This is the first patch of a set of patches that enables DLC.
This patch adds some cleanup in the config.c as well as some
preparatory changes in peer.c to enable dynamic network
configuration
Signed-off-by: Amir Shehata <amir.shehata@intel.com>
Change-Id: I8c8bbf3b55acf4d76f22a8be587b553a70d31889
Reviewed-on: http://review.whamcloud.com/9830
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Liang Zhen <liang.zhen@intel.com>
Reviewed-by: James Simmons <uja.ornl@gmail.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
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,
lnet_nid_t nid);
-void lnet_peer_tables_cleanup(void);
+void lnet_peer_tables_cleanup(lnet_ni_t *ni);
void lnet_peer_tables_destroy(void);
int lnet_peer_tables_create(void);
void lnet_debug_peer(lnet_nid_t nid);
struct lnet_peer_table {
int pt_version; /* /proc validity stamp */
int pt_number; /* # peers extant */
+ int pt_zombies; /* # zombies to go to deathrow
+ * (and not there yet) */
struct list_head pt_deathrow; /* zombie peers */
struct list_head *pt_hash; /* NID->peer hash */
};
/* registered LNDs */
struct list_head ln_lnds;
- /* space for network names */
- char *ln_network_tokens;
- int ln_network_tokens_nob;
/* test protocol compatibility flags */
int ln_testprotocompat;
/* Clear the peer table and wait for all peers to go (they hold refs on
* their NIs) */
- lnet_peer_tables_cleanup();
+ 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
the_lnet.ln_shutdown = 0;
lnet_net_unlock(LNET_LOCK_EX);
-
- if (the_lnet.ln_network_tokens != NULL) {
- LIBCFS_FREE(the_lnet.ln_network_tokens,
- the_lnet.ln_network_tokens_nob);
- the_lnet.ln_network_tokens = NULL;
- }
}
int
if (nets == NULL)
goto failed;
- rc = lnet_parse_networks(&nilist, nets);
- if (rc != 0)
- goto failed;
+ rc = lnet_parse_networks(&nilist, nets);
+ if (rc != 0)
+ goto failed;
while (!list_empty(&nilist)) {
ni = list_entry(nilist.next, lnet_ni_t, ni_list);
void
lnet_ni_free(struct lnet_ni *ni)
{
+ int i;
+
if (ni->ni_refs != NULL)
cfs_percpt_free(ni->ni_refs);
pthread_mutex_destroy(&ni->ni_lock);
# endif
#endif
+ for (i = 0; i < LNET_MAX_INTERFACES &&
+ ni->ni_interfaces[i] != NULL; i++) {
+ LIBCFS_FREE(ni->ni_interfaces[i],
+ strlen(ni->ni_interfaces[i]) + 1);
+ }
LIBCFS_FREE(ni, sizeof(*ni));
}
return -EINVAL;
}
- LIBCFS_ALLOC(tokens, tokensize);
- if (tokens == NULL) {
- CERROR("Can't allocate net tokens\n");
+ LIBCFS_ALLOC(tokens, tokensize);
+ if (tokens == NULL) {
+ CERROR("Can't allocate net tokens\n");
return -ENOMEM;
- }
+ }
- the_lnet.ln_network_tokens = tokens;
- the_lnet.ln_network_tokens_nob = tokensize;
- memcpy (tokens, networks, tokensize);
+ memcpy(tokens, networks, tokensize);
str = tmp = tokens;
/* Add in the loopback network */
goto failed_syntax;
}
- if (niface == LNET_MAX_INTERFACES) {
- LCONSOLE_ERROR_MSG(0x115, "Too many interfaces "
- "for net %s\n",
- libcfs_net2str(net));
- goto failed;
- }
+ if (niface == LNET_MAX_INTERFACES) {
+ LCONSOLE_ERROR_MSG(0x115, "Too many interfaces "
+ "for net %s\n",
+ libcfs_net2str(net));
+ goto failed;
+ }
- ni->ni_interfaces[niface++] = iface;
+ /* Allocate a seperate piece of memory and copy
+ * into it the string, so we don't have
+ * a depencency on the tokens string. This way we
+ * can free the tokens at the end of the function.
+ * The newly allocated ni_interfaces[] can be
+ * freed when freeing the NI */
+ LIBCFS_ALLOC(ni->ni_interfaces[niface],
+ strlen(iface) + 1);
+ if (ni->ni_interfaces[niface] == NULL) {
+ CERROR("Can't allocate net interface name\n");
+ goto failed;
+ }
+ strncpy(ni->ni_interfaces[niface], iface,
+ strlen(iface));
+ niface++;
iface = comma;
} while (iface != NULL);
}
LASSERT(!list_empty(nilist));
+
+ LIBCFS_FREE(tokens, tokensize);
return 0;
failed_syntax:
cfs_expr_list_free(el);
LIBCFS_FREE(tokens, tokensize);
- the_lnet.ln_network_tokens = NULL;
return -EINVAL;
}
char *sep;
char *token = str;
int ntokens = 0;
- int myrc = -1;
- unsigned int hops;
- int got_hops = 0;
+ int myrc = -1;
+ unsigned int hops;
+ int got_hops = 0;
unsigned int priority = 0;
INIT_LIST_HEAD(&gateways);
}
}
- if (!got_hops)
- hops = 1;
+ if (!got_hops)
+ hops = 1;
LASSERT(!list_empty(&nets));
LASSERT(!list_empty(&gateways));
nid = libcfs_str2nid(ltb->ltb_text);
LASSERT(nid != LNET_NID_ANY);
- if (lnet_islocalnid(nid)) {
- *im_a_router = 1;
- continue;
- }
+ if (lnet_islocalnid(nid)) {
+ *im_a_router = 1;
+ continue;
+ }
rc = lnet_add_route(net, hops, nid, priority);
- if (rc != 0) {
- CERROR("Can't create route "
- "to %s via %s\n",
- libcfs_net2str(net),
- libcfs_nid2str(nid));
- goto out;
- }
+ if (rc != 0) {
+ CERROR("Can't create route "
+ "to %s via %s\n",
+ libcfs_net2str(net),
+ libcfs_nid2str(nid));
+ goto out;
+ }
}
}
- myrc = 0;
- goto out;
+ myrc = 0;
+ goto out;
- token_error:
+token_error:
lnet_syntax("routes", cmd, (int)(token - str), strlen(token));
- out:
+out:
lnet_free_text_bufs(&nets);
lnet_free_text_bufs(&gateways);
return myrc;
the_lnet.ln_peer_tables = NULL;
}
+static void
+lnet_peer_table_cleanup_locked(lnet_ni_t *ni, struct lnet_peer_table *ptable)
+{
+ 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);
+ }
+ }
+}
+
+static void
+lnet_peer_table_deathrow_wait_locked(struct lnet_peer_table *ptable,
+ int cpt_locked)
+{
+ int i;
+
+ for (i = 3; ptable->pt_zombies != 0; i++) {
+ lnet_net_unlock(cpt_locked);
+
+ 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(void)
+lnet_peer_tables_cleanup(lnet_ni_t *ni)
{
- struct lnet_peer_table *ptable;
int i;
- int j;
+ struct lnet_peer_table *ptable;
+ lnet_peer_t *lp;
+ struct list_head deathrow;
- LASSERT(the_lnet.ln_shutdown); /* i.e. no new peers */
+ 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);
-
- for (j = 0; j < LNET_PEER_HASH_SIZE; j++) {
- struct list_head *peers = &ptable->pt_hash[j];
-
- while (!list_empty(peers)) {
- lnet_peer_t *lp = list_entry(peers->next,
- lnet_peer_t,
- lp_hashlist);
- list_del_init(&lp->lp_hashlist);
- /* lose hash table's ref */
- lnet_peer_decref_locked(lp);
- }
- }
-
+ 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) {
- struct list_head deathrow = LIST_HEAD_INIT(deathrow);
- lnet_peer_t *lp;
-
lnet_net_lock(i);
+ lnet_peer_table_cleanup_locked(ni, ptable);
+ lnet_net_unlock(i);
+ }
- for (j = 3; ptable->pt_number != 0; j++) {
- lnet_net_unlock(i);
-
- if ((j & (j - 1)) == 0) {
- CDEBUG(D_WARNING,
- "Waiting for %d peers on peer table\n",
- ptable->pt_number);
- }
- cfs_pause(cfs_time_seconds(1) / 2);
- lnet_net_lock(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));
- }
+ while (!list_empty(&deathrow)) {
+ lp = list_entry(deathrow.next, lnet_peer_t, lp_hashlist);
+ list_del(&lp->lp_hashlist);
+ LIBCFS_FREE(lp, sizeof(*lp));
}
}
lp->lp_ni = NULL;
list_add(&lp->lp_hashlist, &ptable->pt_deathrow);
+ LASSERT(ptable->pt_zombies > 0);
+ ptable->pt_zombies--;
}
lnet_peer_t *