]> git.karo-electronics.de Git - linux-beck.git/commitdiff
staging: lustre: DLC Feature dynamic net config
authorAmir Shehata <amir.shehata@intel.com>
Mon, 15 Feb 2016 15:25:54 +0000 (10:25 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 20 Feb 2016 22:29:23 +0000 (14:29 -0800)
This is the third patch of a set of patches that enables DLC.

This patch adds the following features to LNET.  Currently these
features are not driven by user space.
- Adding/Deleting Networks dynamically
Two new functions were added:
 - lnet_dyn_add_ni()
    add an NI. if the NI is already added then fail with
    appropriate error code
 - lnet_dyn_del_ni()
    delete an existing NI.  If NI doesn't exist fail with
    appropriate failure code.
These functions shall be called from IOCTL.

Signed-off-by: Amir Shehata <amir.shehata@intel.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-2456
Reviewed-on: http://review.whamcloud.com/9832
Reviewed-by: James Simmons <uja.ornl@gmail.com>
Reviewed-by: Doug Oucharek <doug.s.oucharek@intel.com>
Reviewed-by: Liang Zhen <liang.zhen@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/include/linux/lnet/lib-lnet.h
drivers/staging/lustre/lnet/lnet/api-ni.c
drivers/staging/lustre/lnet/lnet/config.c

index 3a1cf61018abc1d4ce6494dccb01ac277615f27b..0592e3059820f5087aa2e7bb69f5513d4cdc5891 100644 (file)
@@ -681,12 +681,10 @@ void lnet_router_checker_stop(void);
 void lnet_router_ni_update_locked(lnet_peer_t *gw, __u32 net);
 void lnet_swap_pinginfo(lnet_ping_info_t *info);
 
-int lnet_ping_target_init(void);
-void lnet_ping_target_fini(void);
-
 int lnet_parse_ip2nets(char **networksp, char *ip2nets);
 int lnet_parse_routes(char *route_str, int *im_a_router);
 int lnet_parse_networks(struct list_head *nilist, char *networks);
+int lnet_net_unique(__u32 net, struct list_head *nilist);
 
 int lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid, int cpt);
 lnet_peer_t *lnet_find_peer_locked(struct lnet_peer_table *ptable,
index 06046b216c93380552d6026de0725717bb69dc58..933f345e481f32cb471eff1526e31e063a9368a7 100644 (file)
@@ -533,6 +533,11 @@ lnet_prepare(lnet_pid_t requested_pid)
        struct lnet_res_container **recs;
        int rc = 0;
 
+       if (requested_pid == LNET_PID_ANY) {
+               /* Don't instantiate LNET just for me */
+               return -ENETDOWN;
+       }
+
        LASSERT(!the_lnet.ln_refcount);
 
        the_lnet.ln_routing = 0;
@@ -819,6 +824,229 @@ lnet_count_acceptor_nis(void)
        return count;
 }
 
+static lnet_ping_info_t *
+lnet_ping_info_create(int num_ni)
+{
+       lnet_ping_info_t *ping_info;
+       unsigned int infosz;
+
+       infosz = offsetof(lnet_ping_info_t, pi_ni[num_ni]);
+       LIBCFS_ALLOC(ping_info, infosz);
+       if (!ping_info) {
+               CERROR("Can't allocate ping info[%d]\n", num_ni);
+               return NULL;
+       }
+
+       ping_info->pi_nnis = num_ni;
+       ping_info->pi_pid = the_lnet.ln_pid;
+       ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
+       ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
+
+       return ping_info;
+}
+
+static inline int
+lnet_get_ni_count(void)
+{
+       struct lnet_ni *ni;
+       int count = 0;
+
+       lnet_net_lock(0);
+
+       list_for_each_entry(ni, &the_lnet.ln_nis, ni_list)
+               count++;
+
+       lnet_net_unlock(0);
+
+       return count;
+}
+
+static inline void
+lnet_ping_info_free(lnet_ping_info_t *pinfo)
+{
+       LIBCFS_FREE(pinfo,
+                   offsetof(lnet_ping_info_t,
+                            pi_ni[pinfo->pi_nnis]));
+}
+
+static void
+lnet_ping_info_destroy(void)
+{
+       struct lnet_ni *ni;
+
+       lnet_net_lock(LNET_LOCK_EX);
+
+       list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
+               lnet_ni_lock(ni);
+               ni->ni_status = NULL;
+               lnet_ni_unlock(ni);
+       }
+
+       lnet_ping_info_free(the_lnet.ln_ping_info);
+       the_lnet.ln_ping_info = NULL;
+
+       lnet_net_unlock(LNET_LOCK_EX);
+}
+
+static void
+lnet_ping_event_handler(lnet_event_t *event)
+{
+       lnet_ping_info_t *pinfo = event->md.user_ptr;
+
+       if (event->unlinked)
+               pinfo->pi_features = LNET_PING_FEAT_INVAL;
+}
+
+static int
+lnet_ping_info_setup(lnet_ping_info_t **ppinfo, lnet_handle_md_t *md_handle,
+                    int ni_count, bool set_eq)
+{
+       lnet_process_id_t id = {LNET_NID_ANY, LNET_PID_ANY};
+       lnet_handle_me_t me_handle;
+       lnet_md_t md = {0};
+       int rc, rc2;
+
+       if (set_eq) {
+               rc = LNetEQAlloc(0, lnet_ping_event_handler,
+                                &the_lnet.ln_ping_target_eq);
+               if (rc) {
+                       CERROR("Can't allocate ping EQ: %d\n", rc);
+                       return rc;
+               }
+       }
+
+       *ppinfo = lnet_ping_info_create(ni_count);
+       if (!*ppinfo) {
+               rc = -ENOMEM;
+               goto failed_0;
+       }
+
+       rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
+                         LNET_PROTO_PING_MATCHBITS, 0,
+                         LNET_UNLINK, LNET_INS_AFTER,
+                         &me_handle);
+       if (rc) {
+               CERROR("Can't create ping ME: %d\n", rc);
+               goto failed_1;
+       }
+
+       /* initialize md content */
+       md.start = *ppinfo;
+       md.length = offsetof(lnet_ping_info_t,
+                            pi_ni[(*ppinfo)->pi_nnis]);
+       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;
+       md.user_ptr = *ppinfo;
+
+       rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
+       if (rc) {
+               CERROR("Can't attach ping MD: %d\n", rc);
+               goto failed_2;
+       }
+
+       return 0;
+
+failed_2:
+       rc2 = LNetMEUnlink(me_handle);
+       LASSERT(!rc2);
+failed_1:
+       lnet_ping_info_free(*ppinfo);
+       *ppinfo = NULL;
+failed_0:
+       if (set_eq)
+               LNetEQFree(the_lnet.ln_ping_target_eq);
+       return rc;
+}
+
+static void
+lnet_ping_md_unlink(lnet_ping_info_t *pinfo, lnet_handle_md_t *md_handle)
+{
+       sigset_t blocked = cfs_block_allsigs();
+
+       LNetMDUnlink(*md_handle);
+       LNetInvalidateHandle(md_handle);
+
+       /* NB md could be busy; this just starts the unlink */
+       while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
+               CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
+               set_current_state(TASK_UNINTERRUPTIBLE);
+               schedule_timeout(cfs_time_seconds(1));
+       }
+
+       cfs_restore_sigs(blocked);
+}
+
+static void
+lnet_ping_info_install_locked(lnet_ping_info_t *ping_info)
+{
+       lnet_ni_status_t *ns;
+       lnet_ni_t *ni;
+       int i = 0;
+
+       list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
+               LASSERT(i < ping_info->pi_nnis);
+
+               ns = &ping_info->pi_ni[i];
+
+               ns->ns_nid = ni->ni_nid;
+
+               lnet_ni_lock(ni);
+               ns->ns_status = (ni->ni_status) ?
+                                ni->ni_status->ns_status : LNET_NI_STATUS_UP;
+               ni->ni_status = ns;
+               lnet_ni_unlock(ni);
+
+               i++;
+       }
+}
+
+static void
+lnet_ping_target_update(lnet_ping_info_t *pinfo, lnet_handle_md_t md_handle)
+{
+       lnet_ping_info_t *old_pinfo = NULL;
+       lnet_handle_md_t old_md;
+
+       /* switch the NIs to point to the new ping info created */
+       lnet_net_lock(LNET_LOCK_EX);
+
+       if (!the_lnet.ln_routing)
+               pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
+       lnet_ping_info_install_locked(pinfo);
+
+       if (the_lnet.ln_ping_info) {
+               old_pinfo = the_lnet.ln_ping_info;
+               old_md = the_lnet.ln_ping_target_md;
+       }
+       the_lnet.ln_ping_target_md = md_handle;
+       the_lnet.ln_ping_info = pinfo;
+
+       lnet_net_unlock(LNET_LOCK_EX);
+
+       if (old_pinfo) {
+               /* unlink the old ping info */
+               lnet_ping_md_unlink(old_pinfo, &old_md);
+               lnet_ping_info_free(old_pinfo);
+       }
+}
+
+static void
+lnet_ping_target_fini(void)
+{
+       int rc;
+
+       lnet_ping_md_unlink(the_lnet.ln_ping_info,
+                           &the_lnet.ln_ping_target_md);
+
+       rc = LNetEQFree(the_lnet.ln_ping_target_eq);
+       LASSERT(!rc);
+
+       lnet_ping_info_destroy();
+}
+
 static int
 lnet_ni_tq_credits(lnet_ni_t *ni)
 {
@@ -837,12 +1065,74 @@ lnet_ni_tq_credits(lnet_ni_t *ni)
 }
 
 static void
-lnet_shutdown_lndnis(void)
+lnet_clear_zombies_nis_locked(void)
 {
        int i;
        int islo;
        lnet_ni_t *ni;
 
+       /*
+        * 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 (!list_empty(&the_lnet.ln_nis_zombie)) {
+               int *ref;
+               int j;
+
+               ni = list_entry(the_lnet.ln_nis_zombie.next,
+                               lnet_ni_t, ni_list);
+               list_del_init(&ni->ni_list);
+               cfs_percpt_for_each(ref, j, ni->ni_refs) {
+                       if (!*ref)
+                               continue;
+                       /* still busy, add it back to zombie list */
+                       list_add(&ni->ni_list, &the_lnet.ln_nis_zombie);
+                       break;
+               }
+
+               if (!list_empty(&ni->ni_list)) {
+                       lnet_net_unlock(LNET_LOCK_EX);
+                       ++i;
+                       if ((i & (-i)) == i) {
+                               CDEBUG(D_WARNING, "Waiting for zombie LNI %s\n",
+                                      libcfs_nid2str(ni->ni_nid));
+                       }
+                       set_current_state(TASK_UNINTERRUPTIBLE);
+                       schedule_timeout(cfs_time_seconds(1));
+                       lnet_net_lock(LNET_LOCK_EX);
+                       continue;
+               }
+
+               ni->ni_lnd->lnd_refcount--;
+               lnet_net_unlock(LNET_LOCK_EX);
+
+               islo = ni->ni_lnd->lnd_type == LOLND;
+
+               LASSERT(!in_interrupt());
+               ni->ni_lnd->lnd_shutdown(ni);
+
+               /*
+                * can't deref lnd anymore now; it might have unregistered
+                * itself...
+                */
+               if (!islo)
+                       CDEBUG(D_LNI, "Removed LNI %s\n",
+                              libcfs_nid2str(ni->ni_nid));
+
+               lnet_ni_free(ni);
+               i = 2;
+
+               lnet_net_lock(LNET_LOCK_EX);
+       }
+}
+
+static void
+lnet_shutdown_lndnis(void)
+{
+       lnet_ni_t *ni;
+       int i;
+
        /* NB called holding the global mutex */
 
        /* All quiet on the API front */
@@ -895,92 +1185,90 @@ lnet_shutdown_lndnis(void)
        lnet_peer_tables_cleanup(NULL);
 
        lnet_net_lock(LNET_LOCK_EX);
-       /*
-        * 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 (!list_empty(&the_lnet.ln_nis_zombie)) {
-               int *ref;
-               int j;
 
-               ni = list_entry(the_lnet.ln_nis_zombie.next,
-                               lnet_ni_t, ni_list);
-               list_del_init(&ni->ni_list);
-               cfs_percpt_for_each(ref, j, ni->ni_refs) {
-                       if (!*ref)
-                               continue;
-                       /* still busy, add it back to zombie list */
-                       list_add(&ni->ni_list, &the_lnet.ln_nis_zombie);
-                       break;
-               }
+       lnet_clear_zombies_nis_locked();
+       the_lnet.ln_shutdown = 0;
+       lnet_net_unlock(LNET_LOCK_EX);
+}
 
-               if (!list_empty(&ni->ni_list)) {
-                       lnet_net_unlock(LNET_LOCK_EX);
-                       ++i;
-                       if ((i & (-i)) == i) {
-                               CDEBUG(D_WARNING, "Waiting for zombie LNI %s\n",
-                                      libcfs_nid2str(ni->ni_nid));
-                       }
-                       set_current_state(TASK_UNINTERRUPTIBLE);
-                       schedule_timeout(cfs_time_seconds(1));
-                       lnet_net_lock(LNET_LOCK_EX);
-                       continue;
-               }
+int
+lnet_shutdown_lndni(__u32 net)
+{
+       lnet_ping_info_t *pinfo;
+       lnet_handle_md_t md_handle;
+       lnet_ni_t *found_ni = NULL;
+       int ni_count;
+       int rc;
 
-               ni->ni_lnd->lnd_refcount--;
-               lnet_net_unlock(LNET_LOCK_EX);
+       if (LNET_NETTYP(net) == LOLND)
+               return -EINVAL;
 
-               islo = ni->ni_lnd->lnd_type == LOLND;
+       ni_count = lnet_get_ni_count();
 
-               LASSERT(!in_interrupt());
-               ni->ni_lnd->lnd_shutdown(ni);
+       /* create and link a new ping info, before removing the old one */
+       rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count - 1, false);
+       if (rc)
+               return rc;
 
-               /*
-                * can't deref lnd anymore now; it might have unregistered
-                * itself...
-                */
-               if (!islo)
-                       CDEBUG(D_LNI, "Removed LNI %s\n",
-                              libcfs_nid2str(ni->ni_nid));
+       /* proceed with shutting down the NI */
+       lnet_net_lock(LNET_LOCK_EX);
 
-               lnet_ni_free(ni);
-               i = 2;
+       found_ni = lnet_net2ni_locked(net, 0);
+       if (!found_ni) {
+               lnet_net_unlock(LNET_LOCK_EX);
+               lnet_ping_md_unlink(pinfo, &md_handle);
+               lnet_ping_info_free(pinfo);
+               return -EINVAL;
+       }
 
-               lnet_net_lock(LNET_LOCK_EX);
+       /*
+        * decrement the reference counter on found_ni which was
+        * incremented when we called lnet_net2ni_locked()
+        */
+       lnet_ni_decref_locked(found_ni, 0);
+       /* Move ni to zombie list so nobody can find it anymore */
+       list_move(&found_ni->ni_list, &the_lnet.ln_nis_zombie);
+
+       /* Drop the lock reference for the ln_nis ref. */
+       lnet_ni_decref_locked(found_ni, 0);
+
+       if (!list_empty(&found_ni->ni_cptlist)) {
+               list_del_init(&found_ni->ni_cptlist);
+               lnet_ni_decref_locked(found_ni, 0);
        }
 
-       the_lnet.ln_shutdown = 0;
        lnet_net_unlock(LNET_LOCK_EX);
+
+       /* Do peer table cleanup for this ni */
+       lnet_peer_tables_cleanup(found_ni);
+
+       lnet_net_lock(LNET_LOCK_EX);
+       lnet_clear_zombies_nis_locked();
+       lnet_net_unlock(LNET_LOCK_EX);
+
+       lnet_ping_target_update(pinfo, md_handle);
+
+       return 0;
 }
 
 static int
-lnet_startup_lndnis(void)
+lnet_startup_lndnis(struct list_head *nilist, __s32 peer_timeout,
+                   __s32 peer_cr, __s32 peer_buf_cr, __s32 credits,
+                   int *ni_count)
 {
        lnd_t *lnd;
        struct lnet_ni *ni;
        struct lnet_tx_queue *tq;
-       struct list_head nilist;
        int i;
        int rc = 0;
        __u32 lnd_type;
-       int nicount = 0;
-       char *nets = lnet_get_networks();
-
-       INIT_LIST_HEAD(&nilist);
-
-       if (!nets)
-               goto failed;
-
-       rc = lnet_parse_networks(&nilist, nets);
-       if (rc)
-               goto failed;
 
-       while (!list_empty(&nilist)) {
-               ni = list_entry(nilist.next, lnet_ni_t, ni_list);
+       while (!list_empty(nilist)) {
+               ni = list_entry(nilist->next, lnet_ni_t, ni_list);
                lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
 
-               LASSERT(libcfs_isknown_lnd(lnd_type));
+               if (!libcfs_isknown_lnd(lnd_type))
+                       goto failed;
 
                if (lnd_type == CIBLND    ||
                    lnd_type == OPENIBLND ||
@@ -991,6 +1279,24 @@ lnet_startup_lndnis(void)
                        goto failed;
                }
 
+               /* Make sure this new NI is unique. */
+               lnet_net_lock(LNET_LOCK_EX);
+               if (!lnet_net_unique(LNET_NIDNET(ni->ni_nid),
+                                    &the_lnet.ln_nis)) {
+                       if (lnd_type == LOLND) {
+                               lnet_net_unlock(LNET_LOCK_EX);
+                               list_del(&ni->ni_list);
+                               lnet_ni_free(ni);
+                               continue;
+                       }
+
+                       CERROR("Net %s is not unique\n",
+                              libcfs_net2str(LNET_NIDNET(ni->ni_nid)));
+                       lnet_net_unlock(LNET_LOCK_EX);
+                       goto failed;
+               }
+               lnet_net_unlock(LNET_LOCK_EX);
+
                mutex_lock(&the_lnet.ln_lnd_mutex);
                lnd = lnet_find_lnd_by_type(lnd_type);
 
@@ -1029,6 +1335,25 @@ lnet_startup_lndnis(void)
                        goto failed;
                }
 
+               /*
+                * If given some LND tunable parameters, parse those now to
+                * override the values in the NI structure.
+                */
+               if (peer_buf_cr >= 0)
+                       ni->ni_peerrtrcredits = peer_buf_cr;
+               if (peer_timeout >= 0)
+                       ni->ni_peertimeout = peer_timeout;
+               /*
+                * TODO
+                * Note: For now, don't allow the user to change
+                * peertxcredits as this number is used in the
+                * IB LND to control queue depth.
+                * if (peer_cr != -1)
+                *      ni->ni_peertxcredits = peer_cr;
+                */
+               if (credits >= 0)
+                       ni->ni_maxtxcredits = credits;
+
                LASSERT(ni->ni_peertimeout <= 0 || lnd->lnd_query);
 
                list_del(&ni->ni_list);
@@ -1045,6 +1370,14 @@ lnet_startup_lndnis(void)
 
                lnet_net_unlock(LNET_LOCK_EX);
 
+               /* increment the ni_count here to account for the LOLND as
+                * well.  If we increment past this point then the number
+                * of count will be missing the LOLND, and then ping and
+                * will not report the LOLND
+                */
+               if (ni_count)
+                       (*ni_count)++;
+
                if (lnd->lnd_type == LOLND) {
                        lnet_ni_addref(ni);
                        LASSERT(!the_lnet.ln_loni);
@@ -1070,29 +1403,16 @@ lnet_startup_lndnis(void)
                       libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits,
                       lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
                       ni->ni_peerrtrcredits, ni->ni_peertimeout);
-
-               nicount++;
-       }
-
-       if (the_lnet.ln_eq_waitni && 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));
-               goto failed;
        }
 
        return 0;
-
- failed:
-       lnet_shutdown_lndnis();
-
-       while (!list_empty(&nilist)) {
-               ni = list_entry(nilist.next, lnet_ni_t, ni_list);
+failed:
+       while (!list_empty(nilist)) {
+               ni = list_entry(nilist->next, lnet_ni_t, ni_list);
                list_del(&ni->ni_list);
                lnet_ni_free(ni);
        }
-
-       return -ENETDOWN;
+       return -EINVAL;
 }
 
 /**
@@ -1208,6 +1528,15 @@ LNetNIInit(lnet_pid_t requested_pid)
 {
        int im_a_router = 0;
        int rc;
+       int ni_count = 0;
+       int lnd_type;
+       struct lnet_ni *ni;
+       lnet_ping_info_t *pinfo;
+       lnet_handle_md_t md_handle;
+       struct list_head net_head;
+       char *nets;
+
+       INIT_LIST_HEAD(&net_head);
 
        mutex_lock(&the_lnet.ln_api_mutex);
 
@@ -1216,23 +1545,31 @@ LNetNIInit(lnet_pid_t requested_pid)
 
        if (the_lnet.ln_refcount > 0) {
                rc = the_lnet.ln_refcount++;
-               goto out;
+               mutex_unlock(&the_lnet.ln_api_mutex);
+               return rc;
        }
 
-       if (requested_pid == LNET_PID_ANY) {
-               /* Don't instantiate LNET just for me */
-               rc = -ENETDOWN;
-               goto failed0;
-       }
+       nets = lnet_get_networks();
 
        rc = lnet_prepare(requested_pid);
        if (rc)
                goto failed0;
 
-       rc = lnet_startup_lndnis();
+       rc = lnet_parse_networks(&net_head, nets);
+       if (rc < 0)
+               goto failed1;
+
+       rc = lnet_startup_lndnis(&net_head, -1, -1, -1, -1, &ni_count);
        if (rc)
                goto failed1;
 
+       if (the_lnet.ln_eq_waitni && ni_count > 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));
+               goto failed2;
+       }
+
        rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
        if (rc)
                goto failed2;
@@ -1252,26 +1589,30 @@ LNetNIInit(lnet_pid_t requested_pid)
        the_lnet.ln_refcount = 1;
        /* Now I may use my own API functions... */
 
-       /*
-        * NB router checker needs the_lnet.ln_ping_info in
-        * lnet_router_checker -> lnet_update_ni_status_locked
-         */
-       rc = lnet_ping_target_init();
+       rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
        if (rc)
                goto failed3;
 
+       lnet_ping_target_update(pinfo, md_handle);
+
        rc = lnet_router_checker_start();
        if (rc)
                goto failed4;
 
        lnet_router_debugfs_init();
-       goto out;
+
+       mutex_unlock(&the_lnet.ln_api_mutex);
+
+       return 0;
 
  failed4:
-       lnet_ping_target_fini();
- failed3:
        the_lnet.ln_refcount = 0;
+       lnet_ping_md_unlink(pinfo, &md_handle);
+       lnet_ping_info_free(pinfo);
+ failed3:
        lnet_acceptor_stop();
+       rc = LNetEQFree(the_lnet.ln_ping_target_eq);
+       LASSERT(!rc);
  failed2:
        lnet_destroy_routes();
        lnet_shutdown_lndnis();
@@ -1279,8 +1620,12 @@ LNetNIInit(lnet_pid_t requested_pid)
        lnet_unprepare();
  failed0:
        LASSERT(rc < 0);
- out:
        mutex_unlock(&the_lnet.ln_api_mutex);
+       while (!list_empty(&net_head)) {
+               ni = list_entry(net_head.next, struct lnet_ni, ni_list);
+               list_del_init(&ni->ni_list);
+               lnet_ni_free(ni);
+       }
        return rc;
 }
 EXPORT_SYMBOL(LNetNIInit);
@@ -1325,6 +1670,71 @@ LNetNIFini(void)
 }
 EXPORT_SYMBOL(LNetNIFini);
 
+int
+lnet_dyn_add_ni(lnet_pid_t requested_pid, char *nets,
+               __s32 peer_timeout, __s32 peer_cr, __s32 peer_buf_cr,
+               __s32 credits)
+{
+       lnet_ping_info_t *pinfo;
+       lnet_handle_md_t md_handle;
+       struct lnet_ni *ni;
+       struct list_head net_head;
+       int rc;
+
+       INIT_LIST_HEAD(&net_head);
+
+       /* Create a ni structure for the network string */
+       rc = lnet_parse_networks(&net_head, nets);
+       if (rc < 0)
+               return rc;
+
+       mutex_lock(&the_lnet.ln_api_mutex);
+
+       if (rc > 1) {
+               rc = -EINVAL; /* only add one interface per call */
+               goto failed0;
+       }
+
+       rc = lnet_ping_info_setup(&pinfo, &md_handle, 1 + lnet_get_ni_count(),
+                                 false);
+       if (rc)
+               goto failed0;
+
+       rc = lnet_startup_lndnis(&net_head, peer_timeout, peer_cr,
+                                peer_buf_cr, credits, NULL);
+       if (rc)
+               goto failed1;
+
+       lnet_ping_target_update(pinfo, md_handle);
+       mutex_unlock(&the_lnet.ln_api_mutex);
+
+       return 0;
+
+failed1:
+       lnet_ping_md_unlink(pinfo, &md_handle);
+       lnet_ping_info_free(pinfo);
+failed0:
+       mutex_unlock(&the_lnet.ln_api_mutex);
+       while (!list_empty(&net_head)) {
+               ni = list_entry(net_head.next, struct lnet_ni, ni_list);
+               list_del_init(&ni->ni_list);
+               lnet_ni_free(ni);
+       }
+       return rc;
+}
+
+int
+lnet_dyn_del_ni(__u32 net)
+{
+       int rc;
+
+       mutex_lock(&the_lnet.ln_api_mutex);
+       rc = lnet_shutdown_lndni(net);
+       mutex_unlock(&the_lnet.ln_api_mutex);
+
+       return rc;
+}
+
 /**
  * LNet ioctl handler.
  *
@@ -1339,7 +1749,6 @@ LNetCtl(unsigned int cmd, void *arg)
        unsigned long secs_passed;
 
        LASSERT(the_lnet.ln_init);
-       LASSERT(the_lnet.ln_refcount > 0);
 
        switch (cmd) {
        case IOC_LIBCFS_GET_NI:
@@ -1351,12 +1760,17 @@ LNetCtl(unsigned int cmd, void *arg)
                return lnet_fail_nid(data->ioc_nid, data->ioc_count);
 
        case IOC_LIBCFS_ADD_ROUTE:
+               mutex_lock(&the_lnet.ln_api_mutex);
                rc = lnet_add_route(data->ioc_net, data->ioc_count,
                                    data->ioc_nid, data->ioc_priority);
-               return (rc) ? rc : lnet_check_routes();
+               mutex_unlock(&the_lnet.ln_api_mutex);
+               return rc ? rc : lnet_check_routes();
 
        case IOC_LIBCFS_DEL_ROUTE:
-               return lnet_del_route(data->ioc_net, data->ioc_nid);
+               mutex_lock(&the_lnet.ln_api_mutex);
+               rc = lnet_del_route(data->ioc_net, data->ioc_nid);
+               mutex_unlock(&the_lnet.ln_api_mutex);
+               return rc;
 
        case IOC_LIBCFS_GET_ROUTE:
                return lnet_get_route(data->ioc_count,
@@ -1471,194 +1885,6 @@ LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
 }
 EXPORT_SYMBOL(LNetSnprintHandle);
 
-static int
-lnet_create_ping_info(void)
-{
-       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);
-               if (rc == -ENOENT)
-                       break;
-
-               LASSERT(!rc);
-       }
-
-       infosz = offsetof(lnet_ping_info_t, pi_ni[n]);
-       LIBCFS_ALLOC(pinfo, infosz);
-       if (!pinfo) {
-               CERROR("Can't allocate ping info[%d]\n", n);
-               return -ENOMEM;
-       }
-
-       pinfo->pi_nnis    = n;
-       pinfo->pi_pid     = the_lnet.ln_pid;
-       pinfo->pi_magic   = LNET_PROTO_PING_MAGIC;
-       pinfo->pi_features = LNET_PING_FEAT_NI_STATUS;
-       if (!the_lnet.ln_routing)
-               pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
-
-       for (i = 0; i < n; i++) {
-               lnet_ni_status_t *ns = &pinfo->pi_ni[i];
-
-               rc = LNetGetId(i, &id);
-               LASSERT(!rc);
-
-               ns->ns_nid    = id.nid;
-               ns->ns_status = LNET_NI_STATUS_UP;
-
-               lnet_net_lock(0);
-
-               ni = lnet_nid2ni_locked(id.nid, 0);
-               LASSERT(ni);
-
-               lnet_ni_lock(ni);
-               LASSERT(!ni->ni_status);
-               ni->ni_status = ns;
-               lnet_ni_unlock(ni);
-
-               lnet_ni_decref_locked(ni, 0);
-               lnet_net_unlock(0);
-       }
-
-       the_lnet.ln_ping_info = pinfo;
-       return 0;
-}
-
-static void
-lnet_destroy_ping_info(void)
-{
-       struct lnet_ni *ni;
-
-       lnet_net_lock(0);
-
-       list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
-               lnet_ni_lock(ni);
-               ni->ni_status = NULL;
-               lnet_ni_unlock(ni);
-       }
-
-       lnet_net_unlock(0);
-
-       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;
-}
-
-int
-lnet_ping_target_init(void)
-{
-       lnet_md_t md = { NULL };
-       lnet_handle_me_t meh;
-       lnet_process_id_t id;
-       int rc;
-       int rc2;
-       int infosz;
-
-       rc = lnet_create_ping_info();
-       if (rc)
-               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);
-       if (rc) {
-               CERROR("Can't allocate ping EQ: %d\n", rc);
-               goto failed_0;
-       }
-
-       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) {
-               CERROR("Can't create ping ME: %d\n", rc);
-               goto failed_1;
-       }
-
-       /* 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) {
-               CERROR("Can't attach ping MD: %d\n", rc);
-               goto failed_2;
-       }
-
-       return 0;
-
- failed_2:
-       rc2 = LNetMEUnlink(meh);
-       LASSERT(!rc2);
- failed_1:
-       rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
-       LASSERT(!rc2);
- failed_0:
-       lnet_destroy_ping_info();
-       return rc;
-}
-
-void
-lnet_ping_target_fini(void)
-{
-       lnet_event_t event;
-       int rc;
-       int which;
-       int timeout_ms = 1000;
-       sigset_t blocked = cfs_block_allsigs();
-
-       LNetMDUnlink(the_lnet.ln_ping_target_md);
-       /* NB md could be busy; this just starts the unlink */
-
-       for (;;) {
-               rc = LNetEQPoll(&the_lnet.ln_ping_target_eq, 1,
-                               timeout_ms, &event, &which);
-
-               /* I expect overflow... */
-               LASSERT(rc >= 0 || rc == -EOVERFLOW);
-
-               if (!rc) {
-                       /* timed out: provide a diagnostic */
-                       CWARN("Still waiting for ping MD to unlink\n");
-                       timeout_ms *= 2;
-                       continue;
-               }
-
-               /* Got a valid event */
-               if (event.unlinked)
-                       break;
-       }
-
-       rc = LNetEQFree(the_lnet.ln_ping_target_eq);
-       LASSERT(!rc);
-       lnet_destroy_ping_info();
-       cfs_restore_sigs(blocked);
-}
-
 static int lnet_ping(lnet_process_id_t id, int timeout_ms,
                     lnet_process_id_t __user *ids, int n_ids)
 {
@@ -1670,7 +1896,7 @@ static int lnet_ping(lnet_process_id_t id, int timeout_ms,
        int unlinked = 0;
        int replied = 0;
        const int a_long_time = 60000; /* mS */
-       int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
+       int infosz;
        lnet_ping_info_t *info;
        lnet_process_id_t tmpid;
        int i;
@@ -1679,6 +1905,8 @@ static int lnet_ping(lnet_process_id_t id, int timeout_ms,
        int rc2;
        sigset_t blocked;
 
+       infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
+
        if (n_ids <= 0 ||
            id.nid == LNET_NID_ANY ||
            timeout_ms > 500000 ||            /* arbitrary limit! */
index bcc97b422be108137786619e72d93c16180c59da..1ef07cdd5fe0769bb007786e852b5930ce14ac06 100644 (file)
@@ -77,7 +77,7 @@ lnet_issep(char c)
        }
 }
 
-static int
+int
 lnet_net_unique(__u32 net, struct list_head *nilist)
 {
        struct list_head *tmp;