Whamcloud - gitweb
LU-56 lnet: eliminate a few locking dance in LNet
[fs/lustre-release.git] / lnet / lnet / api-ni.c
index cf9226d..f2b4b2b 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, Whamcloud, Inc.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -91,10 +91,10 @@ lnet_get_networks(void)
 void
 lnet_init_locks(void)
 {
-        spin_lock_init (&the_lnet.ln_lock);
-        cfs_waitq_init (&the_lnet.ln_waitq);
-        init_mutex(&the_lnet.ln_lnd_mutex);
-        init_mutex(&the_lnet.ln_api_mutex);
+       cfs_spin_lock_init (&the_lnet.ln_lock);
+       cfs_waitq_init(&the_lnet.ln_eq_waitq);
+       cfs_mutex_init(&the_lnet.ln_lnd_mutex);
+       cfs_mutex_init(&the_lnet.ln_api_mutex);
 }
 
 void
@@ -123,7 +123,7 @@ lnet_get_networks (void)
         int               len;
         int               nob;
         int               rc;
-        struct list_head *tmp;
+        cfs_list_t       *tmp;
 
 #ifdef NOT_YET
         if (networks != NULL && ip2nets != NULL) {
@@ -137,8 +137,8 @@ lnet_get_networks (void)
                 return (rc == 0) ? networks : NULL;
         }
 #else
-        ip2nets = NULL;
-        rc = 0;
+        SET_BUT_UNUSED(ip2nets);
+        SET_BUT_UNUSED(rc);
 #endif
         if (networks != NULL)
                 return networks;
@@ -150,8 +150,8 @@ lnet_get_networks (void)
         *str = 0;
         sep = "";
 
-        list_for_each (tmp, &the_lnet.ln_lnds) {
-                lnd_t *lnd = list_entry(tmp, lnd_t, lnd_list);
+        cfs_list_for_each (tmp, &the_lnet.ln_lnds) {
+                lnd_t *lnd = cfs_list_entry(tmp, lnd_t, lnd_list);
 
                 nob = snprintf(str, len, "%s%s", sep,
                                libcfs_lnd2str(lnd->lnd_type));
@@ -189,18 +189,18 @@ void lnet_fini_locks(void)
 
 void lnet_init_locks(void)
 {
-        pthread_cond_init(&the_lnet.ln_cond, NULL);
-        pthread_mutex_init(&the_lnet.ln_lock, NULL);
-        pthread_mutex_init(&the_lnet.ln_lnd_mutex, NULL);
-        pthread_mutex_init(&the_lnet.ln_api_mutex, NULL);
+       pthread_cond_init(&the_lnet.ln_eq_cond, NULL);
+       pthread_mutex_init(&the_lnet.ln_lock, NULL);
+       pthread_mutex_init(&the_lnet.ln_lnd_mutex, NULL);
+       pthread_mutex_init(&the_lnet.ln_api_mutex, NULL);
 }
 
 void lnet_fini_locks(void)
 {
-        pthread_mutex_destroy(&the_lnet.ln_api_mutex);
-        pthread_mutex_destroy(&the_lnet.ln_lnd_mutex);
-        pthread_mutex_destroy(&the_lnet.ln_lock);
-        pthread_cond_destroy(&the_lnet.ln_cond);
+       pthread_mutex_destroy(&the_lnet.ln_api_mutex);
+       pthread_mutex_destroy(&the_lnet.ln_lnd_mutex);
+       pthread_mutex_destroy(&the_lnet.ln_lock);
+       pthread_cond_destroy(&the_lnet.ln_eq_cond);
 }
 
 # endif
@@ -303,13 +303,13 @@ lnd_t *
 lnet_find_lnd_by_type (int type)
 {
         lnd_t              *lnd;
-        struct list_head   *tmp;
+        cfs_list_t         *tmp;
 
         /* holding lnd mutex */
-        list_for_each (tmp, &the_lnet.ln_lnds) {
-                lnd = list_entry(tmp, lnd_t, lnd_list);
+        cfs_list_for_each (tmp, &the_lnet.ln_lnds) {
+                lnd = cfs_list_entry(tmp, lnd_t, lnd_list);
 
-                if (lnd->lnd_type == type)
+                if ((int)lnd->lnd_type == type)
                         return lnd;
         }
 
@@ -319,49 +319,36 @@ lnet_find_lnd_by_type (int type)
 void
 lnet_register_lnd (lnd_t *lnd)
 {
-        LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex);
+        LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
 
         LASSERT (the_lnet.ln_init);
         LASSERT (libcfs_isknown_lnd(lnd->lnd_type));
         LASSERT (lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 
-        list_add_tail (&lnd->lnd_list, &the_lnet.ln_lnds);
+        cfs_list_add_tail (&lnd->lnd_list, &the_lnet.ln_lnds);
         lnd->lnd_refcount = 0;
 
         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 
-        LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
+        LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
 }
 
 void
 lnet_unregister_lnd (lnd_t *lnd)
 {
-        LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex);
+        LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
 
         LASSERT (the_lnet.ln_init);
         LASSERT (lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
         LASSERT (lnd->lnd_refcount == 0);
 
-        list_del (&lnd->lnd_list);
+        cfs_list_del (&lnd->lnd_list);
         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 
-        LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
+        LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
 }
 
-#ifndef LNET_USE_LIB_FREELIST
-
-int
-lnet_descriptor_setup (void)
-{
-        return 0;
-}
-
-void
-lnet_descriptor_cleanup (void)
-{
-}
-
-#else
+#ifdef LNET_USE_LIB_FREELIST
 
 int
 lnet_freelist_init (lnet_freelist_t *fl, int n, int size)
@@ -384,7 +371,7 @@ lnet_freelist_init (lnet_freelist_t *fl, int n, int size)
         do
         {
                 memset (space, 0, size);
-                list_add ((struct list_head *)space, &fl->fl_list);
+                cfs_list_add ((cfs_list_t *)space, &fl->fl_list);
                 space += size;
         } while (--n != 0);
 
@@ -394,7 +381,7 @@ lnet_freelist_init (lnet_freelist_t *fl, int n, int size)
 void
 lnet_freelist_fini (lnet_freelist_t *fl)
 {
-        struct list_head *el;
+        cfs_list_t       *el;
         int               count;
 
         if (fl->fl_nobjs == 0)
@@ -410,48 +397,7 @@ lnet_freelist_fini (lnet_freelist_t *fl)
         memset (fl, 0, sizeof (*fl));
 }
 
-int
-lnet_descriptor_setup (void)
-{
-        /* NB on failure caller must still call lnet_descriptor_cleanup */
-        /*               ******                                         */
-        int        rc;
-
-        memset (&the_lnet.ln_free_mes,  0, sizeof (the_lnet.ln_free_mes));
-        memset (&the_lnet.ln_free_msgs, 0, sizeof (the_lnet.ln_free_msgs));
-        memset (&the_lnet.ln_free_mds,  0, sizeof (the_lnet.ln_free_mds));
-        memset (&the_lnet.ln_free_eqs,  0, sizeof (the_lnet.ln_free_eqs));
-
-        rc = lnet_freelist_init(&the_lnet.ln_free_mes,
-                                MAX_MES, sizeof (lnet_me_t));
-        if (rc != 0)
-                return (rc);
-
-        rc = lnet_freelist_init(&the_lnet.ln_free_msgs,
-                                MAX_MSGS, sizeof (lnet_msg_t));
-        if (rc != 0)
-                return (rc);
-
-        rc = lnet_freelist_init(&the_lnet.ln_free_mds,
-                                MAX_MDS, sizeof (lnet_libmd_t));
-        if (rc != 0)
-                return (rc);
-
-        rc = lnet_freelist_init(&the_lnet.ln_free_eqs,
-                                MAX_EQS, sizeof (lnet_eq_t));
-        return (rc);
-}
-
-void
-lnet_descriptor_cleanup (void)
-{
-        lnet_freelist_fini (&the_lnet.ln_free_mes);
-        lnet_freelist_fini (&the_lnet.ln_free_msgs);
-        lnet_freelist_fini (&the_lnet.ln_free_mds);
-        lnet_freelist_fini (&the_lnet.ln_free_eqs);
-}
-
-#endif
+#endif /* LNET_USE_LIB_FREELIST */
 
 __u64
 lnet_create_interface_cookie (void)
@@ -466,7 +412,7 @@ lnet_create_interface_cookie (void)
         int            rc = gettimeofday (&tv, NULL);
         LASSERT (rc == 0);
 #else
-        do_gettimeofday(&tv);
+        cfs_gettimeofday(&tv);
 #endif
         cookie = tv.tv_sec;
         cookie *= 1000000;
@@ -474,132 +420,146 @@ lnet_create_interface_cookie (void)
         return cookie;
 }
 
-int
-lnet_setup_handle_hash (void)
+static char *
+lnet_res_type2str(int type)
 {
-        int       i;
-
-        /* Arbitrary choice of hash table size */
-#ifdef __KERNEL__
-        the_lnet.ln_lh_hash_size = CFS_PAGE_SIZE / sizeof (struct list_head);
-#else
-        the_lnet.ln_lh_hash_size = (MAX_MES + MAX_MDS + MAX_EQS)/4;
-#endif
-        LIBCFS_ALLOC(the_lnet.ln_lh_hash_table,
-                     the_lnet.ln_lh_hash_size * sizeof (struct list_head));
-        if (the_lnet.ln_lh_hash_table == NULL)
-                return (-ENOMEM);
-
-        for (i = 0; i < the_lnet.ln_lh_hash_size; i++)
-                CFS_INIT_LIST_HEAD (&the_lnet.ln_lh_hash_table[i]);
-
-        the_lnet.ln_next_object_cookie = LNET_COOKIE_TYPES;
-
-        return (0);
+       switch (type) {
+       default:
+               LBUG();
+       case LNET_COOKIE_TYPE_MD:
+               return "MD";
+       case LNET_COOKIE_TYPE_ME:
+               return "ME";
+       case LNET_COOKIE_TYPE_EQ:
+               return "EQ";
+       }
 }
 
 void
-lnet_cleanup_handle_hash (void)
+lnet_res_container_cleanup(struct lnet_res_container *rec)
 {
-        if (the_lnet.ln_lh_hash_table == NULL)
-                return;
+       int     count = 0;
 
-        LIBCFS_FREE(the_lnet.ln_lh_hash_table,
-                    the_lnet.ln_lh_hash_size * sizeof (struct list_head));
-}
+       if (rec->rec_type == 0) /* not set yet, it's a uninitialized */
+               return;
 
-lnet_libhandle_t *
-lnet_lookup_cookie (__u64 cookie, int type)
-{
-        /* ALWAYS called with LNET_LOCK held */
-        struct list_head    *list;
-        struct list_head    *el;
-        unsigned int         hash;
+       while (!cfs_list_empty(&rec->rec_active)) {
+               cfs_list_t *e = rec->rec_active.next;
 
-        if ((cookie & (LNET_COOKIE_TYPES - 1)) != type)
-                return (NULL);
+               cfs_list_del_init(e);
+               if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
+                       lnet_eq_free(cfs_list_entry(e, lnet_eq_t, eq_list));
 
-        hash = ((unsigned int)cookie) % the_lnet.ln_lh_hash_size;
-        list = &the_lnet.ln_lh_hash_table[hash];
+               } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
+                       lnet_md_free(cfs_list_entry(e, lnet_libmd_t, md_list));
 
-        list_for_each (el, list) {
-                lnet_libhandle_t *lh = list_entry (el, lnet_libhandle_t,
-                                                  lh_hash_chain);
+               } else { /* NB: Active MEs should be attached on portals */
+                       LBUG();
+               }
+               count++;
+       }
 
-                if (lh->lh_cookie == cookie)
-                        return (lh);
-        }
+       if (count > 0) {
+               /* Found alive MD/ME/EQ, user really should unlink/free
+                * all of them before finalize LNet, but if someone didn't,
+                * we have to recycle garbage for him */
+               CERROR("%d active elements on exit of %s container\n",
+                      count, lnet_res_type2str(rec->rec_type));
+       }
+
+#ifdef LNET_USE_LIB_FREELIST
+       lnet_freelist_fini(&rec->rec_freelist);
+#endif
+       if (rec->rec_lh_hash != NULL) {
+               LIBCFS_FREE(rec->rec_lh_hash,
+                           LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
+               rec->rec_lh_hash = NULL;
+       }
 
-        return (NULL);
+       rec->rec_type = 0; /* mark it as finalized */
 }
 
-void
-lnet_initialise_handle (lnet_libhandle_t *lh, int type)
+int
+lnet_res_container_setup(struct lnet_res_container *rec,
+                        int type, int objnum, int objsz)
 {
-        /* ALWAYS called with LNET_LOCK held */
-        unsigned int    hash;
+       int     rc = 0;
+       int     i;
 
-        LASSERT (type >= 0 && type < LNET_COOKIE_TYPES);
-        lh->lh_cookie = the_lnet.ln_next_object_cookie | type;
-        the_lnet.ln_next_object_cookie += LNET_COOKIE_TYPES;
+       LASSERT(rec->rec_type == 0);
 
-        hash = ((unsigned int)lh->lh_cookie) % the_lnet.ln_lh_hash_size;
-        list_add (&lh->lh_hash_chain, &the_lnet.ln_lh_hash_table[hash]);
-}
+       rec->rec_type = type;
+       CFS_INIT_LIST_HEAD(&rec->rec_active);
 
-void
-lnet_invalidate_handle (lnet_libhandle_t *lh)
-{
-        /* ALWAYS called with LNET_LOCK held */
-        list_del (&lh->lh_hash_chain);
+#ifdef LNET_USE_LIB_FREELIST
+       memset(&rec->rec_freelist, 0, sizeof(rec->rec_freelist));
+       rc = lnet_freelist_init(&rec->rec_freelist, objnum, objsz);
+       if (rc != 0)
+               goto out;
+#endif
+       rec->rec_lh_cookie = type;
+
+       /* Arbitrary choice of hash table size */
+       LIBCFS_ALLOC(rec->rec_lh_hash,
+                    LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
+       if (rec->rec_lh_hash == NULL) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       for (i = 0; i < LNET_LH_HASH_SIZE; i++)
+               CFS_INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
+
+       return 0;
+
+out:
+       CERROR("Failed to setup %s resource container\n",
+              lnet_res_type2str(type));
+       lnet_res_container_cleanup(rec);
+       return rc;
 }
 
-int
-lnet_init_finalizers(void)
+lnet_libhandle_t *
+lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
 {
-#ifdef __KERNEL__
-        int    i;
+       /* ALWAYS called with lnet_res_lock held */
+       cfs_list_t              *head;
+       lnet_libhandle_t        *lh;
+       unsigned int            hash;
 
-        the_lnet.ln_nfinalizers = num_online_cpus();
+       if ((cookie & (LNET_COOKIE_TYPES - 1)) != rec->rec_type)
+               return NULL;
 
-        LIBCFS_ALLOC(the_lnet.ln_finalizers,
-                     the_lnet.ln_nfinalizers *
-                     sizeof(*the_lnet.ln_finalizers));
-        if (the_lnet.ln_finalizers == NULL) {
-                CERROR("Can't allocate ln_finalizers\n");
-                return -ENOMEM;
-        }
+       hash = cookie >> LNET_COOKIE_TYPE_BITS;
+       head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
 
-        for (i = 0; i < the_lnet.ln_nfinalizers; i++)
-                the_lnet.ln_finalizers[i] = NULL;
-#else
-        the_lnet.ln_finalizing = 0;
-#endif
+       cfs_list_for_each_entry(lh, head, lh_hash_chain) {
+               if (lh->lh_cookie == cookie)
+                       return lh;
+       }
 
-        CFS_INIT_LIST_HEAD(&the_lnet.ln_finalizeq);
-        return 0;
+       return NULL;
 }
 
 void
-lnet_fini_finalizers(void)
+lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
 {
-#ifdef __KERNEL__
-        int    i;
+       /* ALWAYS called with lnet_res_lock held */
+       unsigned int    ibits = LNET_COOKIE_TYPE_BITS;
+       unsigned int    hash;
 
-        for (i = 0; i < the_lnet.ln_nfinalizers; i++)
-                LASSERT (the_lnet.ln_finalizers[i] == NULL);
+       lh->lh_cookie = rec->rec_lh_cookie;
+       rec->rec_lh_cookie += 1 << ibits;
 
-        LIBCFS_FREE(the_lnet.ln_finalizers,
-                    the_lnet.ln_nfinalizers *
-                    sizeof(*the_lnet.ln_finalizers));
-#else
-        LASSERT (!the_lnet.ln_finalizing);
-#endif
-        LASSERT (list_empty(&the_lnet.ln_finalizeq));
+       hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
+
+       cfs_list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
 }
 
 #ifndef __KERNEL__
-/* Temporary workaround to allow uOSS and test programs force server
+/**
+ * Reserved API - do not use.
+ * Temporary workaround to allow uOSS and test programs force server
  * mode in userspace. See comments near ln_server_mode_flag in
  * lnet/lib-types.h */
 
@@ -614,7 +574,6 @@ lnet_prepare(lnet_pid_t requested_pid)
 {
         /* Prepare to bring up the network */
         int               rc = 0;
-        int               i;
 
         LASSERT (the_lnet.ln_refcount == 0);
 
@@ -638,16 +597,9 @@ lnet_prepare(lnet_pid_t requested_pid)
         }
 #endif
 
-        rc = lnet_descriptor_setup();
-        if (rc != 0)
-                goto failed0;
-
         memset(&the_lnet.ln_counters, 0,
                sizeof(the_lnet.ln_counters));
 
-        CFS_INIT_LIST_HEAD (&the_lnet.ln_active_msgs);
-        CFS_INIT_LIST_HEAD (&the_lnet.ln_active_mds);
-        CFS_INIT_LIST_HEAD (&the_lnet.ln_active_eqs);
         CFS_INIT_LIST_HEAD (&the_lnet.ln_test_peers);
         CFS_INIT_LIST_HEAD (&the_lnet.ln_nis);
         CFS_INIT_LIST_HEAD (&the_lnet.ln_zombie_nis);
@@ -658,51 +610,66 @@ lnet_prepare(lnet_pid_t requested_pid)
 
         lnet_init_rtrpools();
 
-        rc = lnet_setup_handle_hash ();
-        if (rc != 0)
-                goto failed0;
-
-        rc = lnet_create_peer_table();
+       rc = lnet_peer_table_create();
         if (rc != 0)
-                goto failed1;
-
-        rc = lnet_init_finalizers();
-        if (rc != 0)
-                goto failed2;
-
-        the_lnet.ln_nportals = MAX_PORTALS;
-        LIBCFS_ALLOC(the_lnet.ln_portals,
-                     the_lnet.ln_nportals *
-                     sizeof(*the_lnet.ln_portals));
-        if (the_lnet.ln_portals == NULL) {
-                rc = -ENOMEM;
-                goto failed3;
-        }
-
-        for (i = 0; i < the_lnet.ln_nportals; i++) {
-                CFS_INIT_LIST_HEAD(&(the_lnet.ln_portals[i].ptl_ml));
-                CFS_INIT_LIST_HEAD(&(the_lnet.ln_portals[i].ptl_msgq));
-                the_lnet.ln_portals[i].ptl_options = 0;
-        }
-
-        return 0;
+               goto failed0;
+
+       /* NB: we will have instance of message container per CPT soon */
+       rc = lnet_msg_container_setup(&the_lnet.ln_msg_container);
+       if (rc != 0)
+               goto failed1;
+
+       rc = lnet_res_container_setup(&the_lnet.ln_eq_container,
+                                     LNET_COOKIE_TYPE_EQ, LNET_FL_MAX_EQS,
+                                     sizeof(lnet_eq_t));
+       if (rc != 0) {
+               CERROR("Failed to create EQ container for LNet: %d\n", rc);
+               goto failed2;
+       }
+
+       /* NB: we will have instance of ME container per CPT soon */
+       rc = lnet_res_container_setup(&the_lnet.ln_me_container,
+                                     LNET_COOKIE_TYPE_ME, LNET_FL_MAX_MES,
+                                     sizeof(lnet_me_t));
+       if (rc != 0) {
+               CERROR("Failed to create ME container for LNet: %d\n", rc);
+               goto failed3;
+       }
+
+       /* NB: we will have instance of MD container per CPT soon */
+       rc = lnet_res_container_setup(&the_lnet.ln_md_container,
+                                     LNET_COOKIE_TYPE_MD, LNET_FL_MAX_MDS,
+                                     sizeof(lnet_libmd_t));
+       if (rc != 0) {
+               CERROR("Failed to create MD container for LNet: %d\n", rc);
+               goto failed3;
+       }
+
+       rc = lnet_portals_create();
+       if (rc != 0) {
+               CERROR("Failed to create portals for LNet: %d\n", rc);
+               goto failed3;
+       }
+
+       return 0;
 
  failed3:
-        lnet_fini_finalizers();
+       /* NB: lnet_res_container_cleanup is safe to call for
+        * uninitialized container */
+       lnet_res_container_cleanup(&the_lnet.ln_md_container);
+       lnet_res_container_cleanup(&the_lnet.ln_me_container);
+       lnet_res_container_cleanup(&the_lnet.ln_eq_container);
  failed2:
-        lnet_destroy_peer_table();
+       lnet_msg_container_cleanup(&the_lnet.ln_msg_container);
  failed1:
-        lnet_cleanup_handle_hash();
+       lnet_peer_table_destroy();
  failed0:
-        lnet_descriptor_cleanup();
-        return rc;
+       return rc;
 }
 
 int
 lnet_unprepare (void)
 {
-        int       idx;
-
         /* NB no LNET_LOCK since this is the last reference.  All LND instances
          * have shut down already, so it is safe to unlink and free all
          * descriptors, even those that appear committed to a network op (eg MD
@@ -710,74 +677,33 @@ lnet_unprepare (void)
 
         lnet_fail_nid(LNET_NID_ANY, 0);
 
-        LASSERT (list_empty(&the_lnet.ln_test_peers));
+        LASSERT (cfs_list_empty(&the_lnet.ln_test_peers));
         LASSERT (the_lnet.ln_refcount == 0);
-        LASSERT (list_empty(&the_lnet.ln_nis));
-        LASSERT (list_empty(&the_lnet.ln_zombie_nis));
+        LASSERT (cfs_list_empty(&the_lnet.ln_nis));
+        LASSERT (cfs_list_empty(&the_lnet.ln_zombie_nis));
         LASSERT (the_lnet.ln_nzombie_nis == 0);
 
-        for (idx = 0; idx < the_lnet.ln_nportals; idx++) {
-                LASSERT (list_empty(&the_lnet.ln_portals[idx].ptl_msgq));
-
-                while (!list_empty (&the_lnet.ln_portals[idx].ptl_ml)) {
-                        lnet_me_t *me = list_entry (the_lnet.ln_portals[idx].ptl_ml.next,
-                                                    lnet_me_t, me_list);
-
-                        CERROR ("Active me %p on exit\n", me);
-                        list_del (&me->me_list);
-                        lnet_me_free (me);
-                }
-        }
-
-        while (!list_empty (&the_lnet.ln_active_mds)) {
-                lnet_libmd_t *md = list_entry (the_lnet.ln_active_mds.next,
-                                               lnet_libmd_t, md_list);
-
-                CERROR ("Active md %p on exit\n", md);
-                list_del (&md->md_list);
-                lnet_md_free (md);
-        }
-
-        while (!list_empty (&the_lnet.ln_active_eqs)) {
-                lnet_eq_t *eq = list_entry (the_lnet.ln_active_eqs.next,
-                                            lnet_eq_t, eq_list);
-
-                CERROR ("Active eq %p on exit\n", eq);
-                list_del (&eq->eq_list);
-                lnet_eq_free (eq);
-        }
-
-        while (!list_empty (&the_lnet.ln_active_msgs)) {
-                lnet_msg_t *msg = list_entry (the_lnet.ln_active_msgs.next,
-                                              lnet_msg_t, msg_activelist);
-
-                CERROR ("Active msg %p on exit\n", msg);
-                LASSERT (msg->msg_onactivelist);
-                msg->msg_onactivelist = 0;
-                list_del (&msg->msg_activelist);
-                lnet_msg_free (msg);
-        }
+       lnet_portals_destroy();
 
-        LIBCFS_FREE(the_lnet.ln_portals,  
-                    the_lnet.ln_nportals * sizeof(*the_lnet.ln_portals));
+       lnet_res_container_cleanup(&the_lnet.ln_md_container);
+       lnet_res_container_cleanup(&the_lnet.ln_me_container);
+       lnet_res_container_cleanup(&the_lnet.ln_eq_container);
 
         lnet_free_rtrpools();
-        lnet_fini_finalizers();
-        lnet_destroy_peer_table();
-        lnet_cleanup_handle_hash();
-        lnet_descriptor_cleanup();
+       lnet_msg_container_cleanup(&the_lnet.ln_msg_container);
+       lnet_peer_table_destroy();
 
-        return (0);
+       return 0;
 }
 
 lnet_ni_t  *
 lnet_net2ni_locked (__u32 net)
 {
-        struct list_head *tmp;
+        cfs_list_t       *tmp;
         lnet_ni_t        *ni;
 
-        list_for_each (tmp, &the_lnet.ln_nis) {
-                ni = list_entry(tmp, lnet_ni_t, ni_list);
+        cfs_list_for_each (tmp, &the_lnet.ln_nis) {
+                ni = cfs_list_entry(tmp, lnet_ni_t, ni_list);
 
                 if (LNET_NIDNET(ni->ni_nid) == net) {
                         lnet_ni_addref_locked(ni);
@@ -805,11 +731,11 @@ lnet_islocalnet (__u32 net)
 lnet_ni_t  *
 lnet_nid2ni_locked (lnet_nid_t nid)
 {
-        struct list_head *tmp;
+        cfs_list_t       *tmp;
         lnet_ni_t        *ni;
 
-        list_for_each (tmp, &the_lnet.ln_nis) {
-                ni = list_entry(tmp, lnet_ni_t, ni_list);
+        cfs_list_for_each (tmp, &the_lnet.ln_nis) {
+                ni = cfs_list_entry(tmp, lnet_ni_t, ni_list);
 
                 if (ni->ni_nid == nid) {
                         lnet_ni_addref_locked(ni);
@@ -838,14 +764,14 @@ int
 lnet_count_acceptor_nis (void)
 {
         /* Return the # of NIs that need the acceptor. */
-        int                count = 0;
+        int            count = 0;
 #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD)
-        struct list_head  *tmp;
-        lnet_ni_t         *ni;
+        cfs_list_t    *tmp;
+        lnet_ni_t     *ni;
 
         LNET_LOCK();
-        list_for_each (tmp, &the_lnet.ln_nis) {
-                ni = list_entry(tmp, lnet_ni_t, ni_list);
+        cfs_list_for_each (tmp, &the_lnet.ln_nis) {
+                ni = cfs_list_entry(tmp, lnet_ni_t, ni_list);
 
                 if (ni->ni_lnd->lnd_accept != NULL)
                         count++;
@@ -869,28 +795,28 @@ lnet_shutdown_lndnis (void)
         /* All quiet on the API front */
         LASSERT (!the_lnet.ln_shutdown);
         LASSERT (the_lnet.ln_refcount == 0);
-        LASSERT (list_empty(&the_lnet.ln_zombie_nis));
+        LASSERT (cfs_list_empty(&the_lnet.ln_zombie_nis));
         LASSERT (the_lnet.ln_nzombie_nis == 0);
-        LASSERT (list_empty(&the_lnet.ln_remote_nets));
+        LASSERT (cfs_list_empty(&the_lnet.ln_remote_nets));
 
         LNET_LOCK();
         the_lnet.ln_shutdown = 1;               /* flag shutdown */
 
         /* Unlink NIs from the global table */
-        while (!list_empty(&the_lnet.ln_nis)) {
-                ni = list_entry(the_lnet.ln_nis.next,
-                                lnet_ni_t, ni_list);
-                list_del (&ni->ni_list);
+        while (!cfs_list_empty(&the_lnet.ln_nis)) {
+                ni = cfs_list_entry(the_lnet.ln_nis.next,
+                                    lnet_ni_t, ni_list);
+                cfs_list_del (&ni->ni_list);
 
                 the_lnet.ln_nzombie_nis++;
-                lnet_ni_decref_locked(ni); /* drop apini's ref */
+                lnet_ni_decref_locked(ni); /* drop ln_nis' ref */
         }
 
         /* Drop the cached eqwait NI. */
-        if (the_lnet.ln_eqwaitni != NULL) {
-                lnet_ni_decref_locked(the_lnet.ln_eqwaitni);
-                the_lnet.ln_eqwaitni = NULL;
-        }
+       if (the_lnet.ln_eq_waitni != NULL) {
+               lnet_ni_decref_locked(the_lnet.ln_eq_waitni);
+               the_lnet.ln_eq_waitni = NULL;
+       }
 
         /* Drop the cached loopback NI. */
         if (the_lnet.ln_loni != NULL) {
@@ -907,15 +833,15 @@ lnet_shutdown_lndnis (void)
 
         /* Clear the peer table and wait for all peers to go (they hold refs on
          * their NIs) */
-        lnet_clear_peer_table();
+       lnet_peer_table_cleanup();
 
         LNET_LOCK();
-        /* Now wait for the NI's I just nuked to show up on apini_zombie_nis
+        /* Now wait for the NI's I just nuked to show up on ln_zombie_nis
          * and shut them down in guaranteed thread context */
         i = 2;
         while (the_lnet.ln_nzombie_nis != 0) {
 
-                while (list_empty(&the_lnet.ln_zombie_nis)) {
+                while (cfs_list_empty(&the_lnet.ln_zombie_nis)) {
                         LNET_UNLOCK();
                         ++i;
                         if ((i & (-i)) == i)
@@ -925,16 +851,16 @@ lnet_shutdown_lndnis (void)
                         LNET_LOCK();
                 }
 
-                ni = list_entry(the_lnet.ln_zombie_nis.next,
-                                lnet_ni_t, ni_list);
-                list_del(&ni->ni_list);
+                ni = cfs_list_entry(the_lnet.ln_zombie_nis.next,
+                                    lnet_ni_t, ni_list);
+                cfs_list_del(&ni->ni_list);
                 ni->ni_lnd->lnd_refcount--;
 
                 LNET_UNLOCK();
 
                 islo = ni->ni_lnd->lnd_type == LOLND;
 
-                LASSERT (!in_interrupt ());
+                LASSERT (!cfs_in_interrupt ());
                 (ni->ni_lnd->lnd_shutdown)(ni);
 
                 /* can't deref lnd anymore now; it might have unregistered
@@ -965,7 +891,7 @@ lnet_startup_lndnis (void)
 {
         lnd_t             *lnd;
         lnet_ni_t         *ni;
-        struct list_head   nilist;
+        cfs_list_t         nilist;
         int                rc = 0;
         int                lnd_type;
         int                nicount = 0;
@@ -980,38 +906,48 @@ lnet_startup_lndnis (void)
         if (rc != 0)
                 goto failed;
 
-        while (!list_empty(&nilist)) {
-                ni = list_entry(nilist.next, lnet_ni_t, ni_list);
+        while (!cfs_list_empty(&nilist)) {
+                ni = cfs_list_entry(nilist.next, lnet_ni_t, ni_list);
                 lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
 
                 LASSERT (libcfs_isknown_lnd(lnd_type));
 
-                LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex);
+                if (lnd_type == CIBLND    ||
+                    lnd_type == OPENIBLND ||
+                    lnd_type == IIBLND    ||
+                    lnd_type == VIBLND) {
+                        CERROR("LND %s obsoleted\n",
+                               libcfs_lnd2str(lnd_type));
+                        goto failed;
+                }
+
+                LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
                 lnd = lnet_find_lnd_by_type(lnd_type);
 
 #ifdef __KERNEL__
                 if (lnd == NULL) {
-                        LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
-                        rc = request_module(libcfs_lnd2modname(lnd_type));
-                        LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex);
+                        LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
+                        rc = cfs_request_module("%s",
+                                                libcfs_lnd2modname(lnd_type));
+                        LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
 
                         lnd = lnet_find_lnd_by_type(lnd_type);
                         if (lnd == NULL) {
-                                LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
+                                LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
                                        libcfs_lnd2str(lnd_type),
                                        libcfs_lnd2modname(lnd_type), rc);
-#ifndef CONFIG_KMOD
+#ifndef HAVE_MODULE_LOADING_SUPPORT
                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
-                                         "compiled with CONFIG_KMOD set for "
-                                         "automatic module loading.");
+                                         "compiled with kernel module "
+                                         "loading support.");
 #endif
                                 goto failed;
                         }
                 }
 #else
                 if (lnd == NULL) {
-                        LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
+                        LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
                         CERROR("LND %s not supported\n",
                                libcfs_lnd2str(lnd_type));
                         goto failed;
@@ -1028,7 +964,7 @@ lnet_startup_lndnis (void)
 
                 rc = (lnd->lnd_startup)(ni);
 
-                LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
+                LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
 
                 if (rc != 0) {
                         LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s"
@@ -1040,10 +976,12 @@ lnet_startup_lndnis (void)
                         goto failed;
                 }
 
-                list_del(&ni->ni_list);
+                LASSERT (ni->ni_peertimeout <= 0 || lnd->lnd_query != NULL);
+
+                cfs_list_del(&ni->ni_list);
 
                 LNET_LOCK();
-                list_add_tail(&ni->ni_list, &the_lnet.ln_nis);
+                cfs_list_add_tail(&ni->ni_list, &the_lnet.ln_nis);
                 LNET_UNLOCK();
 
                 if (lnd->lnd_type == LOLND) {
@@ -1055,10 +993,10 @@ lnet_startup_lndnis (void)
 
 #ifndef __KERNEL__
                 if (lnd->lnd_wait != NULL) {
-                        if (the_lnet.ln_eqwaitni == NULL) {
-                                lnet_ni_addref(ni);
-                                the_lnet.ln_eqwaitni = ni;
-                        }
+                       if (the_lnet.ln_eq_waitni == NULL) {
+                               lnet_ni_addref(ni);
+                               the_lnet.ln_eq_waitni = ni;
+                       }
                 } else {
 # ifndef HAVE_LIBPTHREAD
                         LCONSOLE_ERROR_MSG(0x106, "LND %s not supported in a "
@@ -1079,15 +1017,16 @@ lnet_startup_lndnis (void)
 
                 ni->ni_txcredits = ni->ni_mintxcredits = ni->ni_maxtxcredits;
 
-                CDEBUG(D_LNI, "Added LNI %s [%d/%d]\n",
+                CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
                        libcfs_nid2str(ni->ni_nid),
-                       ni->ni_peertxcredits, ni->ni_txcredits);
+                       ni->ni_peertxcredits, ni->ni_txcredits,
+                       ni->ni_peerrtrcredits, ni->ni_peertimeout);
 
                 nicount++;
         }
 
-        if (the_lnet.ln_eqwaitni != NULL && nicount > 1) {
-                lnd_type = the_lnet.ln_eqwaitni->ni_lnd->lnd_type;
+       if (the_lnet.ln_eq_waitni != NULL && nicount > 1) {
+               lnd_type = the_lnet.ln_eq_waitni->ni_lnd->lnd_type;
                 LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network"
                                    "\n",
                                    libcfs_lnd2str(lnd_type));
@@ -1099,15 +1038,25 @@ lnet_startup_lndnis (void)
  failed:
         lnet_shutdown_lndnis();
 
-        while (!list_empty(&nilist)) {
-                ni = list_entry(nilist.next, lnet_ni_t, ni_list);
-                list_del(&ni->ni_list);
+        while (!cfs_list_empty(&nilist)) {
+                ni = cfs_list_entry(nilist.next, lnet_ni_t, ni_list);
+                cfs_list_del(&ni->ni_list);
                 LIBCFS_FREE(ni, sizeof(*ni));
         }
 
         return -ENETDOWN;
 }
 
+/**
+ * Initialize LNet library.
+ *
+ * Only userspace program needs to call this function - it's automatically
+ * called in the kernel at module loading time. Caller has to call LNetFini()
+ * after a call to LNetInit(), if and only if the latter returned 0. It must
+ * be called exactly once.
+ *
+ * \return 0 on success, and -ve on failures.
+ */
 int
 LNetInit(void)
 {
@@ -1117,9 +1066,12 @@ LNetInit(void)
         memset(&the_lnet, 0, sizeof(the_lnet));
 
         lnet_init_locks();
-        CFS_INIT_LIST_HEAD(&the_lnet.ln_lnds);
         the_lnet.ln_refcount = 0;
         the_lnet.ln_init = 1;
+        LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
+        CFS_INIT_LIST_HEAD(&the_lnet.ln_lnds);
+       CFS_INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
+       CFS_INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
 
 #ifdef __KERNEL__
         /* All LNDs apart from the LOLND are in separate modules.  They
@@ -1139,27 +1091,52 @@ LNetInit(void)
         return 0;
 }
 
+/**
+ * Finalize LNet library.
+ *
+ * Only userspace program needs to call this function. It can be called
+ * at most once.
+ *
+ * \pre LNetInit() called with success.
+ * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
+ */
 void
 LNetFini(void)
 {
         LASSERT (the_lnet.ln_init);
         LASSERT (the_lnet.ln_refcount == 0);
 
-        while (!list_empty(&the_lnet.ln_lnds))
-                lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
-                                               lnd_t, lnd_list));
+        while (!cfs_list_empty(&the_lnet.ln_lnds))
+                lnet_unregister_lnd(cfs_list_entry(the_lnet.ln_lnds.next,
+                                                   lnd_t, lnd_list));
         lnet_fini_locks();
 
         the_lnet.ln_init = 0;
 }
 
+/**
+ * Set LNet PID and start LNet interfaces, routing, and forwarding.
+ *
+ * Userspace program should call this after a successful call to LNetInit().
+ * Users must call this function at least once before any other functions.
+ * For each successful call there must be a corresponding call to
+ * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
+ * ignored.
+ *
+ * The PID used by LNet may be different from the one requested.
+ * See LNetGetId().
+ *
+ * \param requested_pid PID requested by the caller.
+ *
+ * \return >= 0 on success, and < 0 error code on failures.
+ */
 int
 LNetNIInit(lnet_pid_t requested_pid)
 {
         int         im_a_router = 0;
         int         rc;
 
-        LNET_MUTEX_DOWN(&the_lnet.ln_api_mutex);
+        LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
 
         LASSERT (the_lnet.ln_init);
         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
@@ -1169,6 +1146,8 @@ LNetNIInit(lnet_pid_t requested_pid)
                 goto out;
         }
 
+        lnet_get_tunables();
+
         if (requested_pid == LNET_PID_ANY) {
                 /* Don't instantiate LNET just for me */
                 rc = -ENETDOWN;
@@ -1202,11 +1181,13 @@ LNetNIInit(lnet_pid_t requested_pid)
         the_lnet.ln_refcount = 1;
         /* Now I may use my own API functions... */
 
-        rc = lnet_router_checker_start();
+        /* NB router checker needs the_lnet.ln_ping_info in
+        * lnet_router_checker -> lnet_update_ni_status_locked */
+        rc = lnet_ping_target_init();
         if (rc != 0)
                 goto failed3;
 
-        rc = lnet_ping_target_init();
+        rc = lnet_router_checker_start();
         if (rc != 0)
                 goto failed4;
 
@@ -1214,7 +1195,7 @@ LNetNIInit(lnet_pid_t requested_pid)
         goto out;
 
  failed4:
-        lnet_router_checker_stop();
+        lnet_ping_target_fini();
  failed3:
         the_lnet.ln_refcount = 0;
         lnet_acceptor_stop();
@@ -1226,14 +1207,23 @@ LNetNIInit(lnet_pid_t requested_pid)
  failed0:
         LASSERT (rc < 0);
  out:
-        LNET_MUTEX_UP(&the_lnet.ln_api_mutex);
+        LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
         return rc;
 }
 
+/**
+ * Stop LNet interfaces, routing, and forwarding.
+ *
+ * Users must call this function once for each successful call to LNetNIInit().
+ * Once the LNetNIFini() operation has been started, the results of pending
+ * API operations are undefined.
+ *
+ * \return always 0 for current implementation.
+ */
 int
 LNetNIFini()
 {
-        LNET_MUTEX_DOWN(&the_lnet.ln_api_mutex);
+        LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex);
 
         LASSERT (the_lnet.ln_init);
         LASSERT (the_lnet.ln_refcount > 0);
@@ -1244,8 +1234,8 @@ LNetNIFini()
                 LASSERT (!the_lnet.ln_niinit_self);
 
                 lnet_proc_fini();
-                lnet_ping_target_fini();
                 lnet_router_checker_stop();
+                lnet_ping_target_fini();
 
                 /* Teardown fns that use my own API functions BEFORE here */
                 the_lnet.ln_refcount = 0;
@@ -1256,15 +1246,28 @@ LNetNIFini()
                 lnet_unprepare();
         }
 
-        LNET_MUTEX_UP(&the_lnet.ln_api_mutex);
+        LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
         return 0;
 }
 
+/**
+ * This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and
+ * IOC_LIBCFS_PORTALS_COMPATIBILITY commands to users, by tweaking the LNet
+ * internal ioctl handler.
+ *
+ * IOC_LIBCFS_PORTALS_COMPATIBILITY is now deprecated, don't use it.
+ *
+ * \param cmd IOC_LIBCFS_DEBUG_PEER to print debugging data about a peer.
+ * The data will be printed to system console. Don't use it excessively.
+ * \param arg A pointer to lnet_process_id_t, process ID of the peer.
+ *
+ * \return Always return 0 when called by users directly (i.e., not via ioctl).
+ */
 int
 LNetCtl(unsigned int cmd, void *arg)
 {
         struct libcfs_ioctl_data *data = arg;
-        lnet_process_id_t         id;
+        lnet_process_id_t         id = {0};
         lnet_ni_t                *ni;
         int                       rc;
 
@@ -1294,7 +1297,9 @@ LNetCtl(unsigned int cmd, void *arg)
                                       &data->ioc_nid, &data->ioc_flags);
         case IOC_LIBCFS_NOTIFY_ROUTER:
                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
-                                   (time_t)data->ioc_u64[0]);
+                                   cfs_time_current() -
+                                   cfs_time_seconds(cfs_time_current_sec() -
+                                                    (time_t)data->ioc_u64[0]));
 
         case IOC_LIBCFS_PORTALS_COMPATIBILITY:
                 /* This can be removed once lustre stops calling it */
@@ -1315,9 +1320,9 @@ LNetCtl(unsigned int cmd, void *arg)
                 return 0;
 
         case IOC_LIBCFS_PING:
-                rc = lnet_ping((lnet_process_id_t) {.nid = data->ioc_nid,
-                                                    .pid = data->ioc_u32[0]},
-                               data->ioc_u32[1], /* timeout */
+                id.nid = data->ioc_nid;
+                id.pid = data->ioc_u32[0];
+                rc = lnet_ping(id, data->ioc_u32[1], /* timeout */
                                (lnet_process_id_t *)data->ioc_pbuf1,
                                data->ioc_plen1/sizeof(lnet_process_id_t));
                 if (rc < 0)
@@ -1328,17 +1333,17 @@ LNetCtl(unsigned int cmd, void *arg)
         case IOC_LIBCFS_DEBUG_PEER: {
                 /* CAVEAT EMPTOR: this one designed for calling directly; not
                  * via an ioctl */
-                lnet_process_id_t *id = arg;
+                id = *((lnet_process_id_t *) arg);
 
-                lnet_debug_peer(id->nid);
+                lnet_debug_peer(id.nid);
 
-                ni = lnet_net2ni(LNET_NIDNET(id->nid));
+                ni = lnet_net2ni(LNET_NIDNET(id.nid));
                 if (ni == NULL) {
-                        CDEBUG(D_WARNING, "No NI for %s\n", libcfs_id2str(*id));
+                        CDEBUG(D_WARNING, "No NI for %s\n", libcfs_id2str(id));
                 } else {
                         if (ni->ni_lnd->lnd_ctl == NULL) {
                                 CDEBUG(D_WARNING, "No ctl for %s\n",
-                                       libcfs_id2str(*id));
+                                       libcfs_id2str(id));
                         } else {
                                 (void)ni->ni_lnd->lnd_ctl(ni, cmd, arg);
                         }
@@ -1364,11 +1369,22 @@ LNetCtl(unsigned int cmd, void *arg)
         /* not reached */
 }
 
+/**
+ * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
+ * all interfaces share a same PID, as requested by LNetNIInit().
+ *
+ * \param index Index of the interface to look up.
+ * \param id On successful return, this location will hold the
+ * lnet_process_id_t ID of the interface.
+ *
+ * \retval 0 If an interface exists at \a index.
+ * \retval -ENOENT If no interface has been found.
+ */
 int
 LNetGetId(unsigned int index, lnet_process_id_t *id)
 {
         lnet_ni_t        *ni;
-        struct list_head *tmp;
+        cfs_list_t       *tmp;
         int               rc = -ENOENT;
 
         LASSERT (the_lnet.ln_init);
@@ -1376,11 +1392,11 @@ LNetGetId(unsigned int index, lnet_process_id_t *id)
 
         LNET_LOCK();
 
-        list_for_each(tmp, &the_lnet.ln_nis) {
+        cfs_list_for_each(tmp, &the_lnet.ln_nis) {
                 if (index-- != 0)
                         continue;
 
-                ni = list_entry(tmp, lnet_ni_t, ni_list);
+                ni = cfs_list_entry(tmp, lnet_ni_t, ni_list);
 
                 id->nid = ni->ni_nid;
                 id->pid = the_lnet.ln_pid;
@@ -1393,23 +1409,26 @@ LNetGetId(unsigned int index, lnet_process_id_t *id)
         return rc;
 }
 
+/**
+ * Print a string representation of handle \a h into buffer \a str of
+ * \a len bytes.
+ */
 void
 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
 {
         snprintf(str, len, LPX64, h.cookie);
 }
 
-
-int
-lnet_ping_target_init(void)
+static int
+lnet_create_ping_info(void)
 {
-        lnet_handle_me_t  meh;
-        lnet_process_id_t id;
-        int               rc;
-        int               rc2;
-        int               n;
-        int               infosz;
         int               i;
+        int               n;
+        int               rc;
+        unsigned int      infosz;
+        lnet_ni_t        *ni;
+        lnet_process_id_t id;
+        lnet_ping_info_t *pinfo;
 
         for (n = 0; ; n++) {
                 rc = LNetGetId(n, &id);
@@ -1419,24 +1438,76 @@ lnet_ping_target_init(void)
                 LASSERT (rc == 0);
         }
 
-        infosz = offsetof(lnet_ping_info_t, pi_nid[n]);
-        LIBCFS_ALLOC(the_lnet.ln_ping_info, infosz);
-        if (the_lnet.ln_ping_info == NULL) {
+        infosz = offsetof(lnet_ping_info_t, pi_ni[n]);
+        LIBCFS_ALLOC(pinfo, infosz);
+        if (pinfo == NULL) {
                 CERROR("Can't allocate ping info[%d]\n", n);
                 return -ENOMEM;
         }
 
-        the_lnet.ln_ping_info->pi_magic   = LNET_PROTO_PING_MAGIC;
-        the_lnet.ln_ping_info->pi_version = LNET_PROTO_PING_VERSION;
-        the_lnet.ln_ping_info->pi_pid     = the_lnet.ln_pid;
-        the_lnet.ln_ping_info->pi_nnids   = n;
+        pinfo->pi_nnis    = n;
+        pinfo->pi_pid     = the_lnet.ln_pid;
+        pinfo->pi_magic   = LNET_PROTO_PING_MAGIC;
+        pinfo->pi_version = LNET_PROTO_PING_VERSION;
 
         for (i = 0; i < n; i++) {
+                lnet_ni_status_t *ns = &pinfo->pi_ni[i];
+
                 rc = LNetGetId(i, &id);
                 LASSERT (rc == 0);
-                the_lnet.ln_ping_info->pi_nid[i] = id.nid;
+
+                ns->ns_nid    = id.nid;
+                ns->ns_status = LNET_NI_STATUS_UP;
+
+                LNET_LOCK();
+
+                ni = lnet_nid2ni_locked(id.nid);
+                LASSERT (ni != NULL);
+                LASSERT (ni->ni_status == NULL);
+                ni->ni_status = ns;
+                lnet_ni_decref_locked(ni);
+
+                LNET_UNLOCK();
         }
 
+        the_lnet.ln_ping_info = pinfo;
+        return 0;
+}
+
+static void
+lnet_destroy_ping_info(void)
+{
+        lnet_ni_t *ni;
+
+        LNET_LOCK();
+
+        cfs_list_for_each_entry (ni, &the_lnet.ln_nis, ni_list) {
+                ni->ni_status = NULL;
+        }
+
+        LNET_UNLOCK();
+
+        LIBCFS_FREE(the_lnet.ln_ping_info,
+                    offsetof(lnet_ping_info_t,
+                             pi_ni[the_lnet.ln_ping_info->pi_nnis]));
+        the_lnet.ln_ping_info = NULL;
+        return;
+}
+
+int
+lnet_ping_target_init(void)
+{
+        lnet_md_t         md = {0};
+        lnet_handle_me_t  meh;
+        lnet_process_id_t id;
+        int               rc;
+        int               rc2;
+        int               infosz;
+
+        rc = lnet_create_ping_info();
+        if (rc != 0)
+                return rc;
+
         /* We can have a tiny EQ since we only need to see the unlink event on
          * teardown, which by definition is the last one! */
         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &the_lnet.ln_ping_target_eq);
@@ -1445,10 +1516,12 @@ lnet_ping_target_init(void)
                 goto failed_0;
         }
 
-        rc = LNetMEAttach(LNET_RESERVED_PORTAL,
-                          (lnet_process_id_t){.nid = LNET_NID_ANY,
-                                              .pid = LNET_PID_ANY},
-                          LNET_PROTO_PING_MATCHBITS, 0LL,
+        memset(&id, 0, sizeof(lnet_process_id_t));
+        id.nid = LNET_NID_ANY;
+        id.pid = LNET_PID_ANY;
+
+        rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
+                          LNET_PROTO_PING_MATCHBITS, 0,
                           LNET_UNLINK, LNET_INS_AFTER,
                           &meh);
         if (rc != 0) {
@@ -1456,14 +1529,19 @@ lnet_ping_target_init(void)
                 goto failed_1;
         }
 
-        rc = LNetMDAttach(meh,
-                          (lnet_md_t){.start = the_lnet.ln_ping_info,
-                                      .length = infosz,
-                                      .threshold = LNET_MD_THRESH_INF,
-                                      .options = (LNET_MD_OP_GET |
-                                                  LNET_MD_TRUNCATE |
-                                                  LNET_MD_MANAGE_REMOTE),
-                                      .eq_handle = the_lnet.ln_ping_target_eq},
+        /* initialize md content */
+        infosz = offsetof(lnet_ping_info_t,
+                          pi_ni[the_lnet.ln_ping_info->pi_nnis]);
+        md.start     = the_lnet.ln_ping_info;
+        md.length    = infosz;
+        md.threshold = LNET_MD_THRESH_INF;
+        md.max_size  = 0;
+        md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
+                       LNET_MD_MANAGE_REMOTE;
+        md.user_ptr  = NULL;
+        md.eq_handle = the_lnet.ln_ping_target_eq;
+
+        rc = LNetMDAttach(meh, md,
                           LNET_RETAIN,
                           &the_lnet.ln_ping_target_md);
         if (rc != 0) {
@@ -1480,8 +1558,7 @@ lnet_ping_target_init(void)
         rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
         LASSERT (rc2 == 0);
  failed_0:
-        LIBCFS_FREE(the_lnet.ln_ping_info, infosz);
-
+        lnet_destroy_ping_info();
         return rc;
 }
 
@@ -1518,11 +1595,7 @@ lnet_ping_target_fini(void)
 
         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
         LASSERT (rc == 0);
-
-        LIBCFS_FREE(the_lnet.ln_ping_info,
-                    offsetof(lnet_ping_info_t,
-                             pi_nid[the_lnet.ln_ping_info->pi_nnids]));
-
+        lnet_destroy_ping_info();
         cfs_restore_sigs(blocked);
 }
 
@@ -1532,11 +1605,12 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
         lnet_handle_eq_t     eqh;
         lnet_handle_md_t     mdh;
         lnet_event_t         event;
+        lnet_md_t            md = {0};
         int                  which;
         int                  unlinked = 0;
         int                  replied = 0;
         const int            a_long_time = 60000; /* mS */
-        int                  infosz = offsetof(lnet_ping_info_t, pi_nid[n_ids]);
+        int                  infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
         lnet_ping_info_t    *info;
         lnet_process_id_t    tmpid;
         int                  i;
@@ -1565,13 +1639,16 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
                 goto out_0;
         }
 
-        rc = LNetMDBind((lnet_md_t){.start = info,
-                                    .length = infosz,
-                                    .threshold = 2, /* GET/REPLY */
-                                    .options = LNET_MD_TRUNCATE,
-                                    .eq_handle = eqh},
-                        LNET_UNLINK,
-                        &mdh);
+        /* initialize md content */
+        md.start     = info;
+        md.length    = infosz;
+        md.threshold = 2; /*GET/REPLY*/
+        md.max_size  = 0;
+        md.options   = LNET_MD_TRUNCATE;
+        md.user_ptr  = NULL;
+        md.eq_handle = eqh;
+
+        rc = LNetMDBind(md, LNET_UNLINK, &mdh);
         if (rc != 0) {
                 CERROR("Can't bind MD: %d\n", rc);
                 goto out_1;
@@ -1627,7 +1704,6 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
                                 CWARN("ping %s: late network completion\n",
                                       libcfs_id2str(id));
                         }
-
                 } else if (event.type == LNET_EVENT_REPLY) {
                         replied = 1;
                         rc = event.mlength;
@@ -1656,14 +1732,7 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
         }
 
         if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
-                /* NB I might be swabbing garbage until I check below, but it
-                 * doesn't matter */
-                __swab32s(&info->pi_version);
-                __swab32s(&info->pi_pid);
-                __swab32s(&info->pi_nnids);
-                for (i = 0; i < info->pi_nnids && i < n_ids; i++)
-                        __swab64s(&info->pi_nid[i]);
-
+                lnet_swap_pinginfo(info);
         } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
                 CERROR("%s: Unexpected magic %08x\n", 
                        libcfs_id2str(id), info->pi_magic);
@@ -1676,18 +1745,18 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
                 goto out_1;
         }
 
-        if (nob < offsetof(lnet_ping_info_t, pi_nid[0])) {
+        if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) {
                 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
-                       nob, (int)offsetof(lnet_ping_info_t, pi_nid[0]));
+                       nob, (int)offsetof(lnet_ping_info_t, pi_ni[0]));
                 goto out_1;
         }
 
-        if (info->pi_nnids < n_ids)
-                n_ids = info->pi_nnids;
+        if (info->pi_nnis < n_ids)
+                n_ids = info->pi_nnis;
 
-        if (nob < offsetof(lnet_ping_info_t, pi_nid[n_ids])) {
+        if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) {
                 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
-                       nob, (int)offsetof(lnet_ping_info_t, pi_nid[n_ids]));
+                       nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids]));
                 goto out_1;
         }
 
@@ -1695,15 +1764,15 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
 
         for (i = 0; i < n_ids; i++) {
                 tmpid.pid = info->pi_pid;
-                tmpid.nid = info->pi_nid[i];
+                tmpid.nid = info->pi_ni[i].ns_nid;
 #ifdef __KERNEL__
-                if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
+                if (cfs_copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
                         goto out_1;
 #else
                 ids[i] = tmpid;
 #endif
         }
-        rc = info->pi_nnids;
+        rc = info->pi_nnis;
 
  out_1:
         rc2 = LNetEQFree(eqh);