lnet/lnet/peer.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lnet/lnet/peer.c
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_LNET
  40
  41 #include <lnet/lib-lnet.h>
  42
  43 int
  44 lnet_peer_tables_create(void)
  45 {
  46         struct lnet_peer_table  *ptable;
  47         cfs_list_t              *hash;
  48         int                     i;
  49         int                     j;
  50
  51         the_lnet.ln_peer_tables = cfs_percpt_alloc(lnet_cpt_table(),
  52                                                    sizeof(*ptable));
  53         if (the_lnet.ln_peer_tables == NULL) {
  54                 CERROR("Failed to allocate cpu-partition peer tables\n");
  55                 return -ENOMEM;
  56         }
  57
  58         cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
  59                 CFS_INIT_LIST_HEAD(&ptable->pt_deathrow);
  60
  61                 LIBCFS_CPT_ALLOC(hash, lnet_cpt_table(), i,
  62                                  LNET_PEER_HASH_SIZE * sizeof(*hash));
  63                 if (hash == NULL) {
  64                         CERROR("Failed to create peer hash table\n");
  65                         lnet_peer_tables_destroy();
  66                         return -ENOMEM;
  67                 }
  68
  69                 for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
  70                         CFS_INIT_LIST_HEAD(&hash[j]);
  71                 ptable->pt_hash = hash; /* sign of initialization */
  72         }
  73
  74         return 0;
  75 }
  76
  77 void
  78 lnet_peer_tables_destroy(void)
  79 {
  80         struct lnet_peer_table  *ptable;
  81         cfs_list_t              *hash;
  82         int                     i;
  83         int                     j;
  84
  85         if (the_lnet.ln_peer_tables == NULL)
  86                 return;
  87
  88         cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
  89                 hash = ptable->pt_hash;
  90                 if (hash == NULL) /* not intialized */
  91                         break;
  92
  93                 LASSERT(cfs_list_empty(&ptable->pt_deathrow));
  94
  95                 ptable->pt_hash = NULL;
  96                 for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
  97                         LASSERT(cfs_list_empty(&hash[j]));
  98
  99                 LIBCFS_FREE(hash, LNET_PEER_HASH_SIZE * sizeof(*hash));
 100         }
 101
 102         cfs_percpt_free(the_lnet.ln_peer_tables);
 103         the_lnet.ln_peer_tables = NULL;
 104 }
 105
 106 void
 107 lnet_peer_tables_cleanup(void)
 108 {
 109         struct lnet_peer_table  *ptable;
 110         int                     i;
 111         int                     j;
 112
 113         LASSERT(the_lnet.ln_shutdown);  /* i.e. no new peers */
 114
 115         cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
 116                 lnet_net_lock(i);
 117
 118                 for (j = 0; j < LNET_PEER_HASH_SIZE; j++) {
 119                         cfs_list_t *peers = &ptable->pt_hash[j];
 120
 121                         while (!cfs_list_empty(peers)) {
 122                                 lnet_peer_t *lp = cfs_list_entry(peers->next,
 123                                                                  lnet_peer_t,
 124                                                                  lp_hashlist);
 125                                 cfs_list_del_init(&lp->lp_hashlist);
 126                                 /* lose hash table's ref */
 127                                 lnet_peer_decref_locked(lp);
 128                         }
 129                 }
 130
 131                 lnet_net_unlock(i);
 132         }
 133
 134         cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
 135                 CFS_LIST_HEAD   (deathrow);
 136                 lnet_peer_t     *lp;
 137
 138                 lnet_net_lock(i);
 139
 140                 for (j = 3; ptable->pt_number != 0; j++) {
 141                         lnet_net_unlock(i);
 142
 143                         if ((j & (j - 1)) == 0) {
 144                                 CDEBUG(D_WARNING,
 145                                        "Waiting for %d peers on peer table\n",
 146                                        ptable->pt_number);
 147                         }
 148                         cfs_pause(cfs_time_seconds(1) / 2);
 149                         lnet_net_lock(i);
 150                 }
 151                 cfs_list_splice_init(&ptable->pt_deathrow, &deathrow);
 152
 153                 lnet_net_unlock(i);
 154
 155                 while (!cfs_list_empty(&deathrow)) {
 156                         lp = cfs_list_entry(deathrow.next,
 157                                             lnet_peer_t, lp_hashlist);
 158                         cfs_list_del(&lp->lp_hashlist);
 159                         LIBCFS_FREE(lp, sizeof(*lp));
 160                 }
 161         }
 162 }
 163
 164 void
 165 lnet_destroy_peer_locked(lnet_peer_t *lp)
 166 {
 167         struct lnet_peer_table *ptable;
 168
 169         LASSERT(lp->lp_refcount == 0);
 170         LASSERT(lp->lp_rtr_refcount == 0);
 171         LASSERT(cfs_list_empty(&lp->lp_txq));
 172         LASSERT(cfs_list_empty(&lp->lp_hashlist));
 173         LASSERT(lp->lp_txqnob == 0);
 174
 175         ptable = the_lnet.ln_peer_tables[lp->lp_cpt];
 176         LASSERT(ptable->pt_number > 0);
 177         ptable->pt_number--;
 178
 179         lnet_ni_decref_locked(lp->lp_ni, lp->lp_cpt);
 180         lp->lp_ni = NULL;
 181
 182         cfs_list_add(&lp->lp_hashlist, &ptable->pt_deathrow);
 183 }
 184
 185 lnet_peer_t *
 186 lnet_find_peer_locked(struct lnet_peer_table *ptable, lnet_nid_t nid)
 187 {
 188         cfs_list_t      *peers;
 189         lnet_peer_t     *lp;
 190
 191         LASSERT(!the_lnet.ln_shutdown);
 192
 193         peers = &ptable->pt_hash[lnet_nid2peerhash(nid)];
 194         cfs_list_for_each_entry(lp, peers, lp_hashlist) {
 195                 if (lp->lp_nid == nid) {
 196                         lnet_peer_addref_locked(lp);
 197                         return lp;
 198                 }
 199         }
 200
 201         return NULL;
 202 }
 203
 204 int
 205 lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid, int cpt)
 206 {
 207         struct lnet_peer_table  *ptable;
 208         lnet_peer_t             *lp = NULL;
 209         lnet_peer_t             *lp2;
 210         int                     cpt2;
 211         int                     rc = 0;
 212
 213         *lpp = NULL;
 214         if (the_lnet.ln_shutdown) /* it's shutting down */
 215                 return -ESHUTDOWN;
 216
 217         /* cpt can be LNET_LOCK_EX if it's called from router functions */
 218         cpt2 = cpt != LNET_LOCK_EX ? cpt : lnet_cpt_of_nid_locked(nid);
 219
 220         ptable = the_lnet.ln_peer_tables[cpt2];
 221         lp = lnet_find_peer_locked(ptable, nid);
 222         if (lp != NULL) {
 223                 *lpp = lp;
 224                 return 0;
 225         }
 226
 227         if (!cfs_list_empty(&ptable->pt_deathrow)) {
 228                 lp = cfs_list_entry(ptable->pt_deathrow.next,
 229                                     lnet_peer_t, lp_hashlist);
 230                 cfs_list_del(&lp->lp_hashlist);
 231         }
 232
 233         /*
 234          * take extra refcount in case another thread has shutdown LNet
 235          * and destroyed locks and peer-table before I finish the allocation
 236          */
 237         ptable->pt_number++;
 238         lnet_net_unlock(cpt);
 239
 240         if (lp != NULL)
 241                 memset(lp, 0, sizeof(*lp));
 242         else
 243                 LIBCFS_CPT_ALLOC(lp, lnet_cpt_table(), cpt2, sizeof(*lp));
 244
 245         if (lp == NULL) {
 246                 rc = -ENOMEM;
 247                 lnet_net_lock(cpt);
 248                 goto out;
 249         }
 250
 251         CFS_INIT_LIST_HEAD(&lp->lp_txq);
 252         CFS_INIT_LIST_HEAD(&lp->lp_rtrq);
 253         CFS_INIT_LIST_HEAD(&lp->lp_routes);
 254
 255         lp->lp_notify = 0;
 256         lp->lp_notifylnd = 0;
 257         lp->lp_notifying = 0;
 258         lp->lp_alive_count = 0;
 259         lp->lp_timestamp = 0;
 260         lp->lp_alive = !lnet_peers_start_down(); /* 1 bit!! */
 261         lp->lp_last_alive = cfs_time_current(); /* assumes alive */
 262         lp->lp_last_query = 0; /* haven't asked NI yet */
 263         lp->lp_ping_timestamp = 0;
 264         lp->lp_ping_feats = LNET_PING_FEAT_INVAL;
 265         lp->lp_nid = nid;
 266         lp->lp_cpt = cpt2;
 267         lp->lp_refcount = 2;    /* 1 for caller; 1 for hash */
 268         lp->lp_rtr_refcount = 0;
 269
 270         lnet_net_lock(cpt);
 271
 272         if (the_lnet.ln_shutdown) {
 273                 rc = -ESHUTDOWN;
 274                 goto out;
 275         }
 276
 277         lp2 = lnet_find_peer_locked(ptable, nid);
 278         if (lp2 != NULL) {
 279                 *lpp = lp2;
 280                 goto out;
 281         }
 282
 283         lp->lp_ni = lnet_net2ni_locked(LNET_NIDNET(nid), cpt2);
 284         if (lp->lp_ni == NULL) {
 285                 rc = -EHOSTUNREACH;
 286                 goto out;
 287         }
 288
 289         lp->lp_txcredits    =
 290         lp->lp_mintxcredits = lp->lp_ni->ni_peertxcredits;
 291         lp->lp_rtrcredits    =
 292         lp->lp_minrtrcredits = lnet_peer_buffer_credits(lp->lp_ni);
 293
 294         cfs_list_add_tail(&lp->lp_hashlist,
 295                           &ptable->pt_hash[lnet_nid2peerhash(nid)]);
 296         ptable->pt_version++;
 297         *lpp = lp;
 298
 299         return 0;
 300 out:
 301         if (lp != NULL)
 302                 cfs_list_add(&lp->lp_hashlist, &ptable->pt_deathrow);
 303         ptable->pt_number--;
 304         return rc;
 305 }
 306
 307 void
 308 lnet_debug_peer(lnet_nid_t nid)
 309 {
 310         char            *aliveness = "NA";
 311         lnet_peer_t     *lp;
 312         int             rc;
 313         int             cpt;
 314
 315         cpt = lnet_cpt_of_nid(nid);
 316         lnet_net_lock(cpt);
 317
 318         rc = lnet_nid2peer_locked(&lp, nid, cpt);
 319         if (rc != 0) {
 320                 lnet_net_unlock(cpt);
 321                 CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
 322                 return;
 323         }
 324
 325         if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
 326                 aliveness = lp->lp_alive ? "up" : "down";
 327
 328         CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
 329                libcfs_nid2str(lp->lp_nid), lp->lp_refcount,
 330                aliveness, lp->lp_ni->ni_peertxcredits,
 331                lp->lp_rtrcredits, lp->lp_minrtrcredits,
 332                lp->lp_txcredits, lp->lp_mintxcredits, lp->lp_txqnob);
 333
 334         lnet_peer_decref_locked(lp);
 335
 336         lnet_net_unlock(cpt);
 337 }