Whamcloud - gitweb
b=23179 MDS node unresponsive
[fs/lustre-release.git] / lnet / lnet / api-ni.c
index 58839c7..3b97ea5 100644 (file)
@@ -26,7 +26,7 @@
  * 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.
  */
 /*
@@ -91,10 +91,10 @@ lnet_get_networks(void)
 void
 lnet_init_locks(void)
 {
-        spin_lock_init (&the_lnet.ln_lock);
+        cfs_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_init_mutex(&the_lnet.ln_lnd_mutex);
+        cfs_init_mutex(&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) {
@@ -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));
@@ -303,11 +303,11 @@ 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 ((int)lnd->lnd_type == type)
                         return lnd;
@@ -325,7 +325,7 @@ lnet_register_lnd (lnd_t *lnd)
         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));
@@ -342,7 +342,7 @@ lnet_unregister_lnd (lnd_t *lnd)
         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);
@@ -384,7 +384,7 @@ 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 +394,7 @@ 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)
@@ -466,7 +466,7 @@ 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;
@@ -481,12 +481,13 @@ lnet_setup_handle_hash (void)
 
         /* Arbitrary choice of hash table size */
 #ifdef __KERNEL__
-        the_lnet.ln_lh_hash_size = CFS_PAGE_SIZE / sizeof (struct list_head);
+        the_lnet.ln_lh_hash_size =
+                (2 * CFS_PAGE_SIZE) / sizeof (cfs_list_t);
 #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));
+                     the_lnet.ln_lh_hash_size * sizeof (cfs_list_t));
         if (the_lnet.ln_lh_hash_table == NULL)
                 return (-ENOMEM);
 
@@ -505,26 +506,26 @@ lnet_cleanup_handle_hash (void)
                 return;
 
         LIBCFS_FREE(the_lnet.ln_lh_hash_table,
-                    the_lnet.ln_lh_hash_size * sizeof (struct list_head));
+                    the_lnet.ln_lh_hash_size * sizeof (cfs_list_t));
 }
 
 lnet_libhandle_t *
 lnet_lookup_cookie (__u64 cookie, int type)
 {
         /* ALWAYS called with LNET_LOCK held */
-        struct list_head    *list;
-        struct list_head    *el;
+        cfs_list_t          *list;
+        cfs_list_t          *el;
         unsigned int         hash;
 
         if ((cookie & (LNET_COOKIE_TYPES - 1)) != type)
                 return (NULL);
 
-        hash = ((unsigned int)cookie) % the_lnet.ln_lh_hash_size;
+        hash = ((unsigned int)(cookie >> LNET_COOKIE_TYPE_BITS)) % the_lnet.ln_lh_hash_size;
         list = &the_lnet.ln_lh_hash_table[hash];
 
-        list_for_each (el, list) {
-                lnet_libhandle_t *lh = list_entry (el, lnet_libhandle_t,
-                                                  lh_hash_chain);
+        cfs_list_for_each (el, list) {
+                lnet_libhandle_t *lh = cfs_list_entry (el, lnet_libhandle_t,
+                                                      lh_hash_chain);
 
                 if (lh->lh_cookie == cookie)
                         return (lh);
@@ -543,15 +544,48 @@ lnet_initialise_handle (lnet_libhandle_t *lh, int type)
         lh->lh_cookie = the_lnet.ln_next_object_cookie | type;
         the_lnet.ln_next_object_cookie += LNET_COOKIE_TYPES;
 
-        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]);
+        hash = ((unsigned int)(lh->lh_cookie >> LNET_COOKIE_TYPE_BITS)) % the_lnet.ln_lh_hash_size;
+        cfs_list_add (&lh->lh_hash_chain, &the_lnet.ln_lh_hash_table[hash]);
 }
 
 void
 lnet_invalidate_handle (lnet_libhandle_t *lh)
 {
         /* ALWAYS called with LNET_LOCK held */
-        list_del (&lh->lh_hash_chain);
+        cfs_list_del (&lh->lh_hash_chain);
+}
+
+cfs_list_t *
+lnet_portal_mhash_alloc(void)
+{
+        cfs_list_t       *mhash;
+        int               i;
+
+        LIBCFS_ALLOC(mhash, sizeof(cfs_list_t) * LNET_PORTAL_HASH_SIZE);
+        if (mhash == NULL)
+                return NULL;
+
+        for (i = 0; i < LNET_PORTAL_HASH_SIZE; i++)
+                CFS_INIT_LIST_HEAD(&mhash[i]);
+
+        return mhash;
+}
+
+void
+lnet_portal_mhash_free(cfs_list_t *mhash)
+{
+        int     i;
+
+        for (i = 0; i < LNET_PORTAL_HASH_SIZE; i++) {
+                while (!cfs_list_empty(&mhash[i])) {
+                        lnet_me_t *me = cfs_list_entry(mhash[i].next,
+                                                       lnet_me_t, me_list);
+                        CERROR ("Active ME %p on exit portal mhash\n", me);
+                        cfs_list_del(&me->me_list);
+                        lnet_me_free(me);
+                }
+        }
+        LIBCFS_FREE(mhash, sizeof(cfs_list_t) * LNET_PORTAL_HASH_SIZE);
 }
 
 int
@@ -560,7 +594,7 @@ lnet_init_finalizers(void)
 #ifdef __KERNEL__
         int    i;
 
-        the_lnet.ln_nfinalizers = (int) num_online_cpus();
+        the_lnet.ln_nfinalizers = (int) cfs_num_online_cpus();
 
         LIBCFS_ALLOC(the_lnet.ln_finalizers,
                      the_lnet.ln_nfinalizers *
@@ -595,11 +629,13 @@ lnet_fini_finalizers(void)
 #else
         LASSERT (!the_lnet.ln_finalizing);
 #endif
-        LASSERT (list_empty(&the_lnet.ln_finalizeq));
+        LASSERT (cfs_list_empty(&the_lnet.ln_finalizeq));
 }
 
 #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 */
 
@@ -680,7 +716,7 @@ lnet_prepare(lnet_pid_t requested_pid)
         }
 
         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_mlist));
                 CFS_INIT_LIST_HEAD(&(the_lnet.ln_portals[i].ptl_msgq));
                 the_lnet.ln_portals[i].ptl_options = 0;
         }
@@ -710,51 +746,57 @@ 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));
+                lnet_portal_t *ptl = &the_lnet.ln_portals[idx];
 
-                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);
+                LASSERT (cfs_list_empty(&ptl->ptl_msgq));
 
-                        CERROR ("Active me %p on exit\n", me);
-                        list_del (&me->me_list);
+                while (!cfs_list_empty(&ptl->ptl_mlist)) {
+                        lnet_me_t *me = cfs_list_entry(ptl->ptl_mlist.next,
+                                                       lnet_me_t, me_list);
+                        CERROR ("Active ME %p on exit\n", me);
+                        cfs_list_del (&me->me_list);
                         lnet_me_free (me);
                 }
+
+                if (ptl->ptl_mhash != NULL) {
+                        LASSERT (lnet_portal_is_unique(ptl));
+                        lnet_portal_mhash_free(ptl->ptl_mhash);
+                }
         }
 
-        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);
+        while (!cfs_list_empty (&the_lnet.ln_active_mds)) {
+                lnet_libmd_t *md = cfs_list_entry (the_lnet.ln_active_mds.next,
+                                                   lnet_libmd_t, md_list);
 
-                CERROR ("Active md %p on exit\n", md);
-                list_del_init (&md->md_list);
+                CERROR ("Active MD %p on exit\n", md);
+                cfs_list_del_init (&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);
+        while (!cfs_list_empty (&the_lnet.ln_active_eqs)) {
+                lnet_eq_t *eq = cfs_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);
+                CERROR ("Active EQ %p on exit\n", eq);
+                cfs_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);
+        while (!cfs_list_empty (&the_lnet.ln_active_msgs)) {
+                lnet_msg_t *msg = cfs_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);
+                cfs_list_del (&msg->msg_activelist);
                 lnet_msg_free (msg);
         }
 
@@ -773,11 +815,11 @@ lnet_unprepare (void)
 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 +847,11 @@ 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 +880,14 @@ 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++;
@@ -869,18 +911,18 @@ 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 */
@@ -915,7 +957,7 @@ lnet_shutdown_lndnis (void)
         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 +967,16 @@ 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 +1007,7 @@ 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;
@@ -980,8 +1022,8 @@ 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));
@@ -992,7 +1034,8 @@ lnet_startup_lndnis (void)
 #ifdef __KERNEL__
                 if (lnd == NULL) {
                         LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex);
-                        rc = request_module(libcfs_lnd2modname(lnd_type));
+                        rc = cfs_request_module("%s",
+                                                libcfs_lnd2modname(lnd_type));
                         LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex);
 
                         lnd = lnet_find_lnd_by_type(lnd_type);
@@ -1001,10 +1044,10 @@ lnet_startup_lndnis (void)
                                 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;
                         }
@@ -1042,10 +1085,10 @@ lnet_startup_lndnis (void)
 
                 LASSERT (ni->ni_peertimeout <= 0 || lnd->lnd_query != NULL);
 
-                list_del(&ni->ni_list);
+                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) {
@@ -1081,10 +1124,10 @@ lnet_startup_lndnis (void)
 
                 ni->ni_txcredits = ni->ni_mintxcredits = ni->ni_maxtxcredits;
 
-                CDEBUG(D_LNI, "Added LNI %s [%d/%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_peertimeout);
+                       ni->ni_peerrtrcredits, ni->ni_peertimeout);
 
                 nicount++;
         }
@@ -1102,15 +1145,25 @@ 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)
 {
@@ -1120,9 +1173,11 @@ 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_zombie_rcd);
 
 #ifdef __KERNEL__
         /* All LNDs apart from the LOLND are in separate modules.  They
@@ -1142,20 +1197,45 @@ 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 non-zero error code on failures.
+ */
 int
 LNetNIInit(lnet_pid_t requested_pid)
 {
@@ -1172,6 +1252,8 @@ 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;
@@ -1205,11 +1287,13 @@ 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 */
+        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;
 
@@ -1217,7 +1301,7 @@ 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();
@@ -1233,6 +1317,15 @@ LNetNIInit(lnet_pid_t requested_pid)
         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()
 {
@@ -1247,8 +1340,8 @@ 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;
@@ -1263,6 +1356,19 @@ LNetNIFini()
         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)
 {
@@ -1297,7 +1403,9 @@ 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 */
@@ -1367,11 +1475,22 @@ 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);
@@ -1379,11 +1498,11 @@ 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;
@@ -1396,24 +1515,26 @@ 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;
-        lnet_md_t         md = {0};
-        int               rc;
-        int               rc2;
+        int               i;
         int               n;
+        int               rc;
         unsigned int      infosz;
-        int               i;
+        lnet_ni_t        *ni;
+        lnet_process_id_t id;
+        lnet_ping_info_t *pinfo;
 
         for (n = 0; ; n++) {
                 rc = LNetGetId(n, &id);
@@ -1423,24 +1544,76 @@ 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);
@@ -1463,6 +1636,8 @@ lnet_ping_target_init(void)
         }
 
         /* 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;
@@ -1489,8 +1664,7 @@ 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;
 }
 
@@ -1527,11 +1701,7 @@ 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);
 }
 
@@ -1546,7 +1716,7 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
         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;
@@ -1640,7 +1810,6 @@ 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;
@@ -1669,14 +1838,7 @@ 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 < (int)info->pi_nnids && i < (int)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);
@@ -1689,18 +1851,18 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i
                 goto out_1;
         }
 
-        if (nob < (int)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 ((int) info->pi_nnids < n_ids)
-                n_ids = info->pi_nnids;
+        if (info->pi_nnis < n_ids)
+                n_ids = info->pi_nnis;
 
-        if (nob < (int)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;
         }
 
@@ -1708,15 +1870,15 @@ 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);