Whamcloud - gitweb
LU-6142 lnet: change wire protocol typedefs to proper structure
[fs/lustre-release.git] / lnet / lnet / peer.c
index 182f44d..b4a075d 100644 (file)
  *
  * 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.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2014, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -39,6 +35,7 @@
 #define DEBUG_SUBSYSTEM S_LNET
 
 #include <lnet/lib-lnet.h>
+#include <lnet/lib-dlc.h>
 
 int
 lnet_peer_tables_create(void)
@@ -103,61 +100,112 @@ lnet_peer_tables_destroy(void)
        the_lnet.ln_peer_tables = NULL;
 }
 
+static void
+lnet_peer_table_cleanup_locked(lnet_ni_t *ni, struct lnet_peer_table *ptable)
+{
+       int              i;
+       lnet_peer_t     *lp;
+       lnet_peer_t     *tmp;
+
+       for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
+               list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i],
+                                        lp_hashlist) {
+                       if (ni != NULL && ni != lp->lp_ni)
+                               continue;
+                       list_del_init(&lp->lp_hashlist);
+                       /* Lose hash table's ref */
+                       ptable->pt_zombies++;
+                       lnet_peer_decref_locked(lp);
+               }
+       }
+}
+
+static void
+lnet_peer_table_deathrow_wait_locked(struct lnet_peer_table *ptable,
+                                    int cpt_locked)
+{
+       int     i;
+
+       for (i = 3; ptable->pt_zombies != 0; i++) {
+               lnet_net_unlock(cpt_locked);
+
+               if (IS_PO2(i)) {
+                       CDEBUG(D_WARNING,
+                              "Waiting for %d zombies on peer table\n",
+                              ptable->pt_zombies);
+               }
+               set_current_state(TASK_UNINTERRUPTIBLE);
+               schedule_timeout(cfs_time_seconds(1) >> 1);
+               lnet_net_lock(cpt_locked);
+       }
+}
+
+static void
+lnet_peer_table_del_rtrs_locked(lnet_ni_t *ni, struct lnet_peer_table *ptable,
+                               int cpt_locked)
+{
+       lnet_peer_t     *lp;
+       lnet_peer_t     *tmp;
+       lnet_nid_t       lp_nid;
+       int              i;
+
+       for (i = 0; i < LNET_PEER_HASH_SIZE; i++) {
+               list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i],
+                                        lp_hashlist) {
+                       if (ni != lp->lp_ni)
+                               continue;
+
+                       if (lp->lp_rtr_refcount == 0)
+                               continue;
+
+                       lp_nid = lp->lp_nid;
+
+                       lnet_net_unlock(cpt_locked);
+                       lnet_del_route(LNET_NIDNET(LNET_NID_ANY), lp_nid);
+                       lnet_net_lock(cpt_locked);
+               }
+       }
+}
+
 void
-lnet_peer_tables_cleanup(void)
+lnet_peer_tables_cleanup(lnet_ni_t *ni)
 {
-       struct lnet_peer_table  *ptable;
        int                     i;
-       int                     j;
+       struct lnet_peer_table  *ptable;
+       lnet_peer_t             *lp;
+       struct list_head        deathrow;
 
-       LASSERT(the_lnet.ln_shutdown);  /* i.e. no new peers */
+       INIT_LIST_HEAD(&deathrow);
 
+       LASSERT(the_lnet.ln_shutdown || ni != NULL);
+       /* If just deleting the peers for a NI, get rid of any routes these
+        * peers are gateways for. */
        cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
                lnet_net_lock(i);
-
-               for (j = 0; j < LNET_PEER_HASH_SIZE; j++) {
-                       struct list_head *peers = &ptable->pt_hash[j];
-
-                       while (!list_empty(peers)) {
-                               lnet_peer_t *lp = list_entry(peers->next,
-                                                                lnet_peer_t,
-                                                                lp_hashlist);
-                               list_del_init(&lp->lp_hashlist);
-                               /* lose hash table's ref */
-                               lnet_peer_decref_locked(lp);
-                       }
-               }
-
+               lnet_peer_table_del_rtrs_locked(ni, ptable, i);
                lnet_net_unlock(i);
        }
 
+       /* Start the process of moving the applicable peers to
+        * deathrow. */
        cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
-               struct list_head deathrow = LIST_HEAD_INIT(deathrow);
-               lnet_peer_t     *lp;
-
                lnet_net_lock(i);
+               lnet_peer_table_cleanup_locked(ni, ptable);
+               lnet_net_unlock(i);
+       }
 
-               for (j = 3; ptable->pt_number != 0; j++) {
-                       lnet_net_unlock(i);
-
-                       if ((j & (j - 1)) == 0) {
-                               CDEBUG(D_WARNING,
-                                      "Waiting for %d peers on peer table\n",
-                                      ptable->pt_number);
-                       }
-                       cfs_pause(cfs_time_seconds(1) / 2);
-                       lnet_net_lock(i);
-               }
+       /* Cleanup all entries on deathrow. */
+       cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
+               lnet_net_lock(i);
+               lnet_peer_table_deathrow_wait_locked(ptable, i);
                list_splice_init(&ptable->pt_deathrow, &deathrow);
-
                lnet_net_unlock(i);
+       }
 
-               while (!list_empty(&deathrow)) {
-                       lp = list_entry(deathrow.next,
-                                       lnet_peer_t, lp_hashlist);
-                       list_del(&lp->lp_hashlist);
-                       LIBCFS_FREE(lp, sizeof(*lp));
-               }
+       while (!list_empty(&deathrow)) {
+               lp = list_entry(deathrow.next, lnet_peer_t, lp_hashlist);
+               list_del(&lp->lp_hashlist);
+               LIBCFS_FREE(lp, sizeof(*lp));
        }
 }
 
@@ -180,6 +228,8 @@ lnet_destroy_peer_locked(lnet_peer_t *lp)
        lp->lp_ni = NULL;
 
        list_add(&lp->lp_hashlist, &ptable->pt_deathrow);
+       LASSERT(ptable->pt_zombies > 0);
+       ptable->pt_zombies--;
 }
 
 lnet_peer_t *
@@ -252,15 +302,15 @@ lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid, int cpt)
        INIT_LIST_HEAD(&lp->lp_rtrq);
        INIT_LIST_HEAD(&lp->lp_routes);
 
-        lp->lp_notify = 0;
-        lp->lp_notifylnd = 0;
-        lp->lp_notifying = 0;
-        lp->lp_alive_count = 0;
-        lp->lp_timestamp = 0;
-        lp->lp_alive = !lnet_peers_start_down(); /* 1 bit!! */
-        lp->lp_last_alive = cfs_time_current(); /* assumes alive */
-        lp->lp_last_query = 0; /* haven't asked NI yet */
-        lp->lp_ping_timestamp = 0;
+       lp->lp_notify = 0;
+       lp->lp_notifylnd = 0;
+       lp->lp_notifying = 0;
+       lp->lp_alive_count = 0;
+       lp->lp_timestamp = 0;
+       lp->lp_alive = !lnet_peers_start_down(); /* 1 bit!! */
+       lp->lp_last_alive = cfs_time_current(); /* assumes alive */
+       lp->lp_last_query = 0; /* haven't asked NI yet */
+       lp->lp_ping_timestamp = 0;
        lp->lp_ping_feats = LNET_PING_FEAT_INVAL;
        lp->lp_nid = nid;
        lp->lp_cpt = cpt2;
@@ -318,20 +368,82 @@ lnet_debug_peer(lnet_nid_t nid)
        rc = lnet_nid2peer_locked(&lp, nid, cpt);
        if (rc != 0) {
                lnet_net_unlock(cpt);
-                CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
-                return;
-        }
+               CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
+               return;
+       }
 
-        if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
-                aliveness = lp->lp_alive ? "up" : "down";
+       if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
+               aliveness = lp->lp_alive ? "up" : "down";
 
-        CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
-               libcfs_nid2str(lp->lp_nid), lp->lp_refcount,
-               aliveness, lp->lp_ni->ni_peertxcredits,
-               lp->lp_rtrcredits, lp->lp_minrtrcredits,
-               lp->lp_txcredits, lp->lp_mintxcredits, lp->lp_txqnob);
+       CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
+              libcfs_nid2str(lp->lp_nid), lp->lp_refcount,
+              aliveness, lp->lp_ni->ni_peertxcredits,
+              lp->lp_rtrcredits, lp->lp_minrtrcredits,
+              lp->lp_txcredits, lp->lp_mintxcredits, lp->lp_txqnob);
 
-        lnet_peer_decref_locked(lp);
+       lnet_peer_decref_locked(lp);
 
        lnet_net_unlock(cpt);
 }
+
+int lnet_get_peer_info(__u32 peer_index, __u64 *nid,
+                      char aliveness[LNET_MAX_STR_LEN],
+                      __u32 *cpt_iter, __u32 *refcount,
+                      __u32 *ni_peer_tx_credits, __u32 *peer_tx_credits,
+                      __u32 *peer_rtr_credits, __u32 *peer_min_rtr_credits,
+                      __u32 *peer_tx_qnob)
+{
+       struct lnet_peer_table  *peer_table;
+       lnet_peer_t             *lp;
+       int                     j;
+       int                     lncpt;
+       bool                    found = false;
+
+       /* get the number of CPTs */
+       lncpt = cfs_percpt_number(the_lnet.ln_peer_tables);
+
+       /* if the cpt number to be examined is >= the number of cpts in
+        * the system then indicate that there are no more cpts to examin
+        */
+       if (*cpt_iter > lncpt)
+               return -ENOENT;
+
+       /* get the current table */
+       peer_table = the_lnet.ln_peer_tables[*cpt_iter];
+       /* if the ptable is NULL then there are no more cpts to examine */
+       if (peer_table == NULL)
+               return -ENOENT;
+
+       lnet_net_lock(*cpt_iter);
+
+       for (j = 0; j < LNET_PEER_HASH_SIZE && !found; j++) {
+               struct list_head *peers = &peer_table->pt_hash[j];
+
+               list_for_each_entry(lp, peers, lp_hashlist) {
+                       if (peer_index-- > 0)
+                               continue;
+
+                       snprintf(aliveness, LNET_MAX_STR_LEN, "NA");
+                       if (lnet_isrouter(lp) ||
+                               lnet_peer_aliveness_enabled(lp))
+                               snprintf(aliveness, LNET_MAX_STR_LEN,
+                                        lp->lp_alive ? "up" : "down");
+
+                       *nid = lp->lp_nid;
+                       *refcount = lp->lp_refcount;
+                       *ni_peer_tx_credits = lp->lp_ni->ni_peertxcredits;
+                       *peer_tx_credits = lp->lp_txcredits;
+                       *peer_rtr_credits = lp->lp_rtrcredits;
+                       *peer_min_rtr_credits = lp->lp_mintxcredits;
+                       *peer_tx_qnob = lp->lp_txqnob;
+
+                       found = true;
+               }
+
+       }
+       lnet_net_unlock(*cpt_iter);
+
+       *cpt_iter = lncpt;
+
+       return found ? 0 : -ENOENT;
+}