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) {
* 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;
/*
* from the first send; hence the ++ rather than = below. */
tx->tx_sending++;
list_add(&tx->tx_list, &conn->ibc_active_txs);
- tx->tx_on_activeq = ktime_get();
-
- /* I'm still holding ibc_lock! */
- if (conn->ibc_state != IBLND_CONN_ESTABLISHED) {
- rc = -ECONNABORTED;
- } else if (tx->tx_pool->tpo_pool.po_failed ||
- conn->ibc_hdev != tx->tx_pool->tpo_hdev) {
- /* close_conn will launch failover */
- rc = -ENETDOWN;
- } else {
+
+ /* I'm still holding ibc_lock! */
+ if (conn->ibc_state != IBLND_CONN_ESTABLISHED) {
+ rc = -ECONNABORTED;
+ } else if (tx->tx_pool->tpo_pool.po_failed ||
+ conn->ibc_hdev != tx->tx_pool->tpo_hdev) {
+ /* 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;
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
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
kiblnd_thread_start(int (*fn)(void *arg), void *arg, char *name)
{
- struct task_struct *task = kthread_run(fn, arg, name);
+ struct task_struct *task = kthread_run(fn, arg, "%s", name);
if (IS_ERR(task))
return PTR_ERR(task);
kiblnd_abort_txs(struct kib_conn *conn, struct list_head *txs)
{
LIST_HEAD(zombies);
- struct list_head *tmp;
- struct list_head *nxt;
+ 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_move(&tx->tx_list, &zombies);
+ } else {
+ /* keep tx until cq destroy */
+ list_move(&tx->tx_list, &conn->ibc_zombie_txs);
+ conn->ibc_waits ++;
}
}
kiblnd_txlist_done(&zombies, -ECONNABORTED, LNET_MSG_STATUS_OK);
}
+static int
+kiblnd_tx_may_discard(struct kib_conn *conn)
+{
+ int rc = 0;
+ 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 = 1;
+ }
+ }
+ }
+
+ spin_unlock(&conn->ibc_lock);
+ return rc;
+}
+
static void
kiblnd_finalise_conn(struct kib_conn *conn)
{
CNETERR("Deleting messages for %s: connection failed\n",
libcfs_nid2str(peer_ni->ibp_nid));
- kiblnd_txlist_done(&zombies, error,
- LNET_MSG_STATUS_LOCAL_DROPPED);
+ if (error == -EHOSTUNREACH || error == -ETIMEDOUT)
+ kiblnd_txlist_done(&zombies, error,
+ LNET_MSG_STATUS_NETWORK_TIMEOUT);
+ else
+ kiblnd_txlist_done(&zombies, error,
+ LNET_MSG_STATUS_LOCAL_DROPPED);
}
static void
(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);
+ /* reset retry count */
+ peer_ni->ibp_retries = 0;
+
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... */
{
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));
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),
+ CERROR("Can't accept conn from %s on %s (%s:%d:%pI4h): bad dst nid %s\n", libcfs_nid2str(nid),
+ ni ? libcfs_nid2str(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,
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) {
goto out;
}
- switch (why) {
- default:
- reason = "Unknown";
- break;
+ if (peer_ni->ibp_retries > *kiblnd_tunables.kib_retry_count) {
+ reason = "retry count exceeded due to no listener";
+ goto out;
+ }
+
+ switch (why) {
+ default:
+ reason = "Unknown";
+ break;
case IBLND_REJECT_RDMA_FRAGS: {
struct lnet_ioctl_config_o2iblnd_tunables *tunables;
IBLND_REJECT_CONN_STALE, NULL);
break;
- case IB_CM_REJ_INVALID_SERVICE_ID:
+ case IB_CM_REJ_INVALID_SERVICE_ID:
+ peer_ni->ibp_retries++;
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;
+ 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:
if (priv_nob >= offsetof(struct kib_rej, ibr_padding)) {
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);
{
struct kib_tx *tx;
struct list_head *ttmp;
- bool active_txs = strcmp(kiblnd_queue2str(conn, txs),
- "active_txs") == 0;
list_for_each(ttmp, txs) {
tx = list_entry(ttmp, struct kib_tx, tx_list);
LASSERT(tx->tx_waiting || tx->tx_sending != 0);
}
- if (ktime_compare(ktime_get(), tx->tx_deadline) < 0)
- continue;
-
- if (!active_txs) {
- CERROR("Timed out tx: %s, "
- "outstanding RDMA time: %lld sec\n",
- kiblnd_queue2str(conn, txs),
- *kiblnd_tunables.kib_timeout +
- (ktime_ms_delta(ktime_get(),
- tx->tx_deadline) / MSEC_PER_SEC));
- } else {
- CERROR("Timed out tx: %s, time in internal queue: %lld "
- "sec, time in active queue: %lld sec,"
- " outstanding RDMA time: %lld sec\n",
+ if (ktime_compare(ktime_get(), tx->tx_deadline) >= 0) {
+ CERROR("Timed out tx: %s(WSQ:%d%d%d), %lld seconds\n",
kiblnd_queue2str(conn, txs),
- ktime_ms_delta(tx->tx_deadline,
- tx->tx_on_activeq) / MSEC_PER_SEC,
+ tx->tx_waiting, tx->tx_sending, tx->tx_queued,
+ kiblnd_timeout() +
ktime_ms_delta(ktime_get(),
- tx->tx_on_activeq) / MSEC_PER_SEC,
- *kiblnd_tunables.kib_timeout +
- (ktime_ms_delta(ktime_get(),
- tx->tx_deadline) / MSEC_PER_SEC));
+ tx->tx_deadline) / MSEC_PER_SEC);
+ return 1;
}
-
- return 1;
}
return 0;
LIST_HEAD(closes);
LIST_HEAD(checksends);
LIST_HEAD(timedout_txs);
- struct list_head *peers = &kiblnd_data.kib_peers[idx];
- struct list_head *ptmp;
+ 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) {
}
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. */
+ * has ceased once the QP has been modified.
+ */
while (!list_empty(&closes)) {
conn = list_entry(closes.next,
struct kib_conn, ibc_connd_list);
/* 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... */
+ * free to do it last time...
+ */
while (!list_empty(&checksends)) {
conn = list_entry(checksends.next,
struct kib_conn, ibc_connd_list);
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;
+ int timeout;
+ int i;
+ int 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 = 0;
if (!list_empty(&kiblnd_data.kib_connd_zombies)) {
struct kib_peer_ni *peer_ni = NULL;
}
if (!list_empty(&kiblnd_data.kib_connd_conns)) {
+ int wait;
conn = list_entry(kiblnd_data.kib_connd_conns.next,
struct kib_conn, ibc_list);
list_del(&conn->ibc_list);
dropped_lock = 1;
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 !=
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;
+ if (!list_empty(&kiblnd_data.kib_connd_waits)) {
+ conn = list_entry(kiblnd_data.kib_connd_waits.next,
+ struct kib_conn, ibc_list);
+ list_del(&conn->ibc_list);
+ spin_unlock_irqrestore(lock, flags);
+
+ dropped_lock = kiblnd_tx_may_discard(conn);
+ if (dropped_lock)
+ kiblnd_conn_decref(conn);
+
+ spin_lock_irqsave(lock, flags);
+ if (dropped_lock == 0)
+ 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 = 1;
+ 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. */
+ /* 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)
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 += cfs_time_seconds(p);
unsigned long flags;
struct ib_wc wc;
int did_something;
- int busy_loops = 0;
int rc;
- init_waitqueue_entry(&wait, current);
+ init_wait(&wait);
sched = kiblnd_data.kib_scheds[KIB_THREAD_CPT(id)];
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);
}
spin_unlock_irqrestore(&sched->ibs_lock, flags);
schedule();
- busy_loops = 0;
remove_wait_queue(&sched->ibs_waitq, &wait);
set_current_state(TASK_RUNNING);
LASSERT(*kiblnd_tunables.kib_dev_failover != 0);
- init_waitqueue_entry(&wait, current);
+ init_wait(&wait);
write_lock_irqsave(glock, flags);
while (!kiblnd_data.kib_shutdown) {