Whamcloud - gitweb
LU-12036 ofd: add "no_precreate" mount option
[fs/lustre-release.git] / lnet / lnet / api-ni.c
index 99ccfd7..d6e5e3d 100644 (file)
@@ -221,6 +221,7 @@ MODULE_PARM_DESC(lnet_retry_count,
 
 
 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
+unsigned int lnet_current_net_count;
 
 /*
  * This sequence number keeps track of how many times DLC was used to
@@ -545,7 +546,7 @@ lnet_init_locks(void)
        spin_lock_init(&the_lnet.ln_eq_wait_lock);
        spin_lock_init(&the_lnet.ln_msg_resend_lock);
        init_waitqueue_head(&the_lnet.ln_eq_waitq);
-       init_waitqueue_head(&the_lnet.ln_mt_waitq);
+       init_completion(&the_lnet.ln_mt_wait_complete);
        mutex_init(&the_lnet.ln_lnd_mutex);
 }
 
@@ -827,7 +828,6 @@ lnet_register_lnd(struct lnet_lnd *lnd)
        LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 
        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));
 
@@ -841,7 +841,6 @@ lnet_unregister_lnd(struct lnet_lnd *lnd)
        mutex_lock(&the_lnet.ln_lnd_mutex);
 
        LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
-       LASSERT(lnd->lnd_refcount == 0);
 
        list_del(&lnd->lnd_list);
        CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
@@ -1099,6 +1098,26 @@ lnet_res_lh_initialize(struct lnet_res_container *rec,
        list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
 }
 
+struct list_head **
+lnet_create_array_of_queues(void)
+{
+       struct list_head **qs;
+       struct list_head *q;
+       int i;
+
+       qs = cfs_percpt_alloc(lnet_cpt_table(),
+                             sizeof(struct list_head));
+       if (!qs) {
+               CERROR("Failed to allocate queues\n");
+               return NULL;
+       }
+
+       cfs_percpt_for_each(q, i, qs)
+               INIT_LIST_HEAD(q);
+
+       return qs;
+}
+
 static int lnet_unprepare(void);
 
 static int
@@ -1133,6 +1152,7 @@ lnet_prepare(lnet_pid_t requested_pid)
        INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
        init_waitqueue_head(&the_lnet.ln_dc_waitq);
        LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
+       init_completion(&the_lnet.ln_started);
 
        rc = lnet_descriptor_setup();
        if (rc != 0)
@@ -1191,6 +1211,12 @@ lnet_prepare(lnet_pid_t requested_pid)
                goto failed;
        }
 
+       the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
+       if (!the_lnet.ln_mt_zombie_rstqs) {
+               rc = -ENOMEM;
+               goto failed;
+       }
+
        return 0;
 
  failed:
@@ -1214,6 +1240,11 @@ lnet_unprepare (void)
        LASSERT(list_empty(&the_lnet.ln_test_peers));
        LASSERT(list_empty(&the_lnet.ln_nets));
 
+       if (the_lnet.ln_mt_zombie_rstqs) {
+               lnet_clean_zombie_rstqs();
+               the_lnet.ln_mt_zombie_rstqs = NULL;
+       }
+
        if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
                rc = LNetEQFree(the_lnet.ln_mt_eqh);
                LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
@@ -1368,18 +1399,28 @@ lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
 EXPORT_SYMBOL(lnet_cpt_of_nid);
 
 int
-lnet_islocalnet(__u32 net_id)
+lnet_islocalnet_locked(__u32 net_id)
 {
        struct lnet_net *net;
-       int             cpt;
-       bool            local;
-
-       cpt = lnet_net_lock_current();
+       bool local;
 
        net = lnet_get_net_locked(net_id);
 
        local = net != NULL;
 
+       return local;
+}
+
+int
+lnet_islocalnet(__u32 net_id)
+{
+       int cpt;
+       bool local;
+
+       cpt = lnet_net_lock_current();
+
+       local = lnet_islocalnet_locked(net_id);
+
        lnet_net_unlock(cpt);
 
        return local;
@@ -1537,6 +1578,44 @@ lnet_get_ni_count(void)
 }
 
 int
+lnet_get_net_count(void)
+{
+       struct lnet_net *net;
+       int count = 0;
+
+       lnet_net_lock(0);
+
+       list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
+               count++;
+       }
+
+       lnet_net_unlock(0);
+
+       return count;
+}
+
+void
+lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
+{
+       struct lnet_ni_status *stat;
+       int nnis;
+       int i;
+
+       __swab32s(&pbuf->pb_info.pi_magic);
+       __swab32s(&pbuf->pb_info.pi_features);
+       __swab32s(&pbuf->pb_info.pi_pid);
+       __swab32s(&pbuf->pb_info.pi_nnis);
+       nnis = pbuf->pb_info.pi_nnis;
+       if (nnis > pbuf->pb_nnis)
+               nnis = pbuf->pb_nnis;
+       for (i = 0; i < nnis; i++) {
+               stat = &pbuf->pb_info.pi_ni[i];
+               __swab64s(&stat->ns_nid);
+               __swab32s(&stat->ns_status);
+       }
+}
+
+int
 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
 {
        if (!pinfo)
@@ -2032,8 +2111,6 @@ lnet_shutdown_lndnet(struct lnet_net *net)
 
        lnet_net_lock(LNET_LOCK_EX);
 
-       net->net_state = LNET_NET_STATE_DELETING;
-
        list_del_init(&net->net_list);
 
        while (!list_empty(&net->net_ni_list)) {
@@ -2049,15 +2126,6 @@ lnet_shutdown_lndnet(struct lnet_net *net)
        /* Do peer table cleanup for this net */
        lnet_peer_tables_cleanup(net);
 
-       lnet_net_lock(LNET_LOCK_EX);
-       /*
-        * decrement ref count on lnd only when the entire network goes
-        * away
-        */
-       net->net_lnd->lnd_refcount--;
-
-       lnet_net_unlock(LNET_LOCK_EX);
-
        lnet_net_free(net);
 }
 
@@ -2141,9 +2209,6 @@ lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
        if (rc != 0) {
                LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
                                   rc, libcfs_lnd2str(net->net_lnd->lnd_type));
-               lnet_net_lock(LNET_LOCK_EX);
-               net->net_lnd->lnd_refcount--;
-               lnet_net_unlock(LNET_LOCK_EX);
                goto failed0;
        }
 
@@ -2252,10 +2317,6 @@ lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
                        }
                }
 
-               lnet_net_lock(LNET_LOCK_EX);
-               lnd->lnd_refcount++;
-               lnet_net_unlock(LNET_LOCK_EX);
-
                net->net_lnd = lnd;
 
                mutex_unlock(&the_lnet.ln_lnd_mutex);
@@ -2306,7 +2367,7 @@ lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
                 * up is actually unique. if it's not fail. */
                if (!lnet_ni_unique_net(&net_l->net_ni_list,
                                        ni->ni_interfaces[0])) {
-                       rc = -EINVAL;
+                       rc = -EEXIST;
                        goto failed1;
                }
 
@@ -2343,7 +2404,6 @@ lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
                 */
                lnet_net_free(net);
        } else {
-               net->net_state = LNET_NET_STATE_ACTIVE;
                /*
                 * restore tunables after it has been overwitten by the
                 * lnd
@@ -2360,6 +2420,9 @@ lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
                lnet_net_unlock(LNET_LOCK_EX);
        }
 
+       /* update net count */
+       lnet_current_net_count = lnet_get_net_count();
+
        return ni_count;
 
 failed1:
@@ -2454,12 +2517,9 @@ int lnet_lib_init(void)
        }
 
        the_lnet.ln_refcount = 0;
-       LNetInvalidateEQHandle(&the_lnet.ln_rc_eqh);
        INIT_LIST_HEAD(&the_lnet.ln_lnds);
        INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
-       INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
        INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
-       INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
 
        /* The hash table size is the number of bits it takes to express the set
         * ln_num_routes, minus 1 (better to under estimate than over so we
@@ -2599,29 +2659,34 @@ LNetNIInit(lnet_pid_t requested_pid)
                goto err_stop_ping;
        }
 
-       rc = lnet_monitor_thr_start();
-       if (rc != 0)
-               goto err_stop_ping;
-
        rc = lnet_push_target_init();
        if (rc != 0)
-               goto err_stop_monitor_thr;
+               goto err_stop_ping;
 
        rc = lnet_peer_discovery_start();
        if (rc != 0)
                goto err_destroy_push_target;
 
+       rc = lnet_monitor_thr_start();
+       if (rc != 0)
+               goto err_stop_discovery_thr;
+
        lnet_fault_init();
        lnet_router_debugfs_init();
 
        mutex_unlock(&the_lnet.ln_api_mutex);
 
+       complete_all(&the_lnet.ln_started);
+
+       /* wait for all routers to start */
+       lnet_wait_router_start();
+
        return 0;
 
+err_stop_discovery_thr:
+       lnet_peer_discovery_stop();
 err_destroy_push_target:
        lnet_push_target_fini();
-err_stop_monitor_thr:
-       lnet_monitor_thr_stop();
 err_stop_ping:
        lnet_ping_target_fini();
 err_acceptor_stop:
@@ -2657,7 +2722,7 @@ EXPORT_SYMBOL(LNetNIInit);
  * \return always 0 for current implementation.
  */
 int
-LNetNIFini()
+LNetNIFini(void)
 {
        mutex_lock(&the_lnet.ln_api_mutex);
 
@@ -2671,9 +2736,9 @@ LNetNIFini()
                lnet_fault_fini();
 
                lnet_router_debugfs_fini();
+               lnet_monitor_thr_stop();
                lnet_peer_discovery_stop();
                lnet_push_target_fini();
-               lnet_monitor_thr_stop();
                lnet_ping_target_fini();
 
                /* Teardown fns that use my own API functions BEFORE here */
@@ -3511,20 +3576,28 @@ LNetCtl(unsigned int cmd, void *arg)
        case IOC_LIBCFS_FAIL_NID:
                return lnet_fail_nid(data->ioc_nid, data->ioc_count);
 
-       case IOC_LIBCFS_ADD_ROUTE:
+       case IOC_LIBCFS_ADD_ROUTE: {
+               /* default router sensitivity to 1 */
+               unsigned int sensitivity = 1;
                config = arg;
 
                if (config->cfg_hdr.ioc_len < sizeof(*config))
                        return -EINVAL;
 
+               if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
+                       sensitivity =
+                         config->cfg_config_u.cfg_route.rtr_sensitivity;
+               }
+
                mutex_lock(&the_lnet.ln_api_mutex);
                rc = lnet_add_route(config->cfg_net,
                                    config->cfg_config_u.cfg_route.rtr_hop,
                                    config->cfg_nid,
                                    config->cfg_config_u.cfg_route.
-                                       rtr_priority);
+                                       rtr_priority, sensitivity);
                mutex_unlock(&the_lnet.ln_api_mutex);
                return rc;
+       }
 
        case IOC_LIBCFS_DEL_ROUTE:
                config = arg;
@@ -3550,7 +3623,9 @@ LNetCtl(unsigned int cmd, void *arg)
                                    &config->cfg_nid,
                                    &config->cfg_config_u.cfg_route.rtr_flags,
                                    &config->cfg_config_u.cfg_route.
-                                       rtr_priority);
+                                       rtr_priority,
+                                   &config->cfg_config_u.cfg_route.
+                                       rtr_sensitivity);
                mutex_unlock(&the_lnet.ln_api_mutex);
                return rc;
 
@@ -3824,7 +3899,7 @@ LNetCtl(unsigned int cmd, void *arg)
                 * that deadline to the wall clock.
                 */
                deadline += ktime_get_seconds();
-               return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
+               return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
                                   deadline);
        }
 
@@ -3854,9 +3929,9 @@ LNetCtl(unsigned int cmd, void *arg)
                /* If timeout is negative then set default of 3 minutes */
                if (((s32)data->ioc_u32[1] <= 0) ||
                    data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
-                       timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
+                       timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
                else
-                       timeout = msecs_to_jiffies(data->ioc_u32[1]);
+                       timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
 
                rc = lnet_ping(id, timeout, data->ioc_pbuf1,
                               data->ioc_plen1 / sizeof(struct lnet_process_id));
@@ -3876,9 +3951,9 @@ LNetCtl(unsigned int cmd, void *arg)
                /* If timeout is negative then set default of 3 minutes */
                if (((s32)ping->op_param) <= 0 ||
                    ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
-                       timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
+                       timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
                else
-                       timeout = msecs_to_jiffies(ping->op_param);
+                       timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
 
                rc = lnet_ping(ping->ping_id, timeout,
                               ping->ping_buf,
@@ -4024,7 +4099,7 @@ static int lnet_ping(struct lnet_process_id id, signed long timeout,
        int which;
        int unlinked = 0;
        int replied = 0;
-       const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
+       const signed long a_long_time = cfs_time_seconds(60);
        struct lnet_ping_buffer *pbuf;
        struct lnet_process_id tmpid;
        int i;