/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (C) 2005 Cluster File Systems, Inc. All rights reserved.
- * Author: PJ Kirner <pjkirner@clusterfs.com>
- * E Barton <eeb@bartonsoftware.com>
+ * GPL HEADER START
*
- * This file is part of the Lustre file system, http://www.lustre.org
- * Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * This file is confidential source code owned by Cluster File Systems.
- * No viewing, modification, compilation, redistribution, or any other
- * form of use is permitted except through a signed license agreement.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
*
- * If you have not signed such an agreement, then you have no rights to
- * this file. Please destroy it immediately and contact CFS.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
*
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lnet/klnds/ptllnd/ptllnd_peer.c
+ *
+ * Author: PJ Kirner <pjkirner@clusterfs.com>
+ * Author: E Barton <eeb@bartonsoftware.com>
*/
#include "ptllnd.h"
read_lock_irqsave(g_lock, flags);
for (i = 0; i < kptllnd_data.kptl_peer_hash_size; i++) {
-
list_for_each (ptmp, &kptllnd_data.kptl_peers[i]) {
peer = list_entry(ptmp, kptl_peer_t, peer_list);
void
kptllnd_peer_add_peertable_locked (kptl_peer_t *peer)
{
- LASSERT (!kptllnd_data.kptl_shutdown);
LASSERT (kptllnd_data.kptl_n_active_peers <
kptllnd_data.kptl_expected_peers);
* in MRU order */
peer = list_entry(tmp, kptl_peer_t, peer_list);
- if (peer->peer_id.nid != pid.nid)
+ if (LNET_NIDADDR(peer->peer_id.nid) != LNET_NIDADDR(pid.nid))
continue;
LASSERT (peer->peer_id.pid != pid.pid);
}
kptl_peer_t *
-kptllnd_peer_allocate (lnet_process_id_t lpid, ptl_process_id_t ppid)
+kptllnd_peer_allocate (kptl_net_t *net, lnet_process_id_t lpid, ptl_process_id_t ppid)
{
unsigned long flags;
kptl_peer_t *peer;
memset(peer, 0, sizeof(*peer)); /* zero flags etc */
+ INIT_LIST_HEAD (&peer->peer_noops);
INIT_LIST_HEAD (&peer->peer_sendq);
INIT_LIST_HEAD (&peer->peer_activeq);
spin_lock_init (&peer->peer_lock);
peer->peer_state = PEER_STATE_ALLOCATED;
peer->peer_error = 0;
- peer->peer_last_alive = cfs_time_current();
+ peer->peer_last_alive = 0;
peer->peer_id = lpid;
peer->peer_ptlid = ppid;
peer->peer_credits = 1; /* enough for HELLO */
peer->peer_next_matchbits = PTL_RESERVED_MATCHBITS;
- peer->peer_outstanding_credits = *kptllnd_tunables.kptl_peercredits - 1;
+ peer->peer_outstanding_credits = *kptllnd_tunables.kptl_peertxcredits - 1;
peer->peer_sent_credits = 1; /* HELLO credit is implicit */
peer->peer_max_msg_size = PTLLND_MIN_BUFFER_SIZE; /* until we know better */
/* Only increase # peers under lock, to guarantee we dont grow it
* during shutdown */
- if (kptllnd_data.kptl_shutdown) {
- write_unlock_irqrestore(&kptllnd_data.kptl_peer_rw_lock,
- flags);
+ if (net->net_shutdown) {
+ write_unlock_irqrestore(&kptllnd_data.kptl_peer_rw_lock, flags);
LIBCFS_FREE(peer, sizeof(*peer));
return NULL;
}
kptllnd_data.kptl_npeers++;
write_unlock_irqrestore(&kptllnd_data.kptl_peer_rw_lock, flags);
-
return peer;
}
LASSERT (atomic_read(&peer->peer_refcount) == 0);
LASSERT (peer->peer_state == PEER_STATE_ALLOCATED ||
peer->peer_state == PEER_STATE_ZOMBIE);
+ LASSERT (list_empty(&peer->peer_noops));
LASSERT (list_empty(&peer->peer_sendq));
LASSERT (list_empty(&peer->peer_activeq));
spin_lock_irqsave(&peer->peer_lock, flags);
+ kptllnd_cancel_txlist(&peer->peer_noops, txs);
kptllnd_cancel_txlist(&peer->peer_sendq, txs);
kptllnd_cancel_txlist(&peer->peer_activeq, txs);
kptllnd_peer_notify (kptl_peer_t *peer)
{
unsigned long flags;
- time_t last_alive = 0;
+ kptl_net_t *net;
+ kptl_net_t **nets;
+ int i = 0;
+ int nnets = 0;
int error = 0;
+ time_t last_alive = 0;
spin_lock_irqsave(&peer->peer_lock, flags);
spin_unlock_irqrestore(&peer->peer_lock, flags);
- if (error != 0)
- lnet_notify (kptllnd_data.kptl_ni, peer->peer_id.nid, 0,
- last_alive);
+ if (error == 0)
+ return;
+
+ read_lock(&kptllnd_data.kptl_net_rw_lock);
+ list_for_each_entry (net, &kptllnd_data.kptl_nets, net_list)
+ nnets++;
+ read_unlock(&kptllnd_data.kptl_net_rw_lock);
+
+ if (nnets == 0) /* shutdown in progress */
+ return;
+
+ LIBCFS_ALLOC(nets, nnets * sizeof(*nets));
+ if (nets == NULL) {
+ CERROR("Failed to allocate nets[%d]\n", nnets);
+ return;
+ }
+ memset(nets, 0, nnets * sizeof(*nets));
+
+ read_lock(&kptllnd_data.kptl_net_rw_lock);
+ i = 0;
+ list_for_each_entry (net, &kptllnd_data.kptl_nets, net_list) {
+ LASSERT (i < nnets);
+ nets[i] = net;
+ kptllnd_net_addref(net);
+ i++;
+ }
+ read_unlock(&kptllnd_data.kptl_net_rw_lock);
+
+ for (i = 0; i < nnets; i++) {
+ lnet_nid_t peer_nid;
+
+ net = nets[i];
+ if (net == NULL)
+ break;
+
+ if (!net->net_shutdown) {
+ peer_nid = kptllnd_ptl2lnetnid(net->net_ni->ni_nid,
+ peer->peer_ptlid.nid);
+ lnet_notify(net->net_ni, peer_nid, 0, last_alive);
+ }
+
+ kptllnd_net_decref(net);
+ }
+
+ LIBCFS_FREE(nets, nnets * sizeof(*nets));
}
void
peer = list_entry (ptmp, kptl_peer_t, peer_list);
if (!(id.nid == LNET_NID_ANY ||
- (peer->peer_id.nid == id.nid &&
+ (LNET_NIDADDR(peer->peer_id.nid) == LNET_NIDADDR(id.nid) &&
(id.pid == LNET_PID_ANY ||
peer->peer_id.pid == id.pid))))
continue;
}
void
+kptllnd_queue_tx(kptl_peer_t *peer, kptl_tx_t *tx)
+{
+ /* CAVEAT EMPTOR: I take over caller's ref on 'tx' */
+ unsigned long flags;
+
+ spin_lock_irqsave(&peer->peer_lock, flags);
+
+ /* Ensure HELLO is sent first */
+ if (tx->tx_msg->ptlm_type == PTLLND_MSG_TYPE_NOOP)
+ list_add(&tx->tx_list, &peer->peer_noops);
+ else if (tx->tx_msg->ptlm_type == PTLLND_MSG_TYPE_HELLO)
+ list_add(&tx->tx_list, &peer->peer_sendq);
+ else
+ list_add_tail(&tx->tx_list, &peer->peer_sendq);
+
+ spin_unlock_irqrestore(&peer->peer_lock, flags);
+}
+
+
+void
kptllnd_post_tx(kptl_peer_t *peer, kptl_tx_t *tx, int nfrag)
{
/* CAVEAT EMPTOR: I take over caller's ref on 'tx' */
ptl_handle_md_t msg_mdh;
ptl_md_t md;
ptl_err_t prc;
- unsigned long flags;
LASSERT (!tx->tx_idle);
LASSERT (!tx->tx_active);
kptllnd_tx_decref(tx);
return;
}
-
- spin_lock_irqsave(&peer->peer_lock, flags);
+
tx->tx_deadline = jiffies + (*kptllnd_tunables.kptl_timeout * HZ);
tx->tx_active = 1;
tx->tx_msg_mdh = msg_mdh;
+ kptllnd_queue_tx(peer, tx);
+}
- /* Ensure HELLO is sent first */
- if (tx->tx_msg->ptlm_type == PTLLND_MSG_TYPE_HELLO)
- list_add(&tx->tx_list, &peer->peer_sendq);
- else
- list_add_tail(&tx->tx_list, &peer->peer_sendq);
+/* NB "restarts" comes from peer_sendq of a single peer */
+void
+kptllnd_restart_txs (kptl_net_t *net, lnet_process_id_t target, struct list_head *restarts)
+{
+ kptl_tx_t *tx;
+ kptl_tx_t *tmp;
+ kptl_peer_t *peer;
- spin_unlock_irqrestore(&peer->peer_lock, flags);
+ LASSERT (!list_empty(restarts));
+
+ if (kptllnd_find_target(net, target, &peer) != 0)
+ peer = NULL;
+
+ list_for_each_entry_safe (tx, tmp, restarts, tx_list) {
+ LASSERT (tx->tx_peer != NULL);
+ LASSERT (tx->tx_type == TX_TYPE_GET_REQUEST ||
+ tx->tx_type == TX_TYPE_PUT_REQUEST ||
+ tx->tx_type == TX_TYPE_SMALL_MESSAGE);
+
+ list_del_init(&tx->tx_list);
+
+ if (peer == NULL ||
+ tx->tx_msg->ptlm_type == PTLLND_MSG_TYPE_HELLO) {
+ kptllnd_tx_decref(tx);
+ continue;
+ }
+
+ LASSERT (tx->tx_msg->ptlm_type != PTLLND_MSG_TYPE_NOOP);
+ tx->tx_status = 0;
+ tx->tx_active = 1;
+ kptllnd_peer_decref(tx->tx_peer);
+ tx->tx_peer = NULL;
+ kptllnd_set_tx_peer(tx, peer);
+ kptllnd_queue_tx(peer, tx); /* takes over my ref on tx */
+ }
+
+ if (peer == NULL)
+ return;
+
+ kptllnd_peer_check_sends(peer);
+ kptllnd_peer_decref(peer);
+}
+
+static inline int
+kptllnd_peer_send_noop (kptl_peer_t *peer)
+{
+ if (!peer->peer_sent_hello ||
+ peer->peer_credits == 0 ||
+ !list_empty(&peer->peer_noops) ||
+ peer->peer_outstanding_credits < PTLLND_CREDIT_HIGHWATER)
+ return 0;
+
+ /* No tx to piggyback NOOP onto or no credit to send a tx */
+ return (list_empty(&peer->peer_sendq) || peer->peer_credits == 1);
}
void
ptl_handle_me_t meh;
kptl_tx_t *tx;
int rc;
+ int msg_type;
unsigned long flags;
LASSERT(!in_interrupt());
peer->peer_retry_noop = 0;
- if (list_empty(&peer->peer_sendq) &&
- peer->peer_outstanding_credits >= PTLLND_CREDIT_HIGHWATER &&
- peer->peer_credits != 0) {
-
+ if (kptllnd_peer_send_noop(peer)) {
/* post a NOOP to return credits */
spin_unlock_irqrestore(&peer->peer_lock, flags);
CERROR("Can't return credits to %s: can't allocate descriptor\n",
libcfs_id2str(peer->peer_id));
} else {
- kptllnd_init_msg(tx->tx_msg, PTLLND_MSG_TYPE_NOOP, 0);
+ kptllnd_init_msg(tx->tx_msg, PTLLND_MSG_TYPE_NOOP,
+ peer->peer_id, 0);
kptllnd_post_tx(peer, tx, 0);
}
peer->peer_retry_noop = (tx == NULL);
}
- while (!list_empty(&peer->peer_sendq)) {
- tx = list_entry (peer->peer_sendq.next, kptl_tx_t, tx_list);
+ for (;;) {
+ if (!list_empty(&peer->peer_noops)) {
+ LASSERT (peer->peer_sent_hello);
+ tx = list_entry(peer->peer_noops.next,
+ kptl_tx_t, tx_list);
+ } else if (!list_empty(&peer->peer_sendq)) {
+ tx = list_entry(peer->peer_sendq.next,
+ kptl_tx_t, tx_list);
+ } else {
+ /* nothing to send right now */
+ break;
+ }
LASSERT (tx->tx_active);
LASSERT (!PtlHandleIsEqual(tx->tx_msg_mdh, PTL_INVALID_HANDLE));
LASSERT (peer->peer_sent_credits >= 0);
LASSERT (peer->peer_sent_credits +
peer->peer_outstanding_credits <=
- *kptllnd_tunables.kptl_peercredits);
+ *kptllnd_tunables.kptl_peertxcredits);
LASSERT (peer->peer_credits >= 0);
- /* Ensure HELLO is sent first */
- if (!peer->peer_sent_hello) {
- if (tx->tx_msg->ptlm_type != PTLLND_MSG_TYPE_HELLO)
- break;
- peer->peer_sent_hello = 1;
- }
+ msg_type = tx->tx_msg->ptlm_type;
+
+ /* Ensure HELLO is sent first */
+ if (!peer->peer_sent_hello) {
+ LASSERT (list_empty(&peer->peer_noops));
+ if (msg_type != PTLLND_MSG_TYPE_HELLO)
+ break;
+ peer->peer_sent_hello = 1;
+ }
if (peer->peer_credits == 0) {
- CDEBUG(D_NETTRACE, "%s[%d/%d+%d]: no credits for %p\n",
+ CDEBUG(D_NETTRACE, "%s[%d/%d+%d]: no credits for %s[%p]\n",
libcfs_id2str(peer->peer_id),
peer->peer_credits,
peer->peer_outstanding_credits,
- peer->peer_sent_credits, tx);
+ peer->peer_sent_credits,
+ kptllnd_msgtype2str(msg_type), tx);
break;
}
- /* Don't use the last credit unless I've got credits to
- * return */
+ /* Last/Initial credit reserved for NOOP/HELLO */
if (peer->peer_credits == 1 &&
- peer->peer_outstanding_credits == 0) {
+ msg_type != PTLLND_MSG_TYPE_HELLO &&
+ msg_type != PTLLND_MSG_TYPE_NOOP) {
CDEBUG(D_NETTRACE, "%s[%d/%d+%d]: "
- "not using last credit for %p\n",
+ "not using last credit for %s[%p]\n",
libcfs_id2str(peer->peer_id),
peer->peer_credits,
peer->peer_outstanding_credits,
- peer->peer_sent_credits, tx);
+ peer->peer_sent_credits,
+ kptllnd_msgtype2str(msg_type), tx);
break;
}
/* Discard any NOOP I queued if I'm not at the high-water mark
* any more or more messages have been queued */
- if (tx->tx_msg->ptlm_type == PTLLND_MSG_TYPE_NOOP &&
- (!list_empty(&peer->peer_sendq) ||
- peer->peer_outstanding_credits < PTLLND_CREDIT_HIGHWATER)) {
-
+ if (msg_type == PTLLND_MSG_TYPE_NOOP &&
+ !kptllnd_peer_send_noop(peer)) {
tx->tx_active = 0;
spin_unlock_irqrestore(&peer->peer_lock, flags);
tx->tx_msg->ptlm_u.rdma.kptlrm_matchbits =
peer->peer_next_matchbits++;
}
-
+
peer->peer_sent_credits += peer->peer_outstanding_credits;
peer->peer_outstanding_credits = 0;
peer->peer_credits--;
CDEBUG(D_NETTRACE, "%s[%d/%d+%d]: %s tx=%p nob=%d cred=%d\n",
libcfs_id2str(peer->peer_id), peer->peer_credits,
peer->peer_outstanding_credits, peer->peer_sent_credits,
- kptllnd_msgtype2str(tx->tx_msg->ptlm_type),
- tx, tx->tx_msg->ptlm_nob,
+ kptllnd_msgtype2str(msg_type), tx, tx->tx_msg->ptlm_nob,
tx->tx_msg->ptlm_credits);
list_add_tail(&tx->tx_list, &peer->peer_activeq);
/* Nuke everything (including tx we were trying) */
kptllnd_peer_close(peer, -EIO);
kptllnd_tx_decref(tx);
+ kptllnd_schedule_ptltrace_dump();
}
kptl_tx_t *
kptllnd_find_timed_out_tx(kptl_peer_t *peer)
{
kptl_tx_t *tx;
- struct list_head *tmp;
+ struct list_head *ele;
- list_for_each(tmp, &peer->peer_sendq) {
- tx = list_entry(peer->peer_sendq.next, kptl_tx_t, tx_list);
+ list_for_each(ele, &peer->peer_sendq) {
+ tx = list_entry(ele, kptl_tx_t, tx_list);
if (time_after_eq(jiffies, tx->tx_deadline)) {
kptllnd_tx_addref(tx);
}
}
- list_for_each(tmp, &peer->peer_activeq) {
- tx = list_entry(peer->peer_activeq.next, kptl_tx_t, tx_list);
+ list_for_each(ele, &peer->peer_activeq) {
+ tx = list_entry(ele, kptl_tx_t, tx_list);
if (time_after_eq(jiffies, tx->tx_deadline)) {
kptllnd_tx_addref(tx);
kptllnd_peer_check_bucket (int idx, int stamp)
{
struct list_head *peers = &kptllnd_data.kptl_peers[idx];
- struct list_head *ptmp;
kptl_peer_t *peer;
- kptl_tx_t *tx;
unsigned long flags;
- int nsend;
- int nactive;
- int check_sends;
CDEBUG(D_NET, "Bucket=%d, stamp=%d\n", idx, stamp);
/* NB. Shared lock while I just look */
read_lock_irqsave(&kptllnd_data.kptl_peer_rw_lock, flags);
- list_for_each (ptmp, peers) {
- peer = list_entry (ptmp, kptl_peer_t, peer_list);
+ list_for_each_entry (peer, peers, peer_list) {
+ kptl_tx_t *tx;
+ int check_sends;
+ int c = -1, oc = -1, sc = -1;
+ int nsend = -1, nactive = -1;
+ int sent_hello = -1, state = -1;
CDEBUG(D_NET, "Peer=%s Credits=%d Outstanding=%d Send=%d\n",
libcfs_id2str(peer->peer_id), peer->peer_credits,
tx = kptllnd_find_timed_out_tx(peer);
check_sends = peer->peer_retry_noop;
+ if (tx != NULL) {
+ c = peer->peer_credits;
+ sc = peer->peer_sent_credits;
+ oc = peer->peer_outstanding_credits;
+ state = peer->peer_state;
+ sent_hello = peer->peer_sent_hello;
+ nsend = kptllnd_count_queue(&peer->peer_sendq);
+ nactive = kptllnd_count_queue(&peer->peer_activeq);
+ }
+
spin_unlock(&peer->peer_lock);
if (tx == NULL && !check_sends)
goto again;
}
- spin_lock_irqsave(&peer->peer_lock, flags);
- nsend = kptllnd_count_queue(&peer->peer_sendq);
- nactive = kptllnd_count_queue(&peer->peer_activeq);
- spin_unlock_irqrestore(&peer->peer_lock, flags);
-
LCONSOLE_ERROR_MSG(0x126, "Timing out %s: %s\n",
libcfs_id2str(peer->peer_id),
(tx->tx_tposted == 0) ?
"no free peer buffers" :
"please check Portals");
+ if (tx->tx_tposted) {
+ CERROR("Could not send to %s after %ds (sent %lds ago); "
+ "check Portals for possible issues\n",
+ libcfs_id2str(peer->peer_id),
+ *kptllnd_tunables.kptl_timeout,
+ cfs_duration_sec(jiffies - tx->tx_tposted));
+ } else if (state < PEER_STATE_ACTIVE) {
+ CERROR("Could not connect %s (%d) after %ds; "
+ "peer might be down\n",
+ libcfs_id2str(peer->peer_id), state,
+ *kptllnd_tunables.kptl_timeout);
+ } else {
+ CERROR("Could not get credits for %s after %ds; "
+ "possible Lustre networking issues\n",
+ libcfs_id2str(peer->peer_id),
+ *kptllnd_tunables.kptl_timeout);
+ }
+
CERROR("%s timed out: cred %d outstanding %d, sent %d, "
- "sendq %d, activeq %d Tx %p %s (%s%s%s) status %d "
- "%sposted %lu T/O %ds\n",
- libcfs_id2str(peer->peer_id), peer->peer_credits,
- peer->peer_outstanding_credits, peer->peer_sent_credits,
- nsend, nactive, tx, kptllnd_tx_typestr(tx->tx_type),
+ "state %d, sent_hello %d, sendq %d, activeq %d "
+ "Tx %p %s %s (%s%s%s) status %d %sposted %lu T/O %ds\n",
+ libcfs_id2str(peer->peer_id), c, oc, sc,
+ state, sent_hello, nsend, nactive,
+ tx, kptllnd_tx_typestr(tx->tx_type),
+ kptllnd_msgtype2str(tx->tx_msg->ptlm_type),
tx->tx_active ? "A" : "",
PtlHandleIsEqual(tx->tx_msg_mdh, PTL_INVALID_HANDLE) ?
"" : "M",
(tx->tx_tposted == 0) ? 0UL : (jiffies - tx->tx_tposted),
*kptllnd_tunables.kptl_timeout);
- kptllnd_dump_ptltrace();
+#ifdef CRAY_XT3
+ if (*kptllnd_tunables.kptl_ptltrace_on_timeout)
+ kptllnd_dump_ptltrace();
+#endif
kptllnd_tx_decref(tx);
kptl_peer_t *peer;
list_for_each (tmp, peers) {
-
peer = list_entry (tmp, kptl_peer_t, peer_list);
LASSERT(peer->peer_state == PEER_STATE_WAITING_HELLO ||
peer->peer_state == PEER_STATE_ACTIVE);
-
- if (peer->peer_id.nid != id.nid ||
- peer->peer_id.pid != id.pid)
+
+ /* NB logical LNet peers share one kptl_peer_t */
+ if (peer->peer_id.pid != id.pid ||
+ LNET_NIDADDR(id.nid) != LNET_NIDADDR(peer->peer_id.nid))
continue;
kptllnd_peer_addref(peer);
CDEBUG(D_NET, "%s -> %s (%d)\n",
- libcfs_id2str(id),
+ libcfs_id2str(id),
kptllnd_ptlid2str(peer->peer_ptlid),
atomic_read (&peer->peer_refcount));
return peer;
list_for_each (tmp, &kptllnd_data.kptl_closing_peers) {
peer = list_entry (tmp, kptl_peer_t, peer_list);
- if (peer->peer_id.nid == lpid.nid &&
+ if (LNET_NIDADDR(peer->peer_id.nid) == LNET_NIDADDR(lpid.nid) &&
peer->peer_id.pid == lpid.pid)
return peer->peer_last_matchbits_seen;
}
-
+
list_for_each (tmp, &kptllnd_data.kptl_zombie_peers) {
peer = list_entry (tmp, kptl_peer_t, peer_list);
- if (peer->peer_id.nid == lpid.nid &&
+ if (LNET_NIDADDR(peer->peer_id.nid) == LNET_NIDADDR(lpid.nid) &&
peer->peer_id.pid == lpid.pid)
return peer->peer_last_matchbits_seen;
}
-
+
return PTL_RESERVED_MATCHBITS;
}
kptl_peer_t *
-kptllnd_peer_handle_hello (ptl_process_id_t initiator,
- kptl_msg_t *msg)
+kptllnd_peer_handle_hello (kptl_net_t *net,
+ ptl_process_id_t initiator, kptl_msg_t *msg)
{
rwlock_t *g_lock = &kptllnd_data.kptl_peer_rw_lock;
kptl_peer_t *peer;
msg->ptlm_srcstamp, peer->peer_incarnation);
kptllnd_peer_decref(peer);
+ peer = NULL;
}
hello_tx = kptllnd_get_idle_tx(TX_TYPE_SMALL_MESSAGE);
}
kptllnd_init_msg(hello_tx->tx_msg, PTLLND_MSG_TYPE_HELLO,
- sizeof(kptl_hello_msg_t));
+ lpid, sizeof(kptl_hello_msg_t));
- new_peer = kptllnd_peer_allocate(lpid, initiator);
+ new_peer = kptllnd_peer_allocate(net, lpid, initiator);
if (new_peer == NULL) {
kptllnd_tx_decref(hello_tx);
return NULL;
write_lock_irqsave(g_lock, flags);
again:
- if (kptllnd_data.kptl_shutdown) {
+ if (net->net_shutdown) {
write_unlock_irqrestore(g_lock, flags);
CERROR ("Shutdown started, refusing connection from %s\n",
new_peer->peer_last_matchbits_seen = last_matchbits_seen;
new_peer->peer_max_msg_size = msg->ptlm_u.hello.kptlhm_max_msg_size;
+ LASSERT (!net->net_shutdown);
kptllnd_peer_add_peertable_locked(new_peer);
write_unlock_irqrestore(g_lock, flags);
}
int
-kptllnd_find_target(kptl_peer_t **peerp, lnet_process_id_t target)
+kptllnd_find_target(kptl_net_t *net, lnet_process_id_t target,
+ kptl_peer_t **peerp)
{
rwlock_t *g_lock = &kptllnd_data.kptl_peer_rw_lock;
ptl_process_id_t ptl_id;
return -EHOSTUNREACH;
}
- /* The new peer is a kernel ptllnd, and kernel ptllnds all have
- * the same portals PID */
+ /* The new peer is a kernel ptllnd, and kernel ptllnds all have the
+ * same portals PID, which has nothing to do with LUSTRE_SRV_LNET_PID */
ptl_id.nid = kptllnd_lnet2ptlnid(target.nid);
ptl_id.pid = kptllnd_data.kptl_portals_id.pid;
return -ENOMEM;
}
+ hello_tx->tx_acked = 1;
kptllnd_init_msg(hello_tx->tx_msg, PTLLND_MSG_TYPE_HELLO,
- sizeof(kptl_hello_msg_t));
+ target, sizeof(kptl_hello_msg_t));
- new_peer = kptllnd_peer_allocate(target, ptl_id);
+ new_peer = kptllnd_peer_allocate(net, target, ptl_id);
if (new_peer == NULL) {
rc = -ENOMEM;
goto unwind_0;
write_lock_irqsave(g_lock, flags);
again:
- if (kptllnd_data.kptl_shutdown) {
- write_unlock_irqrestore(g_lock, flags);
- rc = -ESHUTDOWN;
- goto unwind_2;
- }
+ /* Called only in lnd_send which can't happen after lnd_shutdown */
+ LASSERT (!net->net_shutdown);
*peerp = kptllnd_id2peer_locked(target);
if (*peerp != NULL) {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff