add_ns(ktime_get(), timeout_ns);
spin_unlock(&conn->ibc_lock);
}
-static int kiblnd_resolve_addr(struct rdma_cm_id *cmid,
- struct sockaddr_in *srcaddr,
- struct sockaddr_in *dstaddr,
- int timeout_ms)
+static int
+kiblnd_resolve_addr_cap(struct rdma_cm_id *cmid,
+ struct sockaddr_in *srcaddr,
+ struct sockaddr_in *dstaddr,
+ int timeout_ms)
{
unsigned short port;
int rc;
}
}
- CERROR("Failed to bind to a free privileged port\n");
- return rc;
+ CERROR("cannot bind to a free privileged port: rc = %d\n", rc);
+
+ return rc;
+}
+
+static int
+kiblnd_resolve_addr(struct rdma_cm_id *cmid,
+ struct sockaddr_in *srcaddr,
+ struct sockaddr_in *dstaddr,
+ int timeout_ms)
+{
+ const struct cred *old_creds = NULL;
+ struct cred *new_creds;
+ int rc;
+
+ if (!capable(CAP_NET_BIND_SERVICE)) {
+ new_creds = prepare_kernel_cred(NULL);
+ if (!new_creds)
+ return -ENOMEM;
+
+ cap_raise(new_creds->cap_effective, CAP_NET_BIND_SERVICE);
+ old_creds = override_creds(new_creds);
+ }
+
+ rc = kiblnd_resolve_addr_cap(cmid, srcaddr, dstaddr, timeout_ms);
+
+ if (old_creds)
+ revert_creds(old_creds);
+
+ return rc;
}
static void
LASSERT (net != NULL);
LASSERT (peer_ni->ibp_connecting > 0);
- cmid = kiblnd_rdma_create_id(kiblnd_cm_callback, peer_ni, RDMA_PS_TCP,
- IB_QPT_RC);
+ cmid = kiblnd_rdma_create_id(peer_ni->ibp_ni->ni_net_ns,
+ kiblnd_cm_callback, peer_ni,
+ RDMA_PS_TCP, IB_QPT_RC);
if (IS_ERR(cmid)) {
CERROR("Can't create CMID for %s: %ld\n",
if (*kiblnd_tunables.kib_use_priv_port) {
rc = kiblnd_resolve_addr(cmid, &srcaddr, &dstaddr,
- lnet_get_lnd_timeout() * 1000);
+ kiblnd_timeout() * 1000);
} else {
rc = rdma_resolve_addr(cmid,
(struct sockaddr *)&srcaddr,
(struct sockaddr *)&dstaddr,
- lnet_get_lnd_timeout() * 1000);
+ kiblnd_timeout() * 1000);
}
if (rc != 0) {
/* Can't initiate address resolution: */
bool
kiblnd_reconnect_peer(struct kib_peer_ni *peer_ni)
{
- rwlock_t *glock = &kiblnd_data.kib_global_lock;
- char *reason = NULL;
- struct list_head txs;
- unsigned long flags;
-
- INIT_LIST_HEAD(&txs);
+ rwlock_t *glock = &kiblnd_data.kib_global_lock;
+ char *reason = NULL;
+ LIST_HEAD(txs);
+ unsigned long flags;
write_lock_irqsave(glock, flags);
if (peer_ni->ibp_reconnecting == 0) {
struct kib_peer_ni *peer2;
struct kib_conn *conn;
rwlock_t *g_lock = &kiblnd_data.kib_global_lock;
- unsigned long flags;
- int rc;
- int i;
+ unsigned long flags;
+ int rc;
+ int i;
struct lnet_ioctl_config_o2iblnd_tunables *tunables;
- /* If I get here, I've committed to send, so I complete the tx with
- * failure on any problems */
+ /* If I get here, I've committed to send, so I complete the tx with
+ * failure on any problems
+ */
- LASSERT (tx == NULL || tx->tx_conn == NULL); /* only set when assigned a conn */
- LASSERT (tx == NULL || tx->tx_nwrq > 0); /* work items have been set up */
+ LASSERT(!tx || !tx->tx_conn); /* only set when assigned a conn */
+ LASSERT(!tx || tx->tx_nwrq > 0); /* work items have been set up */
- /* First time, just use a read lock since I expect to find my peer_ni
- * connected */
+ /* First time, just use a read lock since I expect to find my peer_ni
+ * connected
+ */
read_lock_irqsave(g_lock, flags);
- peer_ni = kiblnd_find_peer_locked(ni, nid);
+ peer_ni = kiblnd_find_peer_locked(ni, nid);
if (peer_ni != NULL && !list_empty(&peer_ni->ibp_conns)) {
- /* Found a peer_ni with an established connection */
- conn = kiblnd_get_conn_locked(peer_ni);
- kiblnd_conn_addref(conn); /* 1 ref for me... */
+ /* Found a peer_ni with an established connection */
+ conn = kiblnd_get_conn_locked(peer_ni);
+ kiblnd_conn_addref(conn); /* 1 ref for me... */
read_unlock_irqrestore(g_lock, flags);
- if (tx != NULL)
- kiblnd_queue_tx(tx, conn);
- kiblnd_conn_decref(conn); /* ...to here */
- return;
- }
+ if (tx != NULL)
+ kiblnd_queue_tx(tx, conn);
+ kiblnd_conn_decref(conn); /* ...to here */
+ return;
+ }
read_unlock(g_lock);
/* Re-try with a write lock */
write_lock(g_lock);
- peer_ni = kiblnd_find_peer_locked(ni, nid);
- if (peer_ni != NULL) {
+ peer_ni = kiblnd_find_peer_locked(ni, nid);
+ if (peer_ni != NULL) {
if (list_empty(&peer_ni->ibp_conns)) {
- /* found a peer_ni, but it's still connecting... */
+ /* found a peer_ni, but it's still connecting... */
LASSERT(kiblnd_peer_connecting(peer_ni));
- if (tx != NULL)
+ if (tx != NULL)
list_add_tail(&tx->tx_list,
- &peer_ni->ibp_tx_queue);
+ &peer_ni->ibp_tx_queue);
write_unlock_irqrestore(g_lock, flags);
} else {
conn = kiblnd_get_conn_locked(peer_ni);
write_unlock_irqrestore(g_lock, flags);
- if (tx != NULL)
- kiblnd_queue_tx(tx, conn);
- kiblnd_conn_decref(conn); /* ...to here */
- }
- return;
- }
+ if (tx != NULL)
+ kiblnd_queue_tx(tx, conn);
+ kiblnd_conn_decref(conn); /* ...to here */
+ }
+ return;
+ }
write_unlock_irqrestore(g_lock, flags);
write_lock_irqsave(g_lock, flags);
- peer2 = kiblnd_find_peer_locked(ni, nid);
- if (peer2 != NULL) {
+ peer2 = kiblnd_find_peer_locked(ni, nid);
+ if (peer2 != NULL) {
if (list_empty(&peer2->ibp_conns)) {
- /* found a peer_ni, but it's still connecting... */
+ /* found a peer_ni, but it's still connecting... */
LASSERT(kiblnd_peer_connecting(peer2));
- if (tx != NULL)
+ if (tx != NULL)
list_add_tail(&tx->tx_list,
- &peer2->ibp_tx_queue);
+ &peer2->ibp_tx_queue);
write_unlock_irqrestore(g_lock, flags);
} else {
conn = kiblnd_get_conn_locked(peer2);
write_unlock_irqrestore(g_lock, flags);
- if (tx != NULL)
- kiblnd_queue_tx(tx, conn);
- kiblnd_conn_decref(conn); /* ...to here */
- }
+ if (tx != NULL)
+ kiblnd_queue_tx(tx, conn);
+ kiblnd_conn_decref(conn); /* ...to here */
+ }
- kiblnd_peer_decref(peer_ni);
- return;
- }
+ kiblnd_peer_decref(peer_ni);
+ return;
+ }
/* Brand new peer_ni */
LASSERT(peer_ni->ibp_connecting == 0);
if (tx != NULL)
list_add_tail(&tx->tx_list, &peer_ni->ibp_tx_queue);
- kiblnd_peer_addref(peer_ni);
- list_add_tail(&peer_ni->ibp_list, kiblnd_nid2peerlist(nid));
+ kiblnd_peer_addref(peer_ni);
+ hash_add(kiblnd_data.kib_peers, &peer_ni->ibp_list, nid);
write_unlock_irqrestore(g_lock, flags);
for (i = 0; i < tunables->lnd_conns_per_peer; i++)
kiblnd_connect_peer(peer_ni);
- kiblnd_peer_decref(peer_ni);
+ kiblnd_peer_decref(peer_ni);
}
int
int target_is_router = lntmsg->msg_target_is_router;
int routing = lntmsg->msg_routing;
unsigned int payload_niov = lntmsg->msg_niov;
- struct kvec *payload_iov = lntmsg->msg_iov;
- lnet_kiov_t *payload_kiov = lntmsg->msg_kiov;
+ struct bio_vec *payload_kiov = lntmsg->msg_kiov;
unsigned int payload_offset = lntmsg->msg_offset;
unsigned int payload_nob = lntmsg->msg_len;
struct kib_msg *ibmsg;
/* Thread context */
LASSERT (!in_interrupt());
- /* payload is either all vaddrs or all pages */
- LASSERT (!(payload_kiov != NULL && payload_iov != NULL));
+
+ tx = kiblnd_get_idle_tx(ni, target.nid);
+ if (tx == NULL) {
+ CERROR("Can't allocate %s txd for %s\n",
+ lnet_msgtyp2str(type),
+ libcfs_nid2str(target.nid));
+ return -ENOMEM;
+ }
+ ibmsg = tx->tx_msg;
switch (type) {
default:
/* is the REPLY message too small for RDMA? */
nob = offsetof(struct kib_msg, ibm_u.immediate.ibim_payload[lntmsg->msg_md->md_length]);
- if (nob <= IBLND_MSG_SIZE)
- break; /* send IMMEDIATE */
+ if (nob <= IBLND_MSG_SIZE && !lntmsg->msg_rdma_force)
+ break; /* send IMMEDIATE */
- tx = kiblnd_get_idle_tx(ni, target.nid);
- if (tx == NULL) {
- CERROR("Can't allocate txd for GET to %s\n",
- libcfs_nid2str(target.nid));
- return -ENOMEM;
- }
-
- ibmsg = tx->tx_msg;
rd = &ibmsg->ibm_u.get.ibgm_rd;
- if ((lntmsg->msg_md->md_options & LNET_MD_KIOV) == 0)
- rc = kiblnd_setup_rd_iov(ni, tx, rd,
- lntmsg->msg_md->md_niov,
- lntmsg->msg_md->md_iov.iov,
- 0, lntmsg->msg_md->md_length);
- else
- rc = kiblnd_setup_rd_kiov(ni, tx, rd,
- lntmsg->msg_md->md_niov,
- lntmsg->msg_md->md_iov.kiov,
- 0, lntmsg->msg_md->md_length);
+ rc = kiblnd_setup_rd_kiov(ni, tx, rd,
+ lntmsg->msg_md->md_niov,
+ lntmsg->msg_md->md_kiov,
+ 0, lntmsg->msg_md->md_length);
if (rc != 0) {
CERROR("Can't setup GET sink for %s: %d\n",
libcfs_nid2str(target.nid), rc);
return -EIO;
}
- tx->tx_lntmsg[0] = lntmsg; /* finalise lntmsg[0,1] on completion */
- tx->tx_waiting = 1; /* waiting for GET_DONE */
- kiblnd_launch_tx(ni, tx, target.nid);
- return 0;
+ /* finalise lntmsg[0,1] on completion */
+ tx->tx_lntmsg[0] = lntmsg;
+ tx->tx_waiting = 1; /* waiting for GET_DONE */
+ kiblnd_launch_tx(ni, tx, target.nid);
+ return 0;
- case LNET_MSG_REPLY:
- case LNET_MSG_PUT:
- /* Is the payload small enough not to need RDMA? */
+ case LNET_MSG_REPLY:
+ case LNET_MSG_PUT:
+ /* Is the payload small enough not to need RDMA? */
nob = offsetof(struct kib_msg, ibm_u.immediate.ibim_payload[payload_nob]);
- if (nob <= IBLND_MSG_SIZE)
- break; /* send IMMEDIATE */
+ if (nob <= IBLND_MSG_SIZE && !lntmsg->msg_rdma_force)
+ break; /* send IMMEDIATE */
- tx = kiblnd_get_idle_tx(ni, target.nid);
- if (tx == NULL) {
- CERROR("Can't allocate %s txd for %s\n",
- type == LNET_MSG_PUT ? "PUT" : "REPLY",
- libcfs_nid2str(target.nid));
- return -ENOMEM;
- }
-
- if (payload_kiov == NULL)
- rc = kiblnd_setup_rd_iov(ni, tx, tx->tx_rd,
- payload_niov, payload_iov,
- payload_offset, payload_nob);
- else
- rc = kiblnd_setup_rd_kiov(ni, tx, tx->tx_rd,
- payload_niov, payload_kiov,
- payload_offset, payload_nob);
+ rc = kiblnd_setup_rd_kiov(ni, tx, tx->tx_rd,
+ payload_niov, payload_kiov,
+ payload_offset, payload_nob);
if (rc != 0) {
CERROR("Can't setup PUT src for %s: %d\n",
libcfs_nid2str(target.nid), rc);
return -EIO;
}
- ibmsg = tx->tx_msg;
- ibmsg->ibm_u.putreq.ibprm_hdr = *hdr;
- ibmsg->ibm_u.putreq.ibprm_cookie = tx->tx_cookie;
+ ibmsg->ibm_u.putreq.ibprm_hdr = *hdr;
+ ibmsg->ibm_u.putreq.ibprm_cookie = tx->tx_cookie;
kiblnd_init_tx_msg(ni, tx, IBLND_MSG_PUT_REQ,
sizeof(struct kib_putreq_msg));
- tx->tx_lntmsg[0] = lntmsg; /* finalise lntmsg on completion */
- tx->tx_waiting = 1; /* waiting for PUT_{ACK,NAK} */
- kiblnd_launch_tx(ni, tx, target.nid);
- return 0;
- }
+ /* finalise lntmsg[0,1] on completion */
+ tx->tx_lntmsg[0] = lntmsg;
+ tx->tx_waiting = 1; /* waiting for PUT_{ACK,NAK} */
+ kiblnd_launch_tx(ni, tx, target.nid);
+ return 0;
+ }
/* send IMMEDIATE */
LASSERT(offsetof(struct kib_msg, ibm_u.immediate.ibim_payload[payload_nob])
<= IBLND_MSG_SIZE);
- tx = kiblnd_get_idle_tx(ni, target.nid);
- if (tx == NULL) {
- CERROR ("Can't send %d to %s: tx descs exhausted\n",
- type, libcfs_nid2str(target.nid));
- return -ENOMEM;
- }
-
- ibmsg = tx->tx_msg;
- ibmsg->ibm_u.immediate.ibim_hdr = *hdr;
+ ibmsg = tx->tx_msg;
+ ibmsg->ibm_u.immediate.ibim_hdr = *hdr;
- if (payload_kiov != NULL)
- lnet_copy_kiov2flat(IBLND_MSG_SIZE, ibmsg,
- offsetof(struct kib_msg, ibm_u.immediate.ibim_payload),
- payload_niov, payload_kiov,
- payload_offset, payload_nob);
- else
- lnet_copy_iov2flat(IBLND_MSG_SIZE, ibmsg,
- offsetof(struct kib_msg, ibm_u.immediate.ibim_payload),
- payload_niov, payload_iov,
- payload_offset, payload_nob);
+ lnet_copy_kiov2flat(IBLND_MSG_SIZE, ibmsg,
+ offsetof(struct kib_msg,
+ ibm_u.immediate.ibim_payload),
+ payload_niov, payload_kiov,
+ payload_offset, payload_nob);
nob = offsetof(struct kib_immediate_msg, ibim_payload[payload_nob]);
- kiblnd_init_tx_msg(ni, tx, IBLND_MSG_IMMEDIATE, nob);
- tx->tx_lntmsg[0] = lntmsg; /* finalise lntmsg on completion */
- kiblnd_launch_tx(ni, tx, target.nid);
- return 0;
+ kiblnd_init_tx_msg(ni, tx, IBLND_MSG_IMMEDIATE, nob);
+
+ /* finalise lntmsg on completion */
+ tx->tx_lntmsg[0] = lntmsg;
+
+ kiblnd_launch_tx(ni, tx, target.nid);
+ return 0;
}
static void
kiblnd_reply(struct lnet_ni *ni, struct kib_rx *rx, struct lnet_msg *lntmsg)
{
struct lnet_process_id target = lntmsg->msg_target;
- unsigned int niov = lntmsg->msg_niov;
- struct kvec *iov = lntmsg->msg_iov;
- lnet_kiov_t *kiov = lntmsg->msg_kiov;
- unsigned int offset = lntmsg->msg_offset;
- unsigned int nob = lntmsg->msg_len;
+ unsigned int niov = lntmsg->msg_niov;
+ struct bio_vec *kiov = lntmsg->msg_kiov;
+ unsigned int offset = lntmsg->msg_offset;
+ unsigned int nob = lntmsg->msg_len;
struct kib_tx *tx;
- int rc;
+ int rc;
tx = kiblnd_get_idle_tx(ni, rx->rx_conn->ibc_peer->ibp_nid);
if (tx == NULL) {
if (nob == 0)
rc = 0;
- else if (kiov == NULL)
- rc = kiblnd_setup_rd_iov(ni, tx, tx->tx_rd,
- niov, iov, offset, nob);
else
rc = kiblnd_setup_rd_kiov(ni, tx, tx->tx_rd,
niov, kiov, offset, nob);
lnet_finalize(lntmsg, -EIO);
}
+unsigned int
+kiblnd_get_dev_prio(struct lnet_ni *ni, unsigned int dev_idx)
+{
+ struct kib_net *net = ni->ni_data;
+ struct device *dev = NULL;
+
+ if (net)
+ dev = net->ibn_dev->ibd_hdev->ibh_ibdev->dma_device;
+
+ return lnet_get_dev_prio(dev, dev_idx);
+
+}
+
int
kiblnd_recv(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg,
- int delayed, unsigned int niov, struct kvec *iov, lnet_kiov_t *kiov,
+ int delayed, unsigned int niov, struct bio_vec *kiov,
unsigned int offset, unsigned int mlen, unsigned int rlen)
{
struct kib_rx *rx = private;
LASSERT (mlen <= rlen);
LASSERT (!in_interrupt());
- /* Either all pages or all vaddrs */
- LASSERT (!(kiov != NULL && iov != NULL));
switch (rxmsg->ibm_type) {
default:
break;
}
- if (kiov != NULL)
- lnet_copy_flat2kiov(niov, kiov, offset,
- IBLND_MSG_SIZE, rxmsg,
- offsetof(struct kib_msg, ibm_u.immediate.ibim_payload),
- mlen);
- else
- lnet_copy_flat2iov(niov, iov, offset,
- IBLND_MSG_SIZE, rxmsg,
- offsetof(struct kib_msg, ibm_u.immediate.ibim_payload),
- mlen);
+ lnet_copy_flat2kiov(niov, kiov, offset,
+ IBLND_MSG_SIZE, rxmsg,
+ offsetof(struct kib_msg,
+ ibm_u.immediate.ibim_payload),
+ mlen);
lnet_finalize(lntmsg, 0);
break;
txmsg = tx->tx_msg;
rd = &txmsg->ibm_u.putack.ibpam_rd;
- if (kiov == NULL)
- rc = kiblnd_setup_rd_iov(ni, tx, rd,
- niov, iov, offset, mlen);
- else
- rc = kiblnd_setup_rd_kiov(ni, tx, rd,
- niov, kiov, offset, mlen);
+ rc = kiblnd_setup_rd_kiov(ni, tx, rd,
+ niov, kiov, offset, mlen);
if (rc != 0) {
CERROR("Can't setup PUT sink for %s: %d\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid), rc);
return rc;
}
-int
-kiblnd_thread_start(int (*fn)(void *arg), void *arg, char *name)
-{
- struct task_struct *task = kthread_run(fn, arg, name);
-
- if (IS_ERR(task))
- return PTR_ERR(task);
-
- atomic_inc(&kiblnd_data.kib_nthreads);
- return 0;
-}
-
static void
kiblnd_thread_fini (void)
{
static void
kiblnd_peer_notify(struct kib_peer_ni *peer_ni)
{
- int error = 0;
+ int error = 0;
time64_t last_alive = 0;
- unsigned long flags;
+ unsigned long flags;
read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
if (kiblnd_peer_idle(peer_ni) && peer_ni->ibp_error != 0) {
- error = peer_ni->ibp_error;
- peer_ni->ibp_error = 0;
+ error = peer_ni->ibp_error;
+ peer_ni->ibp_error = 0;
- last_alive = peer_ni->ibp_last_alive;
- }
+ last_alive = peer_ni->ibp_last_alive;
+ }
read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
- if (error != 0)
- lnet_notify(peer_ni->ibp_ni,
- peer_ni->ibp_nid, 0, last_alive);
+ if (error != 0)
+ lnet_notify(peer_ni->ibp_ni,
+ peer_ni->ibp_nid, false, false, last_alive);
}
void
LASSERT(conn->ibc_state >= IBLND_CONN_ESTABLISHED);
write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- while (!list_empty(&conn->ibc_early_rxs)) {
- rx = list_entry(conn->ibc_early_rxs.next,
- struct kib_rx, rx_list);
+ while ((rx = list_first_entry_or_null(&conn->ibc_early_rxs,
+ struct kib_rx,
+ rx_list)) != NULL) {
list_del(&rx->rx_list);
write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
}
-static void
+void
kiblnd_abort_txs(struct kib_conn *conn, struct list_head *txs)
{
- struct list_head zombies = LIST_HEAD_INIT(zombies);
- struct list_head *tmp;
- struct list_head *nxt;
+ LIST_HEAD(zombies);
+ struct kib_tx *nxt;
struct kib_tx *tx;
spin_lock(&conn->ibc_lock);
- list_for_each_safe(tmp, nxt, txs) {
- tx = list_entry(tmp, struct kib_tx, tx_list);
-
+ list_for_each_entry_safe(tx, nxt, txs, tx_list) {
if (txs == &conn->ibc_active_txs) {
LASSERT(!tx->tx_queued);
LASSERT(tx->tx_waiting ||
*/
if (tx->tx_sending == 0) {
tx->tx_queued = 0;
- list_del(&tx->tx_list);
- list_add(&tx->tx_list, &zombies);
+ list_move(&tx->tx_list, &zombies);
+ } else {
+ /* keep tx until cq destroy */
+ list_move(&tx->tx_list, &conn->ibc_zombie_txs);
+ conn->ibc_waits ++;
}
}
/*
* aborting transmits occurs when finalizing the connection.
- * The connection is finalized on error
+ * The connection is finalized on error.
+ * Passing LNET_MSG_STATUS_OK to txlist_done() will not
+ * override the value already set in tx->tx_hstatus above.
*/
- kiblnd_txlist_done(&zombies, -ECONNABORTED, -1);
+ kiblnd_txlist_done(&zombies, -ECONNABORTED, LNET_MSG_STATUS_OK);
+}
+
+static bool
+kiblnd_tx_may_discard(struct kib_conn *conn)
+{
+ bool rc = false;
+ struct kib_tx *nxt;
+ struct kib_tx *tx;
+
+ spin_lock(&conn->ibc_lock);
+
+ list_for_each_entry_safe(tx, nxt, &conn->ibc_zombie_txs, tx_list) {
+ if (tx->tx_sending > 0 && tx->tx_lntmsg[0] &&
+ lnet_md_discarded(tx->tx_lntmsg[0]->msg_md)) {
+ tx->tx_sending --;
+ if (tx->tx_sending == 0) {
+ kiblnd_conn_decref(tx->tx_conn);
+ tx->tx_conn = NULL;
+ rc = true;
+ }
+ }
+ }
+
+ spin_unlock(&conn->ibc_lock);
+ return rc;
}
static void
LASSERT (!in_interrupt());
LASSERT (conn->ibc_state > IBLND_CONN_INIT);
- kiblnd_set_conn_state(conn, IBLND_CONN_DISCONNECTED);
-
/* abort_receives moves QP state to IB_QPS_ERR. This is only required
* for connections that didn't get as far as being connected, because
* rdma_disconnect() does this for free. */
kiblnd_abort_receives(conn);
+ kiblnd_set_conn_state(conn, IBLND_CONN_DISCONNECTED);
+
/* Complete all tx descs not waiting for sends to complete.
* NB we should be safe from RDMA now that the QP has changed state */
kiblnd_peer_connect_failed(struct kib_peer_ni *peer_ni, int active,
int error)
{
- struct list_head zombies = LIST_HEAD_INIT(zombies);
- unsigned long flags;
+ LIST_HEAD(zombies);
+ unsigned long flags;
+ enum lnet_msg_hstatus hstatus;
- LASSERT (error != 0);
- LASSERT (!in_interrupt());
+ LASSERT(error != 0);
+ LASSERT(!in_interrupt());
write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
CNETERR("Deleting messages for %s: connection failed\n",
libcfs_nid2str(peer_ni->ibp_nid));
- kiblnd_txlist_done(&zombies, error,
- LNET_MSG_STATUS_LOCAL_DROPPED);
+ switch (error) {
+ case -EHOSTUNREACH:
+ case -ETIMEDOUT:
+ hstatus = LNET_MSG_STATUS_NETWORK_TIMEOUT;
+ break;
+ case -ECONNREFUSED:
+ hstatus = LNET_MSG_STATUS_REMOTE_DROPPED;
+ break;
+ default:
+ hstatus = LNET_MSG_STATUS_LOCAL_DROPPED;
+ break;
+ }
+
+ kiblnd_txlist_done(&zombies, error, hstatus);
}
static void
{
struct kib_peer_ni *peer_ni = conn->ibc_peer;
struct kib_tx *tx;
- struct list_head txs;
+ LIST_HEAD(txs);
unsigned long flags;
int active;
(conn->ibc_state == IBLND_CONN_PASSIVE_WAIT &&
peer_ni->ibp_accepting > 0));
- LIBCFS_FREE(conn->ibc_connvars, sizeof(*conn->ibc_connvars));
- conn->ibc_connvars = NULL;
+ LIBCFS_FREE(conn->ibc_connvars, sizeof(*conn->ibc_connvars));
+ conn->ibc_connvars = NULL;
- if (status != 0) {
- /* failed to establish connection */
- kiblnd_peer_connect_failed(peer_ni, active, status);
- kiblnd_finalise_conn(conn);
- return;
- }
+ if (status != 0) {
+ /* failed to establish connection */
+ kiblnd_peer_connect_failed(peer_ni, active, status);
+ kiblnd_finalise_conn(conn);
+ return;
+ }
- /* connection established */
+ /* connection established */
write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
conn->ibc_last_send = ktime_get();
- kiblnd_set_conn_state(conn, IBLND_CONN_ESTABLISHED);
- kiblnd_peer_alive(peer_ni);
+ kiblnd_set_conn_state(conn, IBLND_CONN_ESTABLISHED);
+ kiblnd_peer_alive(peer_ni);
/* Add conn to peer_ni's list and nuke any dangling conns from a different
* peer_ni instance... */
}
/* grab pending txs while I have the lock */
- INIT_LIST_HEAD(&txs);
list_splice_init(&peer_ni->ibp_tx_queue, &txs);
if (!kiblnd_peer_active(peer_ni) || /* peer_ni has been deleted */
* scheduled. We won't be using round robin on this first batch.
*/
spin_lock(&conn->ibc_lock);
- while (!list_empty(&txs)) {
- tx = list_entry(txs.next, struct kib_tx, tx_list);
+ while ((tx = list_first_entry_or_null(&txs, struct kib_tx,
+ tx_list)) != NULL) {
list_del(&tx->tx_list);
kiblnd_queue_tx_locked(tx, conn);
{
int rc;
+#ifdef HAVE_RDMA_REJECT_4ARGS
+ rc = rdma_reject(cmid, rej, sizeof(*rej), IB_CM_REJ_CONSUMER_DEFINED);
+#else
rc = rdma_reject(cmid, rej, sizeof(*rej));
+#endif
if (rc != 0)
CWARN("Error %d sending reject\n", rc);
static int
kiblnd_passive_connect(struct rdma_cm_id *cmid, void *priv, int priv_nob)
{
- rwlock_t *g_lock = &kiblnd_data.kib_global_lock;
+ rwlock_t *g_lock = &kiblnd_data.kib_global_lock;
struct kib_msg *reqmsg = priv;
struct kib_msg *ackmsg;
struct kib_dev *ibdev;
struct kib_conn *conn;
struct lnet_ni *ni = NULL;
struct kib_net *net = NULL;
- lnet_nid_t nid;
- struct rdma_conn_param cp;
+ lnet_nid_t nid;
+ struct rdma_conn_param cp;
struct kib_rej rej;
- int version = IBLND_MSG_VERSION;
- unsigned long flags;
- int rc;
- struct sockaddr_in *peer_addr;
- LASSERT (!in_interrupt());
+ int version = IBLND_MSG_VERSION;
+ unsigned long flags;
+ int rc;
+ struct sockaddr_in *peer_addr;
+ LASSERT(!in_interrupt());
/* cmid inherits 'context' from the corresponding listener id */
ibdev = cmid->context;
LASSERT(ibdev);
- memset(&rej, 0, sizeof(rej));
- rej.ibr_magic = IBLND_MSG_MAGIC;
- rej.ibr_why = IBLND_REJECT_FATAL;
- rej.ibr_cp.ibcp_max_msg_size = IBLND_MSG_SIZE;
+ memset(&rej, 0, sizeof(rej));
+ rej.ibr_magic = IBLND_MSG_MAGIC;
+ rej.ibr_why = IBLND_REJECT_FATAL;
+ rej.ibr_cp.ibcp_max_msg_size = IBLND_MSG_SIZE;
- peer_addr = (struct sockaddr_in *)&(cmid->route.addr.dst_addr);
- if (*kiblnd_tunables.kib_require_priv_port &&
- ntohs(peer_addr->sin_port) >= PROT_SOCK) {
+ peer_addr = (struct sockaddr_in *)&(cmid->route.addr.dst_addr);
+ if (*kiblnd_tunables.kib_require_priv_port &&
+ ntohs(peer_addr->sin_port) >= PROT_SOCK) {
__u32 ip = ntohl(peer_addr->sin_addr.s_addr);
CERROR("peer_ni's port (%pI4h:%hu) is not privileged\n",
&ip, ntohs(peer_addr->sin_port));
rej.ibr_incarnation = net->ibn_incarnation;
}
- if (ni == NULL || /* no matching net */
- ni->ni_nid != reqmsg->ibm_dstnid || /* right NET, wrong NID! */
- net->ibn_dev != ibdev) { /* wrong device */
- CERROR("Can't accept conn from %s on %s (%s:%d:%pI4h): "
- "bad dst nid %s\n", libcfs_nid2str(nid),
- ni == NULL ? "NA" : libcfs_nid2str(ni->ni_nid),
+ if (ni == NULL || /* no matching net */
+ lnet_nid_to_nid4(&ni->ni_nid) !=
+ reqmsg->ibm_dstnid || /* right NET, wrong NID! */
+ net->ibn_dev != ibdev) { /* wrong device */
+ CERROR("Can't accept conn from %s on %s (%s:%d:%pI4h): bad dst nid %s\n", libcfs_nid2str(nid),
+ ni ? libcfs_nidstr(&ni->ni_nid) : "NA",
ibdev->ibd_ifname, ibdev->ibd_nnets,
- &ibdev->ibd_ifip,
+ &ibdev->ibd_ifip,
libcfs_nid2str(reqmsg->ibm_dstnid));
goto failed;
}
- /* check time stamp as soon as possible */
+ /* check time stamp as soon as possible */
if (reqmsg->ibm_dststamp != 0 &&
reqmsg->ibm_dststamp != net->ibn_incarnation) {
CWARN("Stale connection request\n");
if (reqmsg->ibm_u.connparams.ibcp_queue_depth >
kiblnd_msg_queue_size(version, ni)) {
- CERROR("Can't accept conn from %s, queue depth too large: "
- " %d (<=%d wanted)\n",
+ CERROR("Can't accept conn from %s, queue depth too large: %d (<=%d wanted)\n",
libcfs_nid2str(nid),
reqmsg->ibm_u.connparams.ibcp_queue_depth,
kiblnd_msg_queue_size(version, ni));
if (reqmsg->ibm_u.connparams.ibcp_max_frags >
IBLND_MAX_RDMA_FRAGS) {
- CWARN("Can't accept conn from %s (version %x): "
- "max_frags %d too large (%d wanted)\n",
+ CWARN("Can't accept conn from %s (version %x): max_frags %d too large (%d wanted)\n",
libcfs_nid2str(nid), version,
reqmsg->ibm_u.connparams.ibcp_max_frags,
IBLND_MAX_RDMA_FRAGS);
} else if (reqmsg->ibm_u.connparams.ibcp_max_frags <
IBLND_MAX_RDMA_FRAGS &&
net->ibn_fmr_ps == NULL) {
- CWARN("Can't accept conn from %s (version %x): "
- "max_frags %d incompatible without FMR pool "
- "(%d wanted)\n",
+ CWARN("Can't accept conn from %s (version %x): max_frags %d incompatible without FMR pool (%d wanted)\n",
libcfs_nid2str(nid), version,
reqmsg->ibm_u.connparams.ibcp_max_frags,
IBLND_MAX_RDMA_FRAGS);
goto failed;
}
- if (reqmsg->ibm_u.connparams.ibcp_max_msg_size > IBLND_MSG_SIZE) {
- CERROR("Can't accept %s: message size %d too big (%d max)\n",
- libcfs_nid2str(nid),
- reqmsg->ibm_u.connparams.ibcp_max_msg_size,
- IBLND_MSG_SIZE);
- goto failed;
- }
+ if (reqmsg->ibm_u.connparams.ibcp_max_msg_size > IBLND_MSG_SIZE) {
+ CERROR("Can't accept %s: message size %d too big (%d max)\n",
+ libcfs_nid2str(nid),
+ reqmsg->ibm_u.connparams.ibcp_max_msg_size,
+ IBLND_MSG_SIZE);
+ goto failed;
+ }
/* assume 'nid' is a new peer_ni; create */
rc = kiblnd_create_peer(ni, &peer_ni, nid);
write_lock_irqsave(g_lock, flags);
- peer2 = kiblnd_find_peer_locked(ni, nid);
- if (peer2 != NULL) {
- if (peer2->ibp_version == 0) {
- peer2->ibp_version = version;
- peer2->ibp_incarnation = reqmsg->ibm_srcstamp;
- }
+ peer2 = kiblnd_find_peer_locked(ni, nid);
+ if (peer2 != NULL) {
+ if (peer2->ibp_version == 0) {
+ peer2->ibp_version = version;
+ peer2->ibp_incarnation = reqmsg->ibm_srcstamp;
+ }
- /* not the guy I've talked with */
- if (peer2->ibp_incarnation != reqmsg->ibm_srcstamp ||
- peer2->ibp_version != version) {
+ /* not the guy I've talked with */
+ if (peer2->ibp_incarnation != reqmsg->ibm_srcstamp ||
+ peer2->ibp_version != version) {
kiblnd_close_peer_conns_locked(peer2, -ESTALE);
if (kiblnd_peer_active(peer2)) {
libcfs_nid2str(nid), peer2->ibp_version, version,
peer2->ibp_incarnation, reqmsg->ibm_srcstamp);
- kiblnd_peer_decref(peer_ni);
- rej.ibr_why = IBLND_REJECT_CONN_STALE;
- goto failed;
- }
+ kiblnd_peer_decref(peer_ni);
+ rej.ibr_why = IBLND_REJECT_CONN_STALE;
+ goto failed;
+ }
/* Tie-break connection race in favour of the higher NID.
* If we keep running into a race condition multiple times,
* the lower NID connection win so we can move forward.
*/
if (peer2->ibp_connecting != 0 &&
- nid < ni->ni_nid && peer2->ibp_races <
- MAX_CONN_RACES_BEFORE_ABORT) {
+ nid < lnet_nid_to_nid4(&ni->ni_nid) &&
+ peer2->ibp_races < MAX_CONN_RACES_BEFORE_ABORT) {
peer2->ibp_races++;
write_unlock_irqrestore(g_lock, flags);
peer2->ibp_queue_depth = peer_ni->ibp_queue_depth;
write_unlock_irqrestore(g_lock, flags);
- kiblnd_peer_decref(peer_ni);
- peer_ni = peer2;
- } else {
- /* Brand new peer_ni */
- LASSERT (peer_ni->ibp_accepting == 0);
- LASSERT (peer_ni->ibp_version == 0 &&
- peer_ni->ibp_incarnation == 0);
+ kiblnd_peer_decref(peer_ni);
+ peer_ni = peer2;
+ } else {
+ /* Brand new peer_ni */
+ LASSERT(peer_ni->ibp_accepting == 0);
+ LASSERT(peer_ni->ibp_version == 0 &&
+ peer_ni->ibp_incarnation == 0);
- peer_ni->ibp_accepting = 1;
- peer_ni->ibp_version = version;
- peer_ni->ibp_incarnation = reqmsg->ibm_srcstamp;
+ peer_ni->ibp_accepting = 1;
+ peer_ni->ibp_version = version;
+ peer_ni->ibp_incarnation = reqmsg->ibm_srcstamp;
- /* I have a ref on ni that prevents it being shutdown */
- LASSERT (net->ibn_shutdown == 0);
+ /* I have a ref on ni that prevents it being shutdown */
+ LASSERT(net->ibn_shutdown == 0);
- kiblnd_peer_addref(peer_ni);
- list_add_tail(&peer_ni->ibp_list, kiblnd_nid2peerlist(nid));
+ kiblnd_peer_addref(peer_ni);
+ hash_add(kiblnd_data.kib_peers, &peer_ni->ibp_list, nid);
write_unlock_irqrestore(g_lock, flags);
- }
+ }
- conn = kiblnd_create_conn(peer_ni, cmid, IBLND_CONN_PASSIVE_WAIT, version);
- if (conn == NULL) {
- kiblnd_peer_connect_failed(peer_ni, 0, -ENOMEM);
- kiblnd_peer_decref(peer_ni);
- rej.ibr_why = IBLND_REJECT_NO_RESOURCES;
- goto failed;
- }
+ conn = kiblnd_create_conn(peer_ni, cmid, IBLND_CONN_PASSIVE_WAIT,
+ version);
+ if (!conn) {
+ kiblnd_peer_connect_failed(peer_ni, 0, -ENOMEM);
+ kiblnd_peer_decref(peer_ni);
+ rej.ibr_why = IBLND_REJECT_NO_RESOURCES;
+ goto failed;
+ }
- /* conn now "owns" cmid, so I return success from here on to ensure the
- * CM callback doesn't destroy cmid. */
+ /* conn now "owns" cmid, so I return success from here on to ensure the
+ * CM callback doesn't destroy cmid.
+ */
conn->ibc_incarnation = reqmsg->ibm_srcstamp;
conn->ibc_credits = conn->ibc_queue_depth;
conn->ibc_reserved_credits = conn->ibc_queue_depth;
LASSERT(conn->ibc_credits + conn->ibc_reserved_credits +
IBLND_OOB_MSGS(version) <= IBLND_RX_MSGS(conn));
- ackmsg = &conn->ibc_connvars->cv_msg;
- memset(ackmsg, 0, sizeof(*ackmsg));
+ ackmsg = &conn->ibc_connvars->cv_msg;
+ memset(ackmsg, 0, sizeof(*ackmsg));
- kiblnd_init_msg(ackmsg, IBLND_MSG_CONNACK,
- sizeof(ackmsg->ibm_u.connparams));
+ kiblnd_init_msg(ackmsg, IBLND_MSG_CONNACK,
+ sizeof(ackmsg->ibm_u.connparams));
ackmsg->ibm_u.connparams.ibcp_queue_depth = conn->ibc_queue_depth;
ackmsg->ibm_u.connparams.ibcp_max_frags = conn->ibc_max_frags;
ackmsg->ibm_u.connparams.ibcp_max_msg_size = IBLND_MSG_SIZE;
- kiblnd_pack_msg(ni, ackmsg, version, 0, nid, reqmsg->ibm_srcstamp);
+ kiblnd_pack_msg(ni, ackmsg, version, 0, nid, reqmsg->ibm_srcstamp);
- memset(&cp, 0, sizeof(cp));
- cp.private_data = ackmsg;
- cp.private_data_len = ackmsg->ibm_nob;
- cp.responder_resources = 0; /* No atomic ops or RDMA reads */
- cp.initiator_depth = 0;
- cp.flow_control = 1;
- cp.retry_count = *kiblnd_tunables.kib_retry_count;
- cp.rnr_retry_count = *kiblnd_tunables.kib_rnr_retry_count;
+ memset(&cp, 0, sizeof(cp));
+ cp.private_data = ackmsg;
+ cp.private_data_len = ackmsg->ibm_nob;
+ cp.responder_resources = 0; /* No atomic ops or RDMA reads */
+ cp.initiator_depth = 0;
+ cp.flow_control = 1;
+ cp.retry_count = *kiblnd_tunables.kib_retry_count;
+ cp.rnr_retry_count = *kiblnd_tunables.kib_rnr_retry_count;
- CDEBUG(D_NET, "Accept %s\n", libcfs_nid2str(nid));
+ CDEBUG(D_NET, "Accept %s\n", libcfs_nid2str(nid));
- rc = rdma_accept(cmid, &cp);
- if (rc != 0) {
- CERROR("Can't accept %s: %d\n", libcfs_nid2str(nid), rc);
- rej.ibr_version = version;
- rej.ibr_why = IBLND_REJECT_FATAL;
+ rc = rdma_accept(cmid, &cp);
+ if (rc != 0) {
+ CERROR("Can't accept %s: %d\n", libcfs_nid2str(nid), rc);
+ rej.ibr_version = version;
+ rej.ibr_why = IBLND_REJECT_FATAL;
- kiblnd_reject(cmid, &rej);
- kiblnd_connreq_done(conn, rc);
- kiblnd_conn_decref(conn);
- }
+ kiblnd_reject(cmid, &rej);
+ kiblnd_connreq_done(conn, rc);
+ kiblnd_conn_decref(conn);
+ }
- lnet_ni_decref(ni);
- return 0;
+ lnet_ni_decref(ni);
+ return 0;
failed:
if (ni != NULL) {
}
write_lock_irqsave(glock, flags);
- /* retry connection if it's still needed and no other connection
- * attempts (active or passive) are in progress
- * NB: reconnect is still needed even when ibp_tx_queue is
- * empty if ibp_version != version because reconnect may be
- * initiated by kiblnd_query() */
+ /* retry connection if it's still needed and no other connection
+ * attempts (active or passive) are in progress
+ * NB: reconnect is still needed even when ibp_tx_queue is
+ * empty if ibp_version != version because reconnect may be
+ * initiated.
+ */
reconnect = (!list_empty(&peer_ni->ibp_tx_queue) ||
peer_ni->ibp_version != version) &&
peer_ni->ibp_connecting &&
goto out;
}
- switch (why) {
- default:
- reason = "Unknown";
- break;
+ switch (why) {
+ default:
+ reason = "Unknown";
+ break;
case IBLND_REJECT_RDMA_FRAGS: {
- struct lnet_ioctl_config_o2iblnd_tunables *tunables;
-
if (!cp) {
reason = "can't negotiate max frags";
goto out;
}
- tunables = &peer_ni->ibp_ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib;
-#ifdef HAVE_IB_GET_DMA_MR
- /*
- * This check only makes sense if the kernel supports global
- * memory registration. Otherwise, map_on_demand will never == 0
- */
- if (!tunables->lnd_map_on_demand) {
- reason = "map_on_demand must be enabled";
- goto out;
- }
-#endif
+
if (conn->ibc_max_frags <= frag_num) {
reason = "unsupported max frags";
goto out;
case IBLND_REJECT_CONN_UNCOMPAT:
reason = "version negotiation";
break;
-
- case IBLND_REJECT_INVALID_SRV_ID:
- reason = "invalid service id";
- break;
}
conn->ibc_reconnect = 1;
kiblnd_rejected(struct kib_conn *conn, int reason, void *priv, int priv_nob)
{
struct kib_peer_ni *peer_ni = conn->ibc_peer;
+ int status = -ECONNREFUSED;
LASSERT (!in_interrupt());
LASSERT (conn->ibc_state == IBLND_CONN_ACTIVE_CONNECT);
IBLND_REJECT_CONN_STALE, NULL);
break;
- case IB_CM_REJ_INVALID_SERVICE_ID:
- kiblnd_check_reconnect(conn, IBLND_MSG_VERSION, 0,
- IBLND_REJECT_INVALID_SRV_ID, NULL);
- CNETERR("%s rejected: no listener at %d\n",
- libcfs_nid2str(peer_ni->ibp_nid),
- *kiblnd_tunables.kib_service);
- break;
+ case IB_CM_REJ_INVALID_SERVICE_ID:
+ status = -EHOSTUNREACH;
+ CNETERR("%s rejected: no listener at %d\n",
+ libcfs_nid2str(peer_ni->ibp_nid),
+ *kiblnd_tunables.kib_service);
+ break;
- case IB_CM_REJ_CONSUMER_DEFINED:
+ case IB_CM_REJ_CONSUMER_DEFINED:
if (priv_nob >= offsetof(struct kib_rej, ibr_padding)) {
struct kib_rej *rej = priv;
struct kib_connparams *cp = NULL;
- int flip = 0;
- __u64 incarnation = -1;
-
- /* NB. default incarnation is -1 because:
- * a) V1 will ignore dst incarnation in connreq.
- * b) V2 will provide incarnation while rejecting me,
- * -1 will be overwrote.
- *
- * if I try to connect to a V1 peer_ni with V2 protocol,
- * it rejected me then upgrade to V2, I have no idea
- * about the upgrading and try to reconnect with V1,
- * in this case upgraded V2 can find out I'm trying to
- * talk to the old guy and reject me(incarnation is -1).
- */
-
- if (rej->ibr_magic == __swab32(IBLND_MSG_MAGIC) ||
- rej->ibr_magic == __swab32(LNET_PROTO_MAGIC)) {
- __swab32s(&rej->ibr_magic);
- __swab16s(&rej->ibr_version);
- flip = 1;
- }
+ bool flip = false;
+ __u64 incarnation = -1;
+
+ /* NB. default incarnation is -1 because:
+ * a) V1 will ignore dst incarnation in connreq.
+ * b) V2 will provide incarnation while rejecting me,
+ * -1 will be overwrote.
+ *
+ * if I try to connect to a V1 peer_ni with V2 protocol,
+ * it rejected me then upgrade to V2, I have no idea
+ * about the upgrading and try to reconnect with V1,
+ * in this case upgraded V2 can find out I'm trying to
+ * talk to the old guy and reject me(incarnation is -1).
+ */
+
+ if (rej->ibr_magic == __swab32(IBLND_MSG_MAGIC) ||
+ rej->ibr_magic == __swab32(LNET_PROTO_MAGIC)) {
+ __swab32s(&rej->ibr_magic);
+ __swab16s(&rej->ibr_version);
+ flip = true;
+ }
if (priv_nob >= sizeof(struct kib_rej) &&
- rej->ibr_version > IBLND_MSG_VERSION_1) {
- /* priv_nob is always 148 in current version
- * of OFED, so we still need to check version.
- * (define of IB_CM_REJ_PRIVATE_DATA_SIZE) */
- cp = &rej->ibr_cp;
-
- if (flip) {
- __swab64s(&rej->ibr_incarnation);
- __swab16s(&cp->ibcp_queue_depth);
- __swab16s(&cp->ibcp_max_frags);
- __swab32s(&cp->ibcp_max_msg_size);
- }
-
- incarnation = rej->ibr_incarnation;
- }
-
- if (rej->ibr_magic != IBLND_MSG_MAGIC &&
- rej->ibr_magic != LNET_PROTO_MAGIC) {
- CERROR("%s rejected: consumer defined fatal error\n",
- libcfs_nid2str(peer_ni->ibp_nid));
- break;
- }
-
- if (rej->ibr_version != IBLND_MSG_VERSION &&
- rej->ibr_version != IBLND_MSG_VERSION_1) {
- CERROR("%s rejected: o2iblnd version %x error\n",
- libcfs_nid2str(peer_ni->ibp_nid),
- rej->ibr_version);
- break;
- }
-
- if (rej->ibr_why == IBLND_REJECT_FATAL &&
- rej->ibr_version == IBLND_MSG_VERSION_1) {
- CDEBUG(D_NET, "rejected by old version peer_ni %s: %x\n",
- libcfs_nid2str(peer_ni->ibp_nid), rej->ibr_version);
-
- if (conn->ibc_version != IBLND_MSG_VERSION_1)
- rej->ibr_why = IBLND_REJECT_CONN_UNCOMPAT;
- }
-
- switch (rej->ibr_why) {
- case IBLND_REJECT_CONN_RACE:
- case IBLND_REJECT_CONN_STALE:
- case IBLND_REJECT_CONN_UNCOMPAT:
+ rej->ibr_version > IBLND_MSG_VERSION_1) {
+ /* priv_nob is always 148 in current version
+ * of OFED, so we still need to check version.
+ * (define of IB_CM_REJ_PRIVATE_DATA_SIZE)
+ */
+ cp = &rej->ibr_cp;
+
+ if (flip) {
+ __swab64s(&rej->ibr_incarnation);
+ __swab16s(&cp->ibcp_queue_depth);
+ __swab16s(&cp->ibcp_max_frags);
+ __swab32s(&cp->ibcp_max_msg_size);
+ }
+
+ incarnation = rej->ibr_incarnation;
+ }
+
+ if (rej->ibr_magic != IBLND_MSG_MAGIC &&
+ rej->ibr_magic != LNET_PROTO_MAGIC) {
+ CERROR("%s rejected: consumer defined fatal error\n",
+ libcfs_nid2str(peer_ni->ibp_nid));
+ break;
+ }
+
+ if (rej->ibr_version != IBLND_MSG_VERSION &&
+ rej->ibr_version != IBLND_MSG_VERSION_1) {
+ CERROR("%s rejected: o2iblnd version %x error\n",
+ libcfs_nid2str(peer_ni->ibp_nid),
+ rej->ibr_version);
+ break;
+ }
+
+ if (rej->ibr_why == IBLND_REJECT_FATAL &&
+ rej->ibr_version == IBLND_MSG_VERSION_1) {
+ CDEBUG(D_NET, "rejected by old version peer_ni %s: %x\n",
+ libcfs_nid2str(peer_ni->ibp_nid),
+ rej->ibr_version);
+
+ if (conn->ibc_version != IBLND_MSG_VERSION_1)
+ rej->ibr_why = IBLND_REJECT_CONN_UNCOMPAT;
+ }
+
+ switch (rej->ibr_why) {
+ case IBLND_REJECT_CONN_RACE:
+ case IBLND_REJECT_CONN_STALE:
+ case IBLND_REJECT_CONN_UNCOMPAT:
case IBLND_REJECT_MSG_QUEUE_SIZE:
case IBLND_REJECT_RDMA_FRAGS:
kiblnd_check_reconnect(conn, rej->ibr_version,
- incarnation, rej->ibr_why, cp);
- break;
-
- case IBLND_REJECT_NO_RESOURCES:
- CERROR("%s rejected: o2iblnd no resources\n",
- libcfs_nid2str(peer_ni->ibp_nid));
- break;
-
- case IBLND_REJECT_FATAL:
- CERROR("%s rejected: o2iblnd fatal error\n",
- libcfs_nid2str(peer_ni->ibp_nid));
- break;
-
- default:
- CERROR("%s rejected: o2iblnd reason %d\n",
- libcfs_nid2str(peer_ni->ibp_nid),
- rej->ibr_why);
- break;
- }
- break;
- }
- /* fall through */
- default:
- CNETERR("%s rejected: reason %d, size %d\n",
- libcfs_nid2str(peer_ni->ibp_nid), reason, priv_nob);
- break;
- }
+ incarnation,
+ rej->ibr_why, cp);
+ break;
+
+ case IBLND_REJECT_NO_RESOURCES:
+ CERROR("%s rejected: o2iblnd no resources\n",
+ libcfs_nid2str(peer_ni->ibp_nid));
+ break;
+
+ case IBLND_REJECT_FATAL:
+ CERROR("%s rejected: o2iblnd fatal error\n",
+ libcfs_nid2str(peer_ni->ibp_nid));
+ break;
+
+ default:
+ CERROR("%s rejected: o2iblnd reason %d\n",
+ libcfs_nid2str(peer_ni->ibp_nid),
+ rej->ibr_why);
+ break;
+ }
+ break;
+ }
+ /* fall through */
+ default:
+ CNETERR("%s rejected: reason %d, size %d\n",
+ libcfs_nid2str(peer_ni->ibp_nid), reason, priv_nob);
+ break;
+ }
- kiblnd_connreq_done(conn, -ECONNREFUSED);
+ kiblnd_connreq_done(conn, status);
}
static void
}
read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- if (msg->ibm_dstnid == ni->ni_nid &&
+ if (msg->ibm_dstnid == lnet_nid_to_nid4(&ni->ni_nid) &&
msg->ibm_dststamp == net->ibn_incarnation)
rc = 0;
else
LASSERT(cmid->context == (void *)conn);
LASSERT(conn->ibc_cmid == cmid);
-
- rc = rdma_connect(cmid, &cp);
+ rc = rdma_connect_locked(cmid, &cp);
if (rc != 0) {
CERROR("Can't connect to %s: %d\n",
libcfs_nid2str(peer_ni->ibp_nid), rc);
rc = event->status;
} else {
rc = rdma_resolve_route(
- cmid, lnet_get_lnd_timeout() * 1000);
+ cmid, kiblnd_timeout() * 1000);
if (rc == 0) {
struct kib_net *net = peer_ni->ibp_ni->ni_data;
struct kib_dev *dev = net->ibn_dev;
kiblnd_check_txs_locked(struct kib_conn *conn, struct list_head *txs)
{
struct kib_tx *tx;
- struct list_head *ttmp;
-
- list_for_each(ttmp, txs) {
- tx = list_entry(ttmp, struct kib_tx, tx_list);
+ list_for_each_entry(tx, txs, tx_list) {
if (txs != &conn->ibc_active_txs) {
LASSERT(tx->tx_queued);
} else {
}
if (ktime_compare(ktime_get(), tx->tx_deadline) >= 0) {
- CERROR("Timed out tx: %s, %lld seconds\n",
+ CERROR("Timed out tx: %s(WSQ:%d%d%d), %lld seconds\n",
kiblnd_queue2str(conn, txs),
+ tx->tx_waiting, tx->tx_sending, tx->tx_queued,
+ kiblnd_timeout() +
ktime_ms_delta(ktime_get(),
tx->tx_deadline) / MSEC_PER_SEC);
return 1;
static void
kiblnd_check_conns (int idx)
{
- struct list_head closes = LIST_HEAD_INIT(closes);
- struct list_head checksends = LIST_HEAD_INIT(checksends);
- struct list_head timedout_txs = LIST_HEAD_INIT(timedout_txs);
- struct list_head *peers = &kiblnd_data.kib_peers[idx];
- struct list_head *ptmp;
+ LIST_HEAD(closes);
+ LIST_HEAD(checksends);
+ LIST_HEAD(timedout_txs);
+ struct hlist_head *peers = &kiblnd_data.kib_peers[idx];
struct kib_peer_ni *peer_ni;
- struct kib_conn *conn;
+ struct kib_conn *conn;
struct kib_tx *tx, *tx_tmp;
- struct list_head *ctmp;
- unsigned long flags;
+ unsigned long flags;
/* NB. We expect to have a look at all the peers and not find any
* RDMAs to time out, so we just use a shared lock while we
- * take a look... */
+ * take a look...
+ */
write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
- list_for_each(ptmp, peers) {
- peer_ni = list_entry(ptmp, struct kib_peer_ni, ibp_list);
-
+ hlist_for_each_entry(peer_ni, peers, ibp_list) {
/* Check tx_deadline */
list_for_each_entry_safe(tx, tx_tmp, &peer_ni->ibp_tx_queue, tx_list) {
if (ktime_compare(ktime_get(), tx->tx_deadline) >= 0) {
}
}
- list_for_each(ctmp, &peer_ni->ibp_conns) {
+ list_for_each_entry(conn, &peer_ni->ibp_conns, ibc_list) {
int timedout;
int sendnoop;
- conn = list_entry(ctmp, struct kib_conn, ibc_list);
-
LASSERT(conn->ibc_state == IBLND_CONN_ESTABLISHED);
spin_lock(&conn->ibc_lock);
}
if (timedout) {
- CERROR("Timed out RDMA with %s (%lld): "
- "c: %u, oc: %u, rc: %u\n",
+ CERROR("Timed out RDMA with %s (%lld): c: %u, oc: %u, rc: %u\n",
libcfs_nid2str(peer_ni->ibp_nid),
- ktime_get_seconds() - peer_ni->ibp_last_alive,
+ ktime_get_seconds()
+ - peer_ni->ibp_last_alive,
conn->ibc_credits,
conn->ibc_outstanding_credits,
conn->ibc_reserved_credits);
if (!list_empty(&timedout_txs))
kiblnd_txlist_done(&timedout_txs, -ETIMEDOUT,
- LNET_MSG_STATUS_LOCAL_TIMEOUT);
+ LNET_MSG_STATUS_NETWORK_TIMEOUT);
/* Handle timeout by closing the whole
* connection. We can only be sure RDMA activity
- * has ceased once the QP has been modified. */
- while (!list_empty(&closes)) {
- conn = list_entry(closes.next,
- struct kib_conn, ibc_connd_list);
+ * has ceased once the QP has been modified.
+ */
+ while ((conn = list_first_entry_or_null(&closes,
+ struct kib_conn,
+ ibc_connd_list)) != NULL) {
list_del(&conn->ibc_connd_list);
kiblnd_close_conn(conn, -ETIMEDOUT);
kiblnd_conn_decref(conn);
/* In case we have enough credits to return via a
* NOOP, but there were no non-blocking tx descs
- * free to do it last time... */
- while (!list_empty(&checksends)) {
- conn = list_entry(checksends.next,
- struct kib_conn, ibc_connd_list);
+ * free to do it last time...
+ */
+ while ((conn = list_first_entry_or_null(&checksends,
+ struct kib_conn,
+ ibc_connd_list)) != NULL) {
list_del(&conn->ibc_connd_list);
spin_lock(&conn->ibc_lock);
int
kiblnd_connd (void *arg)
{
- spinlock_t *lock= &kiblnd_data.kib_connd_lock;
+ spinlock_t *lock = &kiblnd_data.kib_connd_lock;
wait_queue_entry_t wait;
- unsigned long flags;
+ unsigned long flags;
struct kib_conn *conn;
- int timeout;
- int i;
- int dropped_lock;
- int peer_index = 0;
- unsigned long deadline = jiffies;
-
- cfs_block_allsigs();
+ int timeout;
+ int i;
+ bool dropped_lock;
+ int peer_index = 0;
+ unsigned long deadline = jiffies;
- init_waitqueue_entry(&wait, current);
+ init_wait(&wait);
kiblnd_data.kib_connd = current;
spin_lock_irqsave(lock, flags);
while (!kiblnd_data.kib_shutdown) {
int reconn = 0;
- dropped_lock = 0;
+ dropped_lock = false;
- if (!list_empty(&kiblnd_data.kib_connd_zombies)) {
+ conn = list_first_entry_or_null(&kiblnd_data.kib_connd_zombies,
+ struct kib_conn, ibc_list);
+ if (conn) {
struct kib_peer_ni *peer_ni = NULL;
- conn = list_entry(kiblnd_data.kib_connd_zombies.next,
- struct kib_conn, ibc_list);
list_del(&conn->ibc_list);
if (conn->ibc_reconnect) {
peer_ni = conn->ibc_peer;
}
spin_unlock_irqrestore(lock, flags);
- dropped_lock = 1;
+ dropped_lock = true;
kiblnd_destroy_conn(conn);
&kiblnd_data.kib_reconn_wait);
}
- if (!list_empty(&kiblnd_data.kib_connd_conns)) {
- conn = list_entry(kiblnd_data.kib_connd_conns.next,
- struct kib_conn, ibc_list);
+ conn = list_first_entry_or_null(&kiblnd_data.kib_connd_conns,
+ struct kib_conn, ibc_list);
+ if (conn) {
+ int wait;
+
list_del(&conn->ibc_list);
spin_unlock_irqrestore(lock, flags);
- dropped_lock = 1;
+ dropped_lock = true;
kiblnd_disconnect_conn(conn);
- kiblnd_conn_decref(conn);
+ wait = conn->ibc_waits;
+ if (wait == 0) /* keep ref for connd_wait, see below */
+ kiblnd_conn_decref(conn);
spin_lock_irqsave(lock, flags);
- }
+
+ if (wait)
+ list_add_tail(&conn->ibc_list,
+ &kiblnd_data.kib_connd_waits);
+ }
while (reconn < KIB_RECONN_BREAK) {
if (kiblnd_data.kib_reconn_sec !=
&kiblnd_data.kib_reconn_list);
}
- if (list_empty(&kiblnd_data.kib_reconn_list))
+ conn = list_first_entry_or_null(&kiblnd_data.kib_reconn_list,
+ struct kib_conn, ibc_list);
+ if (!conn)
break;
- conn = list_entry(kiblnd_data.kib_reconn_list.next,
- struct kib_conn, ibc_list);
list_del(&conn->ibc_list);
spin_unlock_irqrestore(lock, flags);
- dropped_lock = 1;
+ dropped_lock = true;
reconn += kiblnd_reconnect_peer(conn->ibc_peer);
kiblnd_peer_decref(conn->ibc_peer);
spin_lock_irqsave(lock, flags);
}
- /* careful with the jiffy wrap... */
- timeout = (int)(deadline - jiffies);
- if (timeout <= 0) {
- const int n = 4;
- const int p = 1;
- int chunk = kiblnd_data.kib_peer_hash_size;
- unsigned int lnd_timeout;
-
+ conn = list_first_entry_or_null(&kiblnd_data.kib_connd_waits,
+ struct kib_conn, ibc_list);
+ if (conn) {
+ list_del(&conn->ibc_list);
spin_unlock_irqrestore(lock, flags);
- dropped_lock = 1;
- /* Time to check for RDMA timeouts on a few more
- * peers: I do checks every 'p' seconds on a
- * proportion of the peer_ni table and I need to check
- * every connection 'n' times within a timeout
- * interval, to ensure I detect a timeout on any
- * connection within (n+1)/n times the timeout
- * interval. */
+ dropped_lock = kiblnd_tx_may_discard(conn);
+ if (dropped_lock)
+ kiblnd_conn_decref(conn);
- lnd_timeout = lnet_get_lnd_timeout();
+ spin_lock_irqsave(lock, flags);
+ if (!dropped_lock)
+ list_add_tail(&conn->ibc_list,
+ &kiblnd_data.kib_connd_waits);
+ }
+
+ /* careful with the jiffy wrap... */
+ timeout = (int)(deadline - jiffies);
+ if (timeout <= 0) {
+ const int n = 4;
+ const int p = 1;
+ int chunk = HASH_SIZE(kiblnd_data.kib_peers);
+ unsigned int lnd_timeout;
+
+ spin_unlock_irqrestore(lock, flags);
+ dropped_lock = true;
+
+ /* Time to check for RDMA timeouts on a few more
+ * peers: I do checks every 'p' seconds on a
+ * proportion of the peer_ni table and I need to check
+ * every connection 'n' times within a timeout
+ * interval, to ensure I detect a timeout on any
+ * connection within (n+1)/n times the timeout
+ * interval.
+ */
+
+ lnd_timeout = kiblnd_timeout();
if (lnd_timeout > n * p)
chunk = (chunk * n * p) / lnd_timeout;
if (chunk == 0)
for (i = 0; i < chunk; i++) {
kiblnd_check_conns(peer_index);
peer_index = (peer_index + 1) %
- kiblnd_data.kib_peer_hash_size;
+ HASH_SIZE(kiblnd_data.kib_peers);
}
- deadline += msecs_to_jiffies(p * MSEC_PER_SEC);
+ deadline += cfs_time_seconds(p);
spin_lock_irqsave(lock, flags);
}
schedule_timeout(timeout);
- set_current_state(TASK_RUNNING);
remove_wait_queue(&kiblnd_data.kib_connd_waitq, &wait);
spin_lock_irqsave(lock, flags);
}
{
struct kib_conn *conn = arg;
- switch (event->event) {
- case IB_EVENT_COMM_EST:
- CDEBUG(D_NET, "%s established\n",
- libcfs_nid2str(conn->ibc_peer->ibp_nid));
+ switch (event->event) {
+ case IB_EVENT_COMM_EST:
+ CDEBUG(D_NET, "%s established\n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid));
/* We received a packet but connection isn't established
* probably handshake packet was lost, so free to
* force make connection established */
rdma_notify(conn->ibc_cmid, IB_EVENT_COMM_EST);
- return;
+ return;
- default:
- CERROR("%s: Async QP event type %d\n",
- libcfs_nid2str(conn->ibc_peer->ibp_nid), event->event);
- return;
- }
+ case IB_EVENT_PORT_ERR:
+ case IB_EVENT_DEVICE_FATAL:
+ CERROR("Fatal device error for NI %s\n",
+ libcfs_nidstr(&conn->ibc_peer->ibp_ni->ni_nid));
+ atomic_set(&conn->ibc_peer->ibp_ni->ni_fatal_error_on, 1);
+ return;
+
+ case IB_EVENT_PORT_ACTIVE:
+ CERROR("Port reactivated for NI %s\n",
+ libcfs_nidstr(&conn->ibc_peer->ibp_ni->ni_nid));
+ atomic_set(&conn->ibc_peer->ibp_ni->ni_fatal_error_on, 0);
+ return;
+
+ default:
+ CERROR("%s: Async QP event type %d\n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid), event->event);
+ return;
+ }
}
static void
int
kiblnd_scheduler(void *arg)
{
- long id = (long)arg;
- struct kib_sched_info *sched;
+ long id = (long)arg;
+ struct kib_sched_info *sched;
struct kib_conn *conn;
- wait_queue_entry_t wait;
- unsigned long flags;
- struct ib_wc wc;
- int did_something;
- int busy_loops = 0;
- int rc;
-
- cfs_block_allsigs();
+ wait_queue_entry_t wait;
+ unsigned long flags;
+ struct ib_wc wc;
+ bool did_something;
+ int rc;
- init_waitqueue_entry(&wait, current);
+ init_wait(&wait);
sched = kiblnd_data.kib_scheds[KIB_THREAD_CPT(id)];
rc = cfs_cpt_bind(lnet_cpt_table(), sched->ibs_cpt);
if (rc != 0) {
- CWARN("Unable to bind on CPU partition %d, please verify "
- "whether all CPUs are healthy and reload modules if "
- "necessary, otherwise your system might under risk of "
- "low performance\n", sched->ibs_cpt);
+ CWARN("Unable to bind on CPU partition %d, please verify whether all CPUs are healthy and reload modules if necessary, otherwise your system might under risk of low performance\n", sched->ibs_cpt);
}
spin_lock_irqsave(&sched->ibs_lock, flags);
while (!kiblnd_data.kib_shutdown) {
- if (busy_loops++ >= IBLND_RESCHED) {
+ if (need_resched()) {
spin_unlock_irqrestore(&sched->ibs_lock, flags);
cond_resched();
- busy_loops = 0;
spin_lock_irqsave(&sched->ibs_lock, flags);
}
- did_something = 0;
+ did_something = false;
- if (!list_empty(&sched->ibs_conns)) {
- conn = list_entry(sched->ibs_conns.next,
- struct kib_conn, ibc_sched_list);
+ conn = list_first_entry_or_null(&sched->ibs_conns,
+ struct kib_conn,
+ ibc_sched_list);
+ if (conn) {
/* take over kib_sched_conns' ref on conn... */
LASSERT(conn->ibc_scheduled);
list_del(&conn->ibc_sched_list);
wc.wr_id = IBLND_WID_INVAL;
- rc = ib_poll_cq(conn->ibc_cq, 1, &wc);
- if (rc == 0) {
- rc = ib_req_notify_cq(conn->ibc_cq,
- IB_CQ_NEXT_COMP);
- if (rc < 0) {
- CWARN("%s: ib_req_notify_cq failed: %d, "
- "closing connection\n",
- libcfs_nid2str(conn->ibc_peer->ibp_nid), rc);
- kiblnd_close_conn(conn, -EIO);
- kiblnd_conn_decref(conn);
+ rc = ib_poll_cq(conn->ibc_cq, 1, &wc);
+ if (rc == 0) {
+ rc = ib_req_notify_cq(conn->ibc_cq,
+ IB_CQ_NEXT_COMP);
+ if (rc < 0) {
+ CWARN("%s: ib_req_notify_cq failed: %d, closing connection\n",
+ libcfs_nid2str(conn->ibc_peer->ibp_nid), rc);
+ kiblnd_close_conn(conn, -EIO);
+ kiblnd_conn_decref(conn);
spin_lock_irqsave(&sched->ibs_lock,
- flags);
+ flags);
continue;
}
}
if (rc < 0) {
- CWARN("%s: ib_poll_cq failed: %d, "
- "closing connection\n",
+ CWARN("%s: ib_poll_cq failed: %d, closing connection\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid),
rc);
kiblnd_close_conn(conn, -EIO);
/* +1 ref for sched_conns */
kiblnd_conn_addref(conn);
list_add_tail(&conn->ibc_sched_list,
- &sched->ibs_conns);
+ &sched->ibs_conns);
if (waitqueue_active(&sched->ibs_waitq))
wake_up(&sched->ibs_waitq);
} else {
kiblnd_complete(&wc);
spin_lock_irqsave(&sched->ibs_lock, flags);
- }
+ }
- kiblnd_conn_decref(conn); /* ...drop my ref from above */
- did_something = 1;
- }
+ kiblnd_conn_decref(conn); /* ..drop my ref from above */
+ did_something = true;
+ }
- if (did_something)
- continue;
+ if (did_something)
+ continue;
set_current_state(TASK_INTERRUPTIBLE);
add_wait_queue_exclusive(&sched->ibs_waitq, &wait);
spin_unlock_irqrestore(&sched->ibs_lock, flags);
schedule();
- busy_loops = 0;
remove_wait_queue(&sched->ibs_waitq, &wait);
set_current_state(TASK_RUNNING);
int
kiblnd_failover_thread(void *arg)
{
- rwlock_t *glock = &kiblnd_data.kib_global_lock;
+ rwlock_t *glock = &kiblnd_data.kib_global_lock;
struct kib_dev *dev;
+ struct net *ns = arg;
wait_queue_entry_t wait;
- unsigned long flags;
- int rc;
+ unsigned long flags;
+ int rc;
LASSERT(*kiblnd_tunables.kib_dev_failover != 0);
- cfs_block_allsigs();
-
- init_waitqueue_entry(&wait, current);
+ init_wait(&wait);
write_lock_irqsave(glock, flags);
- while (!kiblnd_data.kib_shutdown) {
- int do_failover = 0;
- int long_sleep;
+ while (!kiblnd_data.kib_shutdown) {
+ bool do_failover = false;
+ int long_sleep;
list_for_each_entry(dev, &kiblnd_data.kib_failed_devs,
- ibd_fail_list) {
+ ibd_fail_list) {
if (ktime_get_seconds() < dev->ibd_next_failover)
- continue;
- do_failover = 1;
- break;
- }
+ continue;
+ do_failover = true;
+ break;
+ }
- if (do_failover) {
+ if (do_failover) {
list_del_init(&dev->ibd_fail_list);
- dev->ibd_failover = 1;
+ dev->ibd_failover = 1;
write_unlock_irqrestore(glock, flags);
- rc = kiblnd_dev_failover(dev);
+ rc = kiblnd_dev_failover(dev, ns);
write_lock_irqsave(glock, flags);
- LASSERT (dev->ibd_failover);
- dev->ibd_failover = 0;
- if (rc >= 0) { /* Device is OK or failover succeed */
+ LASSERT(dev->ibd_failover);
+ dev->ibd_failover = 0;
+ if (rc >= 0) { /* Device is OK or failover succeed */
dev->ibd_next_failover = ktime_get_seconds() + 3;
- continue;
- }
+ continue;
+ }
- /* failed to failover, retry later */
+ /* failed to failover, retry later */
dev->ibd_next_failover = ktime_get_seconds() +
- min(dev->ibd_failed_failover, 10);
- if (kiblnd_dev_can_failover(dev)) {
+ min(dev->ibd_failed_failover, 10);
+ if (kiblnd_dev_can_failover(dev)) {
list_add_tail(&dev->ibd_fail_list,
- &kiblnd_data.kib_failed_devs);
- }
+ &kiblnd_data.kib_failed_devs);
+ }
- continue;
- }
+ continue;
+ }
- /* long sleep if no more pending failover */
+ /* long sleep if no more pending failover */
long_sleep = list_empty(&kiblnd_data.kib_failed_devs);
set_current_state(TASK_INTERRUPTIBLE);
write_unlock_irqrestore(glock, flags);
rc = schedule_timeout(long_sleep ? cfs_time_seconds(10) :
- cfs_time_seconds(1));
+ cfs_time_seconds(1));
set_current_state(TASK_RUNNING);
remove_wait_queue(&kiblnd_data.kib_failover_waitq, &wait);
write_lock_irqsave(glock, flags);
- if (!long_sleep || rc != 0)
- continue;
+ if (!long_sleep || rc != 0)
+ continue;
- /* have a long sleep, routine check all active devices,
- * we need checking like this because if there is not active
- * connection on the dev and no SEND from local, we may listen
- * on wrong HCA for ever while there is a bonding failover */
+ /* have a long sleep, routine check all active devices,
+ * we need checking like this because if there is not active
+ * connection on the dev and no SEND from local, we may listen
+ * on wrong HCA for ever while there is a bonding failover
+ */
list_for_each_entry(dev, &kiblnd_data.kib_devs, ibd_list) {
- if (kiblnd_dev_can_failover(dev)) {
+ if (kiblnd_dev_can_failover(dev)) {
list_add_tail(&dev->ibd_fail_list,
- &kiblnd_data.kib_failed_devs);
- }
- }
- }
+ &kiblnd_data.kib_failed_devs);
+ }
+ }
+ }
write_unlock_irqrestore(glock, flags);
- kiblnd_thread_fini();
- return 0;
+ kiblnd_thread_fini();
+ return 0;
}