Whamcloud - gitweb
Revert "LU-12222 lnet: Check if we're sending to ourselves"
[fs/lustre-release.git] / lnet / lnet / lib-move.c
index 75a0650..615d755 100644 (file)
@@ -194,8 +194,7 @@ lnet_fail_nid(lnet_nid_t nid, unsigned int threshold)
                if (tp->tp_threshold == 0 ||    /* needs culling anyway */
                    nid == LNET_NID_ANY ||      /* removing all entries */
                    tp->tp_nid == nid) {        /* matched this one */
-                       list_del(&tp->tp_list);
-                       list_add(&tp->tp_list, &cull);
+                       list_move(&tp->tp_list, &cull);
                }
        }
 
@@ -233,8 +232,7 @@ fail_peer (lnet_nid_t nid, int outgoing)
                                /* only cull zombies on outgoing tests,
                                 * since we may be at interrupt priority on
                                 * incoming messages. */
-                               list_del(&tp->tp_list);
-                               list_add(&tp->tp_list, &cull);
+                               list_move(&tp->tp_list, &cull);
                        }
                        continue;
                }
@@ -248,8 +246,7 @@ fail_peer (lnet_nid_t nid, int outgoing)
                                if (outgoing &&
                                    tp->tp_threshold == 0) {
                                        /* see above */
-                                       list_del(&tp->tp_list);
-                                       list_add(&tp->tp_list, &cull);
+                                       list_move(&tp->tp_list, &cull);
                                }
                        }
                        break;
@@ -1318,7 +1315,7 @@ routing_off:
 }
 
 static int
-lnet_compare_peers(struct lnet_peer_ni *p1, struct lnet_peer_ni *p2)
+lnet_compare_gw_lpnis(struct lnet_peer_ni *p1, struct lnet_peer_ni *p2)
 {
        if (p1->lpni_txqnob < p2->lpni_txqnob)
                return 1;
@@ -1435,8 +1432,8 @@ lnet_select_peer_ni(struct lnet_ni *best_ni, lnet_nid_t dst_nid,
  * Prerequisite: the best_ni should already be set in the sd
  */
 static inline struct lnet_peer_ni *
-lnet_find_best_lpni_on_net(struct lnet_send_data *sd, struct lnet_peer *peer,
-                          __u32 net_id)
+lnet_find_best_lpni_on_net(struct lnet_ni *lni, lnet_nid_t dst_nid,
+                          struct lnet_peer *peer, __u32 net_id)
 {
        struct lnet_peer_net *peer_net;
 
@@ -1453,111 +1450,101 @@ lnet_find_best_lpni_on_net(struct lnet_send_data *sd, struct lnet_peer *peer,
                return NULL;
        }
 
-       return lnet_select_peer_ni(sd->sd_best_ni, sd->sd_dst_nid,
-                                  peer, peer_net);
+       return lnet_select_peer_ni(lni, dst_nid, peer, peer_net);
 }
 
+/* Compare route priorities and hop counts */
 static int
-lnet_compare_routes(struct lnet_route *r1, struct lnet_route *r2,
-                   struct lnet_peer_ni **best_lpni)
+lnet_compare_routes(struct lnet_route *r1, struct lnet_route *r2)
 {
        int r1_hops = (r1->lr_hops == LNET_UNDEFINED_HOPS) ? 1 : r1->lr_hops;
        int r2_hops = (r2->lr_hops == LNET_UNDEFINED_HOPS) ? 1 : r2->lr_hops;
-       struct lnet_peer *lp1 = r1->lr_gateway;
-       struct lnet_peer *lp2 = r2->lr_gateway;
-       struct lnet_peer_ni *lpni1;
-       struct lnet_peer_ni *lpni2;
-       struct lnet_send_data sd;
-       int rc;
-
-       sd.sd_best_ni = NULL;
-       sd.sd_dst_nid = LNET_NID_ANY;
-       lpni1 = lnet_find_best_lpni_on_net(&sd, lp1, r1->lr_lnet);
-       lpni2 = lnet_find_best_lpni_on_net(&sd, lp2, r2->lr_lnet);
-       LASSERT(lpni1 && lpni2);
 
-       if (r1->lr_priority < r2->lr_priority) {
-               *best_lpni = lpni1;
+       if (r1->lr_priority < r2->lr_priority)
                return 1;
-       }
 
-       if (r1->lr_priority > r2->lr_priority) {
-               *best_lpni = lpni2;
+       if (r1->lr_priority > r2->lr_priority)
                return -1;
-       }
 
-       if (r1_hops < r2_hops) {
-               *best_lpni = lpni1;
+       if (r1_hops < r2_hops)
                return 1;
-       }
 
-       if (r1_hops > r2_hops) {
-               *best_lpni = lpni2;
+       if (r1_hops > r2_hops)
                return -1;
-       }
-
-       rc = lnet_compare_peers(lpni1, lpni2);
-       if (rc == 1) {
-               *best_lpni = lpni1;
-               return rc;
-       } else if (rc == -1) {
-               *best_lpni = lpni2;
-               return rc;
-       }
 
-       if (r1->lr_seq - r2->lr_seq <= 0) {
-               *best_lpni = lpni1;
-               return 1;
-       }
-
-       *best_lpni = lpni2;
-       return -1;
+       return 0;
 }
 
 static struct lnet_route *
-lnet_find_route_locked(struct lnet_net *net, __u32 remote_net,
-                      lnet_nid_t rtr_nid, struct lnet_route **prev_route,
+lnet_find_route_locked(struct lnet_remotenet *rnet, __u32 src_net,
+                      struct lnet_route **prev_route,
                       struct lnet_peer_ni **gwni)
 {
-       struct lnet_peer_ni *best_gw_ni = NULL;
+       struct lnet_peer_ni *lpni, *best_gw_ni = NULL;
        struct lnet_route *best_route;
        struct lnet_route *last_route;
-       struct lnet_remotenet *rnet;
-       struct lnet_peer *lp_best;
        struct lnet_route *route;
-       struct lnet_peer *lp;
        int rc;
+       __u32 restrict_net;
+       __u32 any_net = LNET_NIDNET(LNET_NID_ANY);
 
-       /* If @rtr_nid is not LNET_NID_ANY, return the gateway with
-        * rtr_nid nid, otherwise find the best gateway I can use */
-
-       rnet = lnet_find_rnet_locked(remote_net);
-       if (rnet == NULL)
-               return NULL;
-
-       lp_best = NULL;
        best_route = last_route = NULL;
        list_for_each_entry(route, &rnet->lrn_routes, lr_list) {
-               lp = route->lr_gateway;
-
                if (!lnet_is_route_alive(route))
                        continue;
 
-               if (lp_best == NULL) {
-                       best_route = last_route = route;
-                       lp_best = lp;
+               /* If the src_net is specified then we need to find an lpni
+                * on that network
+                */
+               restrict_net = src_net == any_net ? route->lr_lnet : src_net;
+               if (!best_route) {
+                       lpni = lnet_find_best_lpni_on_net(NULL, LNET_NID_ANY,
+                                                         route->lr_gateway,
+                                                         restrict_net);
+                       if (lpni) {
+                               best_route = last_route = route;
+                               best_gw_ni = lpni;
+                       } else
+                               CERROR("Gateway %s does not have a peer NI on net %s\n",
+                                      libcfs_nid2str(route->lr_gateway->lp_primary_nid),
+                                      libcfs_net2str(restrict_net));
+
+                       continue;
                }
 
                /* no protection on below fields, but it's harmless */
                if (last_route->lr_seq - route->lr_seq < 0)
                        last_route = route;
 
-               rc = lnet_compare_routes(route, best_route, &best_gw_ni);
-               if (rc < 0)
+               rc = lnet_compare_routes(route, best_route);
+               if (rc == -1)
+                       continue;
+
+               lpni = lnet_find_best_lpni_on_net(NULL, LNET_NID_ANY,
+                                                 route->lr_gateway,
+                                                 restrict_net);
+               if (!lpni) {
+                       CERROR("Gateway %s does not have a peer NI on net %s\n",
+                              libcfs_nid2str(route->lr_gateway->lp_primary_nid),
+                              libcfs_net2str(restrict_net));
                        continue;
+               }
 
-               best_route = route;
-               lp_best = lp;
+               if (rc == 1) {
+                       best_route = route;
+                       best_gw_ni = lpni;
+                       continue;
+               }
+
+               rc = lnet_compare_gw_lpnis(lpni, best_gw_ni);
+               if (rc == -1)
+                       continue;
+
+               if (rc == 1 || route->lr_seq <= best_route->lr_seq) {
+                       best_route = route;
+                       best_gw_ni = lpni;
+                       continue;
+               }
        }
 
        *prev_route = last_route;
@@ -1847,13 +1834,14 @@ lnet_handle_send(struct lnet_send_data *sd)
        rc = lnet_post_send_locked(msg, 0);
 
        if (!rc)
-               CDEBUG(D_NET, "TRACE: %s(%s:%s) -> %s(%s:%s) : %s try# %d\n",
+               CDEBUG(D_NET, "TRACE: %s(%s:%s) -> %s(%s:%s) %s : %s try# %d\n",
                       libcfs_nid2str(msg->msg_hdr.src_nid),
                       libcfs_nid2str(msg->msg_txni->ni_nid),
                       libcfs_nid2str(sd->sd_src_nid),
                       libcfs_nid2str(msg->msg_hdr.dest_nid),
                       libcfs_nid2str(sd->sd_dst_nid),
                       libcfs_nid2str(msg->msg_txpeer->lpni_nid),
+                      libcfs_nid2str(sd->sd_rtr_nid),
                       lnet_msgtyp2str(msg->msg_type), msg->msg_retry_count);
 
        return rc;
@@ -2028,69 +2016,97 @@ lnet_handle_find_routed_path(struct lnet_send_data *sd,
                             struct lnet_peer **gw_peer)
 {
        int rc;
+       __u32 local_lnet;
        struct lnet_peer *gw;
        struct lnet_peer *lp;
        struct lnet_peer_net *lpn;
        struct lnet_peer_net *best_lpn = NULL;
-       struct lnet_remotenet *rnet;
-       struct lnet_route *best_route;
-       struct lnet_route *last_route;
+       struct lnet_remotenet *rnet, *best_rnet = NULL;
+       struct lnet_route *best_route = NULL;
+       struct lnet_route *last_route = NULL;
        struct lnet_peer_ni *lpni = NULL;
        struct lnet_peer_ni *gwni = NULL;
        lnet_nid_t src_nid = sd->sd_src_nid;
 
-       /* we've already looked up the initial lpni using dst_nid */
-       lpni = sd->sd_best_lpni;
-       /* the peer tree must be in existence */
-       LASSERT(lpni && lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer);
-       lp = lpni->lpni_peer_net->lpn_peer;
+       /* If a router nid was specified then we are replying to a GET or
+        * sending an ACK. In this case we use the gateway associated with the
+        * specified router nid.
+        */
+       if (sd->sd_rtr_nid != LNET_NID_ANY) {
+               gwni = lnet_find_peer_ni_locked(sd->sd_rtr_nid);
+               if (!gwni) {
+                       CERROR("No peer NI for gateway %s\n",
+                              libcfs_nid2str(sd->sd_rtr_nid));
+                       return -EHOSTUNREACH;
+               }
+               gw = gwni->lpni_peer_net->lpn_peer;
+               lnet_peer_ni_decref_locked(gwni);
+               local_lnet = LNET_NIDNET(sd->sd_rtr_nid);
+       } else {
+               /* we've already looked up the initial lpni using dst_nid */
+               lpni = sd->sd_best_lpni;
+               /* the peer tree must be in existence */
+               LASSERT(lpni && lpni->lpni_peer_net &&
+                       lpni->lpni_peer_net->lpn_peer);
+               lp = lpni->lpni_peer_net->lpn_peer;
+
+               list_for_each_entry(lpn, &lp->lp_peer_nets, lpn_peer_nets) {
+                       /* is this remote network reachable?  */
+                       rnet = lnet_find_rnet_locked(lpn->lpn_net_id);
+                       if (!rnet)
+                               continue;
 
-       list_for_each_entry(lpn, &lp->lp_peer_nets, lpn_peer_nets) {
-               /* is this remote network reachable?  */
-               rnet = lnet_find_rnet_locked(lpn->lpn_net_id);
-               if (!rnet)
-                       continue;
+                       if (!best_lpn) {
+                               best_lpn = lpn;
+                               best_rnet = rnet;
+                       }
+
+                       if (best_lpn->lpn_seq <= lpn->lpn_seq)
+                               continue;
 
-               if (!best_lpn)
                        best_lpn = lpn;
+                       best_rnet = rnet;
+               }
 
-               if (best_lpn->lpn_seq <= lpn->lpn_seq)
-                       continue;
+               if (!best_lpn) {
+                       CERROR("peer %s has no available nets\n",
+                              libcfs_nid2str(sd->sd_dst_nid));
+                       return -EHOSTUNREACH;
+               }
 
-               best_lpn = lpn;
-       }
+               sd->sd_best_lpni = lnet_find_best_lpni_on_net(sd->sd_best_ni,
+                                                             sd->sd_dst_nid,
+                                                             lp,
+                                                             best_lpn->lpn_net_id);
+               if (!sd->sd_best_lpni) {
+                       CERROR("peer %s down\n",
+                              libcfs_nid2str(sd->sd_dst_nid));
+                       return -EHOSTUNREACH;
+               }
 
-       if (!best_lpn) {
-               CERROR("peer %s has no available nets \n",
-                      libcfs_nid2str(sd->sd_dst_nid));
-               return -EHOSTUNREACH;
-       }
+               best_route = lnet_find_route_locked(best_rnet,
+                                                   LNET_NIDNET(src_nid),
+                                                   &last_route, &gwni);
+               if (!best_route) {
+                       CERROR("no route to %s from %s\n",
+                              libcfs_nid2str(dst_nid),
+                              libcfs_nid2str(src_nid));
+                       return -EHOSTUNREACH;
+               }
 
-       sd->sd_best_lpni = lnet_find_best_lpni_on_net(sd, lp, best_lpn->lpn_net_id);
-       if (!sd->sd_best_lpni) {
-               CERROR("peer %s down\n", libcfs_nid2str(sd->sd_dst_nid));
-               return -EHOSTUNREACH;
-       }
+               if (!gwni) {
+                       CERROR("Internal Error. Route expected to %s from %s\n",
+                              libcfs_nid2str(dst_nid),
+                              libcfs_nid2str(src_nid));
+                       return -EFAULT;
+               }
 
-       best_route = lnet_find_route_locked(NULL, best_lpn->lpn_net_id,
-                                           sd->sd_rtr_nid, &last_route,
-                                           &gwni);
-       if (!best_route) {
-               CERROR("no route to %s from %s\n",
-                      libcfs_nid2str(dst_nid), libcfs_nid2str(src_nid));
-               return -EHOSTUNREACH;
-       }
+               gw = best_route->lr_gateway;
+               LASSERT(gw == gwni->lpni_peer_net->lpn_peer);
+               local_lnet = best_route->lr_lnet;
 
-       if (!gwni) {
-               CERROR("Internal Error. Route expected to %s from %s\n",
-                       libcfs_nid2str(dst_nid),
-                       libcfs_nid2str(src_nid));
-               return -EFAULT;
        }
 
-       gw = best_route->lr_gateway;
-       LASSERT(gw == gwni->lpni_peer_net->lpn_peer);
-
        /*
         * Discover this gateway if it hasn't already been discovered.
         * This means we might delay the message until discovery has
@@ -2105,14 +2121,13 @@ lnet_handle_find_routed_path(struct lnet_send_data *sd,
        if (!sd->sd_best_ni)
                sd->sd_best_ni = lnet_find_best_ni_on_spec_net(NULL, gw,
                                        lnet_peer_get_net_locked(gw,
-                                               best_route->lr_lnet),
+                                                                local_lnet),
                                        sd->sd_md_cpt,
                                        true);
 
        if (!sd->sd_best_ni) {
-               CERROR("Internal Error. Expected local ni on %s "
-                      "but non found :%s\n",
-                      libcfs_net2str(best_route->lr_lnet),
+               CERROR("Internal Error. Expected local ni on %s but non found :%s\n",
+                      libcfs_net2str(local_lnet),
                       libcfs_nid2str(sd->sd_src_nid));
                return -EFAULT;
        }
@@ -2124,9 +2139,11 @@ lnet_handle_find_routed_path(struct lnet_send_data *sd,
         * increment the sequence numbers since now we're sure we're
         * going to use this path
         */
-       LASSERT(best_route && last_route);
-       best_route->lr_seq = last_route->lr_seq + 1;
-       best_lpn->lpn_seq++;
+       if (sd->sd_rtr_nid == LNET_NID_ANY) {
+               LASSERT(best_route && last_route);
+               best_route->lr_seq = last_route->lr_seq + 1;
+               best_lpn->lpn_seq++;
+       }
 
        return 0;
 }
@@ -2390,7 +2407,8 @@ lnet_handle_any_mr_dsta(struct lnet_send_data *sd)
                                        lnet_msg_discovery(sd->sd_msg));
        if (sd->sd_best_ni) {
                sd->sd_best_lpni =
-                 lnet_find_best_lpni_on_net(sd, sd->sd_peer,
+                 lnet_find_best_lpni_on_net(sd->sd_best_ni, sd->sd_dst_nid,
+                                            sd->sd_peer,
                                             sd->sd_best_ni->ni_net->net_id);
 
                /*
@@ -4631,7 +4649,6 @@ lnet_attach_rsp_tracker(struct lnet_rsp_tracker *rspt, int cpt,
                        struct lnet_libmd *md, struct lnet_handle_md mdh)
 {
        s64 timeout_ns;
-       bool new_entry = true;
        struct lnet_rsp_tracker *local_rspt;
 
        /*
@@ -4650,8 +4667,7 @@ lnet_attach_rsp_tracker(struct lnet_rsp_tracker *rspt, int cpt,
                 * we already have an rspt attached to the md, so we'll
                 * update the deadline on that one.
                 */
-               LIBCFS_FREE(rspt, sizeof(*rspt));
-               new_entry = false;
+               lnet_rspt_free(rspt, cpt);
        } else {
                /* new md */
                rspt->rspt_mdh = mdh;
@@ -4667,9 +4683,7 @@ lnet_attach_rsp_tracker(struct lnet_rsp_tracker *rspt, int cpt,
         * list in order to expire all the older entries first.
         */
        lnet_net_lock(cpt);
-       if (!new_entry && !list_empty(&local_rspt->rspt_on_list))
-               list_del_init(&local_rspt->rspt_on_list);
-       list_add_tail(&local_rspt->rspt_on_list, the_lnet.ln_mt_rstq[cpt]);
+       list_move_tail(&local_rspt->rspt_on_list, the_lnet.ln_mt_rstq[cpt]);
        lnet_net_unlock(cpt);
        lnet_res_unlock(cpt);
 }
@@ -4771,7 +4785,7 @@ LNetPut(lnet_nid_t self, struct lnet_handle_md mdh, enum lnet_ack_req ack,
                               md->md_me->me_portal);
                lnet_res_unlock(cpt);
 
-               LIBCFS_FREE(rspt, sizeof(*rspt));
+               lnet_rspt_free(rspt, cpt);
                lnet_msg_free(msg);
                return -ENOENT;
        }
@@ -4998,7 +5012,7 @@ LNetGet(lnet_nid_t self, struct lnet_handle_md mdh,
                lnet_res_unlock(cpt);
 
                lnet_msg_free(msg);
-               LIBCFS_FREE(rspt, sizeof(*rspt));
+               lnet_rspt_free(rspt, cpt);
                return -ENOENT;
        }