X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lnet%2Flnet%2Fpeer.c;h=363d27093ce703f8ffb642f33ce16ed75ca01b13;hb=2bcfce77a2712960fadac2f034c1df4826780608;hp=99230ee1143afe72918c3e7b3f7c6fb530487556;hpb=70e80ade90af09300396706b8910e196a7928520;p=fs%2Flustre-release.git diff --git a/lnet/lnet/peer.c b/lnet/lnet/peer.c index 99230ee..363d270 100644 --- a/lnet/lnet/peer.c +++ b/lnet/lnet/peer.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -16,8 +14,8 @@ * in the LICENSE file that accompanied this code). * * You should have received a copy of the GNU General Public License - * version 2 along with this program; If not, see [sun.com URL with a - * copy of GPLv2]. + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, * CA 95054 USA or visit www.sun.com if you need additional information or @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2012, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -39,226 +39,413 @@ #define DEBUG_SUBSYSTEM S_LNET #include +#include int -lnet_create_peer_table(void) +lnet_peer_tables_create(void) { - struct list_head *hash; - int i; - - LASSERT (the_lnet.ln_peer_hash == NULL); - LIBCFS_ALLOC(hash, LNET_PEER_HASHSIZE * sizeof(struct list_head)); - - if (hash == NULL) { - CERROR("Can't allocate peer hash table\n"); + struct lnet_peer_table *ptable; + struct list_head *hash; + int i; + int j; + + the_lnet.ln_peer_tables = cfs_percpt_alloc(lnet_cpt_table(), + sizeof(*ptable)); + if (the_lnet.ln_peer_tables == NULL) { + CERROR("Failed to allocate cpu-partition peer tables\n"); return -ENOMEM; } - for (i = 0; i < LNET_PEER_HASHSIZE; i++) - CFS_INIT_LIST_HEAD(&hash[i]); + cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) { + INIT_LIST_HEAD(&ptable->pt_deathrow); + + LIBCFS_CPT_ALLOC(hash, lnet_cpt_table(), i, + LNET_PEER_HASH_SIZE * sizeof(*hash)); + if (hash == NULL) { + CERROR("Failed to create peer hash table\n"); + lnet_peer_tables_destroy(); + return -ENOMEM; + } + + for (j = 0; j < LNET_PEER_HASH_SIZE; j++) + INIT_LIST_HEAD(&hash[j]); + ptable->pt_hash = hash; /* sign of initialization */ + } - the_lnet.ln_peer_hash = hash; return 0; } void -lnet_destroy_peer_table(void) +lnet_peer_tables_destroy(void) { - int i; + struct lnet_peer_table *ptable; + struct list_head *hash; + int i; + int j; - if (the_lnet.ln_peer_hash == NULL) - return; + if (the_lnet.ln_peer_tables == NULL) + return; + + cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) { + hash = ptable->pt_hash; + if (hash == NULL) /* not intialized */ + break; - for (i = 0; i < LNET_PEER_HASHSIZE; i++) - LASSERT (list_empty(&the_lnet.ln_peer_hash[i])); - - LIBCFS_FREE(the_lnet.ln_peer_hash, - LNET_PEER_HASHSIZE * sizeof (struct list_head)); - the_lnet.ln_peer_hash = NULL; + LASSERT(list_empty(&ptable->pt_deathrow)); + + ptable->pt_hash = NULL; + for (j = 0; j < LNET_PEER_HASH_SIZE; j++) + LASSERT(list_empty(&hash[j])); + + LIBCFS_FREE(hash, LNET_PEER_HASH_SIZE * sizeof(*hash)); + } + + cfs_percpt_free(the_lnet.ln_peer_tables); + the_lnet.ln_peer_tables = NULL; } -void -lnet_clear_peer_table(void) +static void +lnet_peer_table_cleanup_locked(lnet_ni_t *ni, struct lnet_peer_table *ptable) { - int i; - - LASSERT (the_lnet.ln_shutdown); /* i.e. no new peers */ - - for (i = 0; i < LNET_PEER_HASHSIZE; i++) { - struct list_head *peers = &the_lnet.ln_peer_hash[i]; - - LNET_LOCK(); - while (!list_empty(peers)) { - lnet_peer_t *lp = list_entry(peers->next, - lnet_peer_t, lp_hashlist); - - list_del(&lp->lp_hashlist); - lnet_peer_decref_locked(lp); /* lose hash table's ref */ + int i; + lnet_peer_t *lp; + lnet_peer_t *tmp; + + for (i = 0; i < LNET_PEER_HASH_SIZE; i++) { + list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i], + lp_hashlist) { + if (ni != NULL && ni != lp->lp_ni) + continue; + list_del_init(&lp->lp_hashlist); + /* Lose hash table's ref */ + ptable->pt_zombies++; + lnet_peer_decref_locked(lp); } - LNET_UNLOCK(); } +} - LNET_LOCK(); - for (i = 3; the_lnet.ln_npeers != 0;i++) { - LNET_UNLOCK(); +static void +lnet_peer_table_deathrow_wait_locked(struct lnet_peer_table *ptable, + int cpt_locked) +{ + int i; - if ((i & (i-1)) == 0) - CDEBUG(D_WARNING,"Waiting for %d peers\n", - the_lnet.ln_npeers); - cfs_pause(cfs_time_seconds(1)); + for (i = 3; ptable->pt_zombies != 0; i++) { + lnet_net_unlock(cpt_locked); - LNET_LOCK(); - } - LNET_UNLOCK(); + if (IS_PO2(i)) { + CDEBUG(D_WARNING, + "Waiting for %d zombies on peer table\n", + ptable->pt_zombies); + } + cfs_pause(cfs_time_seconds(1) >> 1); + lnet_net_lock(cpt_locked); + } +} + +static void +lnet_peer_table_del_rtrs_locked(lnet_ni_t *ni, struct lnet_peer_table *ptable, + int cpt_locked) +{ + lnet_peer_t *lp; + lnet_peer_t *tmp; + lnet_nid_t lp_nid; + int i; + + for (i = 0; i < LNET_PEER_HASH_SIZE; i++) { + list_for_each_entry_safe(lp, tmp, &ptable->pt_hash[i], + lp_hashlist) { + if (ni != lp->lp_ni) + continue; + + if (lp->lp_rtr_refcount == 0) + continue; + + lp_nid = lp->lp_nid; + + lnet_net_unlock(cpt_locked); + lnet_del_route(LNET_NIDNET(LNET_NID_ANY), lp_nid); + lnet_net_lock(cpt_locked); + } + } +} + +void +lnet_peer_tables_cleanup(lnet_ni_t *ni) +{ + int i; + struct lnet_peer_table *ptable; + lnet_peer_t *lp; + struct list_head deathrow; + + INIT_LIST_HEAD(&deathrow); + + LASSERT(the_lnet.ln_shutdown || ni != NULL); + /* If just deleting the peers for a NI, get rid of any routes these + * peers are gateways for. */ + cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) { + lnet_net_lock(i); + lnet_peer_table_del_rtrs_locked(ni, ptable, i); + lnet_net_unlock(i); + } + + /* Start the process of moving the applicable peers to + * deathrow. */ + cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) { + lnet_net_lock(i); + lnet_peer_table_cleanup_locked(ni, ptable); + lnet_net_unlock(i); + } + + /* Cleanup all entries on deathrow. */ + cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) { + lnet_net_lock(i); + lnet_peer_table_deathrow_wait_locked(ptable, i); + list_splice_init(&ptable->pt_deathrow, &deathrow); + lnet_net_unlock(i); + } + + while (!list_empty(&deathrow)) { + lp = list_entry(deathrow.next, lnet_peer_t, lp_hashlist); + list_del(&lp->lp_hashlist); + LIBCFS_FREE(lp, sizeof(*lp)); + } } void -lnet_destroy_peer_locked (lnet_peer_t *lp) +lnet_destroy_peer_locked(lnet_peer_t *lp) { - lnet_ni_decref_locked(lp->lp_ni); - LNET_UNLOCK(); + struct lnet_peer_table *ptable; - LASSERT (lp->lp_refcount == 0); - LASSERT (lp->lp_rtr_refcount == 0); - LASSERT (list_empty(&lp->lp_txq)); - LASSERT (lp->lp_txqnob == 0); + LASSERT(lp->lp_refcount == 0); + LASSERT(lp->lp_rtr_refcount == 0); + LASSERT(list_empty(&lp->lp_txq)); + LASSERT(list_empty(&lp->lp_hashlist)); + LASSERT(lp->lp_txqnob == 0); - LIBCFS_FREE(lp, sizeof(*lp)); + ptable = the_lnet.ln_peer_tables[lp->lp_cpt]; + LASSERT(ptable->pt_number > 0); + ptable->pt_number--; - LNET_LOCK(); + lnet_ni_decref_locked(lp->lp_ni, lp->lp_cpt); + lp->lp_ni = NULL; - LASSERT(the_lnet.ln_npeers > 0); - the_lnet.ln_npeers--; + list_add(&lp->lp_hashlist, &ptable->pt_deathrow); + LASSERT(ptable->pt_zombies > 0); + ptable->pt_zombies--; } lnet_peer_t * -lnet_find_peer_locked (lnet_nid_t nid) +lnet_find_peer_locked(struct lnet_peer_table *ptable, lnet_nid_t nid) { - unsigned int idx = LNET_NIDADDR(nid) % LNET_PEER_HASHSIZE; - struct list_head *peers = &the_lnet.ln_peer_hash[idx]; - struct list_head *tmp; - lnet_peer_t *lp; + struct list_head *peers; + lnet_peer_t *lp; - if (the_lnet.ln_shutdown) - return NULL; + LASSERT(!the_lnet.ln_shutdown); - list_for_each (tmp, peers) { - lp = list_entry(tmp, lnet_peer_t, lp_hashlist); - + peers = &ptable->pt_hash[lnet_nid2peerhash(nid)]; + list_for_each_entry(lp, peers, lp_hashlist) { if (lp->lp_nid == nid) { - lnet_peer_addref_locked(lp); + lnet_peer_addref_locked(lp); return lp; - } + } } - + return NULL; } int -lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid) +lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid, int cpt) { - lnet_peer_t *lp; - lnet_peer_t *lp2; + struct lnet_peer_table *ptable; + lnet_peer_t *lp = NULL; + lnet_peer_t *lp2; + int cpt2; + int rc = 0; + + *lpp = NULL; + if (the_lnet.ln_shutdown) /* it's shutting down */ + return -ESHUTDOWN; + + /* cpt can be LNET_LOCK_EX if it's called from router functions */ + cpt2 = cpt != LNET_LOCK_EX ? cpt : lnet_cpt_of_nid_locked(nid); + + ptable = the_lnet.ln_peer_tables[cpt2]; + lp = lnet_find_peer_locked(ptable, nid); + if (lp != NULL) { + *lpp = lp; + return 0; + } + + if (!list_empty(&ptable->pt_deathrow)) { + lp = list_entry(ptable->pt_deathrow.next, + lnet_peer_t, lp_hashlist); + list_del(&lp->lp_hashlist); + } + + /* + * take extra refcount in case another thread has shutdown LNet + * and destroyed locks and peer-table before I finish the allocation + */ + ptable->pt_number++; + lnet_net_unlock(cpt); + + if (lp != NULL) + memset(lp, 0, sizeof(*lp)); + else + LIBCFS_CPT_ALLOC(lp, lnet_cpt_table(), cpt2, sizeof(*lp)); - lp = lnet_find_peer_locked(nid); - if (lp != NULL) { - *lpp = lp; - return 0; - } - - LNET_UNLOCK(); - - LIBCFS_ALLOC(lp, sizeof(*lp)); if (lp == NULL) { - *lpp = NULL; - LNET_LOCK(); - return -ENOMEM; - } + rc = -ENOMEM; + lnet_net_lock(cpt); + goto out; + } + + INIT_LIST_HEAD(&lp->lp_txq); + INIT_LIST_HEAD(&lp->lp_rtrq); + INIT_LIST_HEAD(&lp->lp_routes); - memset(lp, 0, sizeof(*lp)); /* zero counters etc */ - - CFS_INIT_LIST_HEAD(&lp->lp_txq); - CFS_INIT_LIST_HEAD(&lp->lp_rtrq); - - lp->lp_alive = !lnet_peers_start_down(); /* 1 bit!! */ lp->lp_notify = 0; lp->lp_notifylnd = 0; lp->lp_notifying = 0; lp->lp_alive_count = 0; - lp->lp_timestamp = 0; + lp->lp_timestamp = 0; + lp->lp_alive = !lnet_peers_start_down(); /* 1 bit!! */ + lp->lp_last_alive = cfs_time_current(); /* assumes alive */ + lp->lp_last_query = 0; /* haven't asked NI yet */ lp->lp_ping_timestamp = 0; + lp->lp_ping_feats = LNET_PING_FEAT_INVAL; lp->lp_nid = nid; - lp->lp_refcount = 2; /* 1 for caller; 1 for hash */ - lp->lp_rtr_refcount = 0; - - LNET_LOCK(); + lp->lp_cpt = cpt2; + lp->lp_refcount = 2; /* 1 for caller; 1 for hash */ + lp->lp_rtr_refcount = 0; - lp2 = lnet_find_peer_locked(nid); - if (lp2 != NULL) { - LNET_UNLOCK(); - LIBCFS_FREE(lp, sizeof(*lp)); - LNET_LOCK(); + lnet_net_lock(cpt); - if (the_lnet.ln_shutdown) { - lnet_peer_decref_locked(lp2); - *lpp = NULL; - return -ESHUTDOWN; - } + if (the_lnet.ln_shutdown) { + rc = -ESHUTDOWN; + goto out; + } - *lpp = lp2; - return 0; - } - - lp->lp_ni = lnet_net2ni_locked(LNET_NIDNET(nid)); - if (lp->lp_ni == NULL) { - LNET_UNLOCK(); - LIBCFS_FREE(lp, sizeof(*lp)); - LNET_LOCK(); - - *lpp = NULL; - return the_lnet.ln_shutdown ? -ESHUTDOWN : -EHOSTUNREACH; - } + lp2 = lnet_find_peer_locked(ptable, nid); + if (lp2 != NULL) { + *lpp = lp2; + goto out; + } - lp->lp_txcredits = - lp->lp_mintxcredits = lp->lp_ni->ni_peertxcredits; + lp->lp_ni = lnet_net2ni_locked(LNET_NIDNET(nid), cpt2); + if (lp->lp_ni == NULL) { + rc = -EHOSTUNREACH; + goto out; + } - /* As a first approximation; allow this peer the same number of router - * buffers as it is allowed outstanding sends */ - lp->lp_rtrcredits = lp->lp_minrtrcredits = lp->lp_txcredits; + lp->lp_txcredits = + lp->lp_mintxcredits = lp->lp_ni->ni_peertxcredits; + lp->lp_rtrcredits = + lp->lp_minrtrcredits = lnet_peer_buffer_credits(lp->lp_ni); - LASSERT (!the_lnet.ln_shutdown); - /* can't add peers after shutdown starts */ + list_add_tail(&lp->lp_hashlist, + &ptable->pt_hash[lnet_nid2peerhash(nid)]); + ptable->pt_version++; + *lpp = lp; - list_add_tail(&lp->lp_hashlist, lnet_nid2peerhash(nid)); - the_lnet.ln_npeers++; - the_lnet.ln_peertable_version++; - *lpp = lp; - return 0; + return 0; +out: + if (lp != NULL) + list_add(&lp->lp_hashlist, &ptable->pt_deathrow); + ptable->pt_number--; + return rc; } void lnet_debug_peer(lnet_nid_t nid) { - int rc; - lnet_peer_t *lp; - - LNET_LOCK(); - - rc = lnet_nid2peer_locked(&lp, nid); - if (rc != 0) { - LNET_UNLOCK(); + char *aliveness = "NA"; + lnet_peer_t *lp; + int rc; + int cpt; + + cpt = lnet_cpt_of_nid(nid); + lnet_net_lock(cpt); + + rc = lnet_nid2peer_locked(&lp, nid, cpt); + if (rc != 0) { + lnet_net_unlock(cpt); CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid)); return; } + if (lnet_peer_aliveness_enabled(lp)) + aliveness = lp->lp_alive ? "up" : "down"; + CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n", - libcfs_nid2str(lp->lp_nid), lp->lp_refcount, - !lnet_isrouter(lp) ? "~rtr" : (lp->lp_alive ? "up" : "down"), - lp->lp_ni->ni_peertxcredits, - lp->lp_rtrcredits, lp->lp_minrtrcredits, + libcfs_nid2str(lp->lp_nid), lp->lp_refcount, + aliveness, lp->lp_ni->ni_peertxcredits, + lp->lp_rtrcredits, lp->lp_minrtrcredits, lp->lp_txcredits, lp->lp_mintxcredits, lp->lp_txqnob); lnet_peer_decref_locked(lp); - LNET_UNLOCK(); + lnet_net_unlock(cpt); +} + +int lnet_get_peer_info(__u32 peer_index, __u64 *nid, + char aliveness[LNET_MAX_STR_LEN], + __u32 *cpt_iter, __u32 *refcount, + __u32 *ni_peer_tx_credits, __u32 *peer_tx_credits, + __u32 *peer_rtr_credits, __u32 *peer_min_rtr_credits, + __u32 *peer_tx_qnob) +{ + struct lnet_peer_table *peer_table; + lnet_peer_t *lp; + int j; + int lncpt; + bool found = false; + + /* get the number of CPTs */ + lncpt = cfs_percpt_number(the_lnet.ln_peer_tables); + + /* if the cpt number to be examined is >= the number of cpts in + * the system then indicate that there are no more cpts to examin + */ + if (*cpt_iter > lncpt) + return -ENOENT; + + /* get the current table */ + peer_table = the_lnet.ln_peer_tables[*cpt_iter]; + /* if the ptable is NULL then there are no more cpts to examine */ + if (peer_table == NULL) + return -ENOENT; + + lnet_net_lock(*cpt_iter); + + for (j = 0; j < LNET_PEER_HASH_SIZE && !found; j++) { + struct list_head *peers = &peer_table->pt_hash[j]; + + list_for_each_entry(lp, peers, lp_hashlist) { + if (peer_index-- > 0) + continue; + + snprintf(aliveness, LNET_MAX_STR_LEN, "NA"); + if (lnet_peer_aliveness_enabled(lp)) + snprintf(aliveness, LNET_MAX_STR_LEN, + lp->lp_alive ? "up" : "down"); + + *nid = lp->lp_nid; + *refcount = lp->lp_refcount; + *ni_peer_tx_credits = lp->lp_ni->ni_peertxcredits; + *peer_tx_credits = lp->lp_txcredits; + *peer_rtr_credits = lp->lp_rtrcredits; + *peer_min_rtr_credits = lp->lp_mintxcredits; + *peer_tx_qnob = lp->lp_txqnob; + + found = true; + } + + } + lnet_net_unlock(*cpt_iter); + + *cpt_iter = lncpt; + + return found ? 0 : -ENOENT; }