* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_LNET
#include <lnet/lib-lnet.h>
+#ifdef __KERNEL__
+#include <linux/log2.h>
+#endif
#ifdef __KERNEL__
#define D_LNI D_CONSOLE
#endif
lnet_t the_lnet; /* THE state of the network */
+EXPORT_SYMBOL(the_lnet);
#ifdef __KERNEL__
CFS_MODULE_PARM(routes, "s", charp, 0444,
"routes to non-local networks");
+static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
+CFS_MODULE_PARM(rnet_htable_size, "i", int, 0444,
+ "size of remote network hash table");
+
char *
lnet_get_routes(void)
{
void
lnet_init_locks(void)
{
- cfs_spin_lock_init(&the_lnet.ln_eq_wait_lock);
- cfs_waitq_init(&the_lnet.ln_eq_waitq);
- cfs_mutex_init(&the_lnet.ln_lnd_mutex);
- cfs_mutex_init(&the_lnet.ln_api_mutex);
+ spin_lock_init(&the_lnet.ln_eq_wait_lock);
+ init_waitqueue_head(&the_lnet.ln_eq_waitq);
+ mutex_init(&the_lnet.ln_lnd_mutex);
+ mutex_init(&the_lnet.ln_api_mutex);
}
void
{
static char default_networks[256];
char *networks = getenv ("LNET_NETWORKS");
- char *ip2nets = getenv ("LNET_IP2NETS");
char *str;
char *sep;
int len;
int nob;
- int rc;
cfs_list_t *tmp;
-#ifdef NOT_YET
- if (networks != NULL && ip2nets != NULL) {
- LCONSOLE_ERROR_MSG(0x103, "Please set EITHER 'LNET_NETWORKS' or"
- " 'LNET_IP2NETS' but not both at once\n");
- return NULL;
- }
-
- if (ip2nets != NULL) {
- rc = lnet_parse_ip2nets(&networks, ip2nets);
- return (rc == 0) ? networks : NULL;
- }
-#else
- SET_BUT_UNUSED(ip2nets);
- SET_BUT_UNUSED(rc);
-#endif
if (networks != NULL)
return networks;
/* In userland, the default 'networks=' is the list of known net types */
-
len = sizeof(default_networks);
str = default_networks;
*str = 0;
nob = snprintf(str, len, "%s%s", sep,
libcfs_lnd2str(lnd->lnd_type));
+ if (nob >= len) {
+ /* overflowed the string; leave it where it was */
+ *str = 0;
+ break;
+ }
len -= nob;
- if (len < 0) {
- /* overflowed the string; leave it where it was */
- *str = 0;
- break;
- }
-
str += nob;
sep = ",";
}
#endif
static int
-lnet_create_locks(void)
+lnet_create_remote_nets_table(void)
{
- lnet_init_locks();
+ int i;
+ cfs_list_t *hash;
+
+ LASSERT(the_lnet.ln_remote_nets_hash == NULL);
+ LASSERT(the_lnet.ln_remote_nets_hbits > 0);
+ LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
+ if (hash == NULL) {
+ CERROR("Failed to create remote nets hash table\n");
+ return -ENOMEM;
+ }
- the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
- if (the_lnet.ln_res_lock == NULL)
- goto failed;
+ for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
+ CFS_INIT_LIST_HEAD(&hash[i]);
+ the_lnet.ln_remote_nets_hash = hash;
+ return 0;
+}
- the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
- if (the_lnet.ln_net_lock == NULL)
- goto failed;
+static void
+lnet_destroy_remote_nets_table(void)
+{
+ int i;
+ cfs_list_t *hash;
- return 0;
+ if (the_lnet.ln_remote_nets_hash == NULL)
+ return;
- failed:
- lnet_fini_locks();
- return -ENOMEM;
+ for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
+ LASSERT(cfs_list_empty(&the_lnet.ln_remote_nets_hash[i]));
+
+ LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
+ LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
+ the_lnet.ln_remote_nets_hash = NULL;
}
static void
lnet_fini_locks();
}
+static int
+lnet_create_locks(void)
+{
+ lnet_init_locks();
+
+ the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
+ if (the_lnet.ln_res_lock == NULL)
+ goto failed;
+
+ the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
+ if (the_lnet.ln_net_lock == NULL)
+ goto failed;
+
+ return 0;
+
+ failed:
+ lnet_destroy_locks();
+ return -ENOMEM;
+}
+
void lnet_assert_wire_constants (void)
{
/* Wire protocol assertions generated by 'wirecheck'
LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
}
+EXPORT_SYMBOL(lnet_register_lnd);
void
lnet_unregister_lnd (lnd_t *lnd)
LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
}
+EXPORT_SYMBOL(lnet_unregister_lnd);
void
lnet_counters_get(lnet_counters_t *counters)
counters->send_count += ctr->send_count;
counters->recv_count += ctr->recv_count;
counters->route_count += ctr->route_count;
- counters->drop_length += ctr->drop_length;
+ counters->drop_count += ctr->drop_count;
counters->send_length += ctr->send_length;
counters->recv_length += ctr->recv_length;
counters->route_length += ctr->route_length;
#endif /* LNET_USE_LIB_FREELIST */
-__u64
-lnet_create_interface_cookie (void)
+__u64 lnet_create_interface_cookie (void)
{
- /* NB the interface cookie in wire handles guards against delayed
- * replies and ACKs appearing valid after reboot. Initialisation time,
- * even if it's only implemented to millisecond resolution is probably
- * easily good enough. */
- struct timeval tv;
- __u64 cookie;
+ /* NB the interface cookie in wire handles guards against delayed
+ * replies and ACKs appearing valid after reboot. Initialisation time,
+ * even if it's only implemented to millisecond resolution is probably
+ * easily good enough. */
+ struct timeval tv;
+ __u64 cookie;
#ifndef __KERNEL__
- int rc = gettimeofday (&tv, NULL);
- LASSERT (rc == 0);
+ int rc = gettimeofday (&tv, NULL);
+ LASSERT (rc == 0);
#else
- cfs_gettimeofday(&tv);
+ do_gettimeofday(&tv);
#endif
- cookie = tv.tv_sec;
- cookie *= 1000000;
- cookie += tv.tv_usec;
- return cookie;
+ cookie = tv.tv_sec;
+ cookie *= 1000000;
+ cookie += tv.tv_usec;
+ return cookie;
}
static char *
the_lnet.ln_pid = requested_pid;
#else
if (the_lnet.ln_server_mode_flag) {/* server case (uOSS) */
- LASSERT ((requested_pid & LNET_PID_USERFLAG) == 0);
+ LASSERT ((requested_pid & LNET_PID_USERFLAG) == 0);
- if (cfs_curproc_uid())/* Only root can run user-space server */
- return -EPERM;
- the_lnet.ln_pid = requested_pid;
+ if (current_uid() != 0) /* Only root can run user-space server */
+ return -EPERM;
+ the_lnet.ln_pid = requested_pid;
} else {/* client case (liblustre) */
CFS_INIT_LIST_HEAD(&the_lnet.ln_nis);
CFS_INIT_LIST_HEAD(&the_lnet.ln_nis_cpt);
CFS_INIT_LIST_HEAD(&the_lnet.ln_nis_zombie);
- CFS_INIT_LIST_HEAD(&the_lnet.ln_remote_nets);
CFS_INIT_LIST_HEAD(&the_lnet.ln_routers);
+ rc = lnet_create_remote_nets_table();
+ if (rc != 0)
+ goto failed;
+
the_lnet.ln_interface_cookie = lnet_create_interface_cookie();
the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
cfs_percpt_free(the_lnet.ln_counters);
the_lnet.ln_counters = NULL;
}
+ lnet_destroy_remote_nets_table();
return 0;
}
if (number == 1)
return 0;
- val = cfs_hash_long(key, LNET_CPT_BITS);
+ val = hash_long(key, LNET_CPT_BITS);
/* NB: LNET_CP_NUMBER doesn't have to be PO2 */
if (val < number)
return val;
LASSERT(!the_lnet.ln_shutdown);
LASSERT(the_lnet.ln_refcount == 0);
LASSERT(cfs_list_empty(&the_lnet.ln_nis_zombie));
- LASSERT(cfs_list_empty(&the_lnet.ln_remote_nets));
lnet_net_lock(LNET_LOCK_EX);
the_lnet.ln_shutdown = 1; /* flag shutdown */
ni->ni_lnd->lnd_refcount--;
lnet_net_unlock(LNET_LOCK_EX);
- islo = ni->ni_lnd->lnd_type == LOLND;
+ islo = ni->ni_lnd->lnd_type == LOLND;
- LASSERT (!cfs_in_interrupt ());
- (ni->ni_lnd->lnd_shutdown)(ni);
+ LASSERT (!in_interrupt ());
+ (ni->ni_lnd->lnd_shutdown)(ni);
- /* can't deref lnd anymore now; it might have unregistered
- * itself... */
+ /* can't deref lnd anymore now; it might have unregistered
+ * itself... */
if (!islo)
CDEBUG(D_LNI, "Removed LNI %s\n",
#ifdef __KERNEL__
if (lnd == NULL) {
- LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
- rc = cfs_request_module("%s",
- libcfs_lnd2modname(lnd_type));
- LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
+ LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex);
+ rc = request_module("%s",
+ libcfs_lnd2modname(lnd_type));
+ LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex);
lnd = lnet_find_lnd_by_type(lnd_type);
if (lnd == NULL) {
CFS_INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
#ifdef __KERNEL__
+ /* The hash table size is the number of bits it takes to express the set
+ * ln_num_routes, minus 1 (better to under estimate than over so we
+ * don't waste memory). */
+ if (rnet_htable_size <= 0)
+ rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
+ else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
+ rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
+ the_lnet.ln_remote_nets_hbits = max_t(int, 1,
+ order_base_2(rnet_htable_size) - 1);
+
/* All LNDs apart from the LOLND are in separate modules. They
* register themselves when their module loads, and unregister
* themselves when their module is unloaded. */
#else
+ the_lnet.ln_remote_nets_hbits = 8;
+
/* Register LNDs
* NB the order here determines default 'networks=' order */
-# ifdef CRAY_XT3
- LNET_REGISTER_ULND(the_ptllnd);
-# endif
# ifdef HAVE_LIBPTHREAD
LNET_REGISTER_ULND(the_tcplnd);
# endif
lnet_register_lnd(&the_lolnd);
return 0;
}
+EXPORT_SYMBOL(LNetInit);
/**
* Finalize LNet library.
the_lnet.ln_init = 0;
}
+EXPORT_SYMBOL(LNetFini);
/**
* Set LNet PID and start LNet interfaces, routing, and forwarding.
LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
return rc;
}
+EXPORT_SYMBOL(LNetNIInit);
/**
* Stop LNet interfaces, routing, and forwarding.
LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex);
return 0;
}
+EXPORT_SYMBOL(LNetNIFini);
/**
* This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and
return lnet_fail_nid(data->ioc_nid, data->ioc_count);
case IOC_LIBCFS_ADD_ROUTE:
- rc = lnet_add_route(data->ioc_net, data->ioc_count,
- data->ioc_nid);
+ rc = lnet_add_route(data->ioc_net, data->ioc_count,
+ data->ioc_nid, data->ioc_priority);
return (rc != 0) ? rc : lnet_check_routes();
case IOC_LIBCFS_DEL_ROUTE:
case IOC_LIBCFS_GET_ROUTE:
return lnet_get_route(data->ioc_count,
&data->ioc_net, &data->ioc_count,
- &data->ioc_nid, &data->ioc_flags);
+ &data->ioc_nid, &data->ioc_flags,
+ &data->ioc_priority);
case IOC_LIBCFS_NOTIFY_ROUTER:
return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
cfs_time_current() -
}
/* not reached */
}
+EXPORT_SYMBOL(LNetCtl);
/**
* Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
lnet_net_unlock(cpt);
return rc;
}
+EXPORT_SYMBOL(LNetGetId);
/**
* Print a string representation of handle \a h into buffer \a str of
{
snprintf(str, len, LPX64, h.cookie);
}
+EXPORT_SYMBOL(LNetSnprintHandle);
static int
lnet_create_ping_info(void)
void
lnet_ping_target_fini(void)
{
- lnet_event_t event;
- int rc;
- int which;
- int timeout_ms = 1000;
- cfs_sigset_t blocked = cfs_block_allsigs();
+ 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 */
+ 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,
int
lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_ids)
{
- lnet_handle_eq_t eqh;
- lnet_handle_md_t mdh;
- lnet_event_t event;
- lnet_md_t md = {0};
- int which;
- int unlinked = 0;
- int replied = 0;
- const int a_long_time = 60000; /* mS */
- int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
- lnet_ping_info_t *info;
- lnet_process_id_t tmpid;
- int i;
- int nob;
- int rc;
- int rc2;
- cfs_sigset_t blocked;
-
- if (n_ids <= 0 ||
- id.nid == LNET_NID_ANY ||
- timeout_ms > 500000 || /* arbitrary limit! */
- n_ids > 20) /* arbitrary limit! */
- return -EINVAL;
+ lnet_handle_eq_t eqh;
+ lnet_handle_md_t mdh;
+ lnet_event_t event;
+ lnet_md_t md = {0};
+ int which;
+ int unlinked = 0;
+ int replied = 0;
+ const int a_long_time = 60000; /* mS */
+ int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
+ lnet_ping_info_t *info;
+ lnet_process_id_t tmpid;
+ int i;
+ int nob;
+ int rc;
+ int rc2;
+ sigset_t blocked;
+
+ if (n_ids <= 0 ||
+ id.nid == LNET_NID_ANY ||
+ timeout_ms > 500000 || /* arbitrary limit! */
+ n_ids > 20) /* arbitrary limit! */
+ return -EINVAL;
if (id.pid == LNET_PID_ANY)
id.pid = LUSTRE_SRV_LNET_PID;
for (i = 0; i < n_ids; i++) {
tmpid.pid = info->pi_pid;
tmpid.nid = info->pi_ni[i].ns_nid;
-#ifdef __KERNEL__
- if (cfs_copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
+ if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
goto out_1;
-#else
- ids[i] = tmpid;
-#endif
}
rc = info->pi_nnis;