if (ip_range_len >= LNET_MAX_STR_LEN) {
snprintf(err_str,
LNET_MAX_STR_LEN,
- "\"cannot parse ip_range '%s'\"", ip_range);
+ "\"too long ip_range '%s'\"", nidstr);
err_str[LNET_MAX_STR_LEN - 1] = '\0';
rc = LUSTRE_CFG_RC_BAD_PARAM;
goto out;
if (rc != LUSTRE_CFG_RC_NO_ERR) {
snprintf(err_str,
LNET_MAX_STR_LEN,
- "\"cannot parse ip_range '%s'\"", ip_range);
+ "\"cannot parse ip_range '%.100s'\"", ip_range);
err_str[LNET_MAX_STR_LEN - 1] = '\0';
rc = LUSTRE_CFG_RC_BAD_PARAM;
goto out;
}
int lustre_lnet_config_route(char *nw, char *gw, int hops, int prio,
- int seq_no, struct cYAML **err_rc)
+ int sen, int seq_no, struct cYAML **err_rc)
{
struct lnet_ioctl_config_data data;
lnet_nid_t gateway_nid;
goto out;
}
+ if (sen == -1) {
+ sen = 1;
+ } else if (sen < 1) {
+ snprintf(err_str,
+ sizeof(err_str),
+ "\"invalid health sensitivity %d, must be 1 or greater\"",
+ sen );
+ rc = LUSTRE_CFG_RC_OUT_OF_RANGE_PARAM;
+ goto out;
+ }
+
rc = lnet_expr2ips(gw, ip_list,
&ip2nets, &net, err_str);
if (rc == LUSTRE_CFG_RC_LAST_ELEM)
data.cfg_net = rnet;
data.cfg_config_u.cfg_route.rtr_hop = hops;
data.cfg_config_u.cfg_route.rtr_priority = prio;
+ data.cfg_config_u.cfg_route.rtr_sensitivity = sen;
for (i = MAX_NUM_IPS - 1; i > ip_idx; i--) {
gateway_nid = LNET_MKNID(net, ip_list[i]);
data.cfg_nid = gateway_nid;
rc = l_ioctl(LNET_DEV_ID, IOC_LIBCFS_ADD_ROUTE, &data);
- if (rc != 0) {
+ if (rc != 0 && errno != EEXIST && errno != EHOSTUNREACH) {
rc = -errno;
snprintf(err_str,
sizeof(err_str),
cfg_route.rtr_priority) == NULL)
goto out;
+ if (cYAML_create_number(item, "health_sensitivity",
+ data.cfg_config_u.
+ cfg_route.rtr_sensitivity) == NULL)
+ goto out;
+
if (!backup &&
cYAML_create_string(item, "state",
data.cfg_config_u.cfg_route.
char val[LNET_MAX_STR_LEN];
__u32 ip;
int gni_num;
+ char *endp;
+ unsigned int num;
if (nw == NULL || nids == NULL) {
/* look at the other interfaces */
list_for_each_entry(intf, &nw->nw_intflist, intf_on_network) {
- rc = lustre_lnet_queryip(intf, &ip);
- if (rc != LUSTRE_CFG_RC_NO_ERR) {
- snprintf(err_str, str_len,
- "\"couldn't query intf %s\"", intf->intf_name);
- err_str[str_len - 1] = '\0';
- goto failed;
+ if (LNET_NETTYP(nw->nw_id) == PTL4LND) {
+ /* handle LNDs with numeric interface name */
+ num = strtoul(intf->intf_name, &endp, 0);
+ if (endp == intf->intf_name || *endp != '\0') {
+ rc = LUSTRE_CFG_RC_BAD_PARAM;
+ snprintf(err_str, str_len,
+ "\"couldn't query intf %s\"",
+ intf->intf_name);
+ err_str[str_len - 1] = '\0';
+ goto failed;
+ }
+ (*nids)[i] = LNET_MKNID(nw->nw_id, num);
+ i++;
+ } else {
+ /* handle LNDs with ip interface name */
+ rc = lustre_lnet_queryip(intf, &ip);
+ if (rc != LUSTRE_CFG_RC_NO_ERR) {
+ snprintf(err_str, str_len,
+ "\"couldn't query intf %s\"",
+ intf->intf_name);
+ err_str[str_len - 1] = '\0';
+ goto failed;
+ }
+ (*nids)[i] = LNET_MKNID(nw->nw_id, ip);
+ i++;
}
- (*nids)[i] = LNET_MKNID(nw->nw_id, ip);
- i++;
}
out:
return rc;
}
+int lustre_lnet_config_recov_intrv(int intrv, int seq_no, struct cYAML **err_rc)
+{
+ int rc = LUSTRE_CFG_RC_NO_ERR;
+ char err_str[LNET_MAX_STR_LEN];
+ char val[LNET_MAX_STR_LEN];
+
+ snprintf(err_str, sizeof(err_str), "\"success\"");
+
+ snprintf(val, sizeof(val), "%d", intrv);
+
+ rc = write_sysfs_file(modparam_path, "lnet_recovery_interval", val,
+ 1, strlen(val) + 1);
+ if (rc)
+ snprintf(err_str, sizeof(err_str),
+ "\"cannot configure recovery interval: %s\"",
+ strerror(errno));
+
+ cYAML_build_error(rc, seq_no, ADD_CMD, "recovery_interval", err_str, err_rc);
+
+ return rc;
+}
+
+int lustre_lnet_config_rtr_sensitivity(int sen, int seq_no, struct cYAML **err_rc)
+{
+ int rc = LUSTRE_CFG_RC_NO_ERR;
+ char err_str[LNET_MAX_STR_LEN];
+ char val[LNET_MAX_STR_LEN];
+
+ snprintf(err_str, sizeof(err_str), "\"success\"");
+
+ snprintf(val, sizeof(val), "%d", sen);
+
+ rc = write_sysfs_file(modparam_path, "router_sensitivity_percentage", val,
+ 1, strlen(val) + 1);
+ if (rc)
+ snprintf(err_str, sizeof(err_str),
+ "\"cannot configure router health sensitivity: %s\"",
+ strerror(errno));
+
+ cYAML_build_error(rc, seq_no, ADD_CMD, "router_sensitivity", err_str, err_rc);
+
+ return rc;
+}
+
int lustre_lnet_config_hsensitivity(int sen, int seq_no, struct cYAML **err_rc)
{
int rc = LUSTRE_CFG_RC_NO_ERR;
}
+int lustre_lnet_config_drop_asym_route(int drop, int seq_no,
+ struct cYAML **err_rc)
+{
+ int rc = LUSTRE_CFG_RC_NO_ERR;
+ char err_str[LNET_MAX_STR_LEN];
+ char val[LNET_MAX_STR_LEN];
+
+ snprintf(err_str, sizeof(err_str), "\"success\"");
+
+ snprintf(val, sizeof(val), "%u", (drop) ? 1 : 0);
+
+ rc = write_sysfs_file(modparam_path, "lnet_drop_asym_route", val,
+ 1, strlen(val) + 1);
+ if (rc)
+ snprintf(err_str, sizeof(err_str),
+ "\"cannot configure drop asym route: %s\"",
+ strerror(errno));
+
+ cYAML_build_error(rc, seq_no, ADD_CMD, "drop_asym_route",
+ err_str, err_rc);
+
+ return rc;
+
+}
+
int lustre_lnet_config_numa_range(int range, int seq_no, struct cYAML **err_rc)
{
return ioctl_set_value(range, IOC_LIBCFS_SET_NUMA_RANGE,
data.sv_value, show_rc, err_rc, l_errno);
}
+int lustre_lnet_show_recov_intrv(int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc)
+{
+ int rc = LUSTRE_CFG_RC_OUT_OF_MEM;
+ char val[LNET_MAX_STR_LEN];
+ int intrv = -1, l_errno = 0;
+ char err_str[LNET_MAX_STR_LEN];
+
+ snprintf(err_str, sizeof(err_str), "\"out of memory\"");
+
+ rc = read_sysfs_file(modparam_path, "lnet_recovery_interval", val,
+ 1, sizeof(val));
+ if (rc) {
+ l_errno = -errno;
+ snprintf(err_str, sizeof(err_str),
+ "\"cannot get recovery interval: %d\"", rc);
+ } else {
+ intrv = atoi(val);
+ }
+
+ return build_global_yaml_entry(err_str, sizeof(err_str), seq_no,
+ "recovery_interval", intrv, show_rc,
+ err_rc, l_errno);
+}
+
int lustre_lnet_show_hsensitivity(int seq_no, struct cYAML **show_rc,
struct cYAML **err_rc)
{
err_rc, l_errno);
}
+int lustre_lnet_show_rtr_sensitivity(int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc)
+{
+ int rc = LUSTRE_CFG_RC_OUT_OF_MEM;
+ char val[LNET_MAX_STR_LEN];
+ int sen = -1, l_errno = 0;
+ char err_str[LNET_MAX_STR_LEN];
+
+ snprintf(err_str, sizeof(err_str), "\"out of memory\"");
+
+ rc = read_sysfs_file(modparam_path, "router_sensitivity_percentage", val,
+ 1, sizeof(val));
+ if (rc) {
+ l_errno = -errno;
+ snprintf(err_str, sizeof(err_str),
+ "\"cannot get router sensitivity percentage: %d\"", rc);
+ } else {
+ sen = atoi(val);
+ }
+
+ return build_global_yaml_entry(err_str, sizeof(err_str), seq_no,
+ "router_sensitivity", sen, show_rc,
+ err_rc, l_errno);
+}
+
int lustre_lnet_show_transaction_to(int seq_no, struct cYAML **show_rc,
struct cYAML **err_rc)
{
goto out;
}
+ if (nid_list.rlst_num_nids == 0)
+ goto out;
+
root = cYAML_create_object(NULL, NULL);
if (root == NULL)
goto out;
err_rc, l_errno);
}
+int lustre_lnet_show_drop_asym_route(int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc)
+{
+ int rc = LUSTRE_CFG_RC_OUT_OF_MEM;
+ char val[LNET_MAX_STR_LEN];
+ int drop_asym_route = -1, l_errno = 0;
+ char err_str[LNET_MAX_STR_LEN];
+
+ snprintf(err_str, sizeof(err_str), "\"out of memory\"");
+
+ rc = read_sysfs_file(modparam_path, "lnet_drop_asym_route", val,
+ 1, sizeof(val));
+ if (rc) {
+ l_errno = -errno;
+ snprintf(err_str, sizeof(err_str),
+ "\"cannot get drop asym route setting: %d\"", rc);
+ } else {
+ drop_asym_route = atoi(val);
+ }
+
+ return build_global_yaml_entry(err_str, sizeof(err_str), seq_no,
+ "drop_asym_route", drop_asym_route,
+ show_rc, err_rc, l_errno);
+}
+
int lustre_lnet_show_numa_range(int seq_no, struct cYAML **show_rc,
struct cYAML **err_rc)
{
struct cYAML **err_rc)
{
struct lnet_ioctl_lnet_stats data;
+ struct lnet_counters *cntrs;
int rc;
int l_errno;
char err_str[LNET_MAX_STR_LEN];
LIBCFS_IOC_INIT_V2(data, st_hdr);
rc = l_ioctl(LNET_DEV_ID, IOC_LIBCFS_GET_LNET_STATS, &data);
- if (rc != 0) {
+ if (rc) {
l_errno = errno;
snprintf(err_str,
sizeof(err_str),
rc = LUSTRE_CFG_RC_OUT_OF_MEM;
+ cntrs = &data.st_cntrs;
+
root = cYAML_create_object(NULL, NULL);
- if (root == NULL)
+ if (!root)
goto out;
stats = cYAML_create_object(root, "statistics");
- if (stats == NULL)
+ if (!stats)
goto out;
- if (cYAML_create_number(stats, "msgs_alloc",
- data.st_cntrs.msgs_alloc) == NULL)
+ if (!cYAML_create_number(stats, "msgs_alloc",
+ cntrs->lct_common.lcc_msgs_alloc))
goto out;
- if (cYAML_create_number(stats, "msgs_max",
- data.st_cntrs.msgs_max) == NULL)
+ if (!cYAML_create_number(stats, "msgs_max",
+ cntrs->lct_common.lcc_msgs_max))
goto out;
- if (cYAML_create_number(stats, "rst_alloc",
- data.st_cntrs.rst_alloc) == NULL)
+ if (!cYAML_create_number(stats, "rst_alloc",
+ cntrs->lct_health.lch_rst_alloc))
goto out;
- if (cYAML_create_number(stats, "errors",
- data.st_cntrs.errors) == NULL)
+ if (!cYAML_create_number(stats, "errors",
+ cntrs->lct_common.lcc_errors))
goto out;
- if (cYAML_create_number(stats, "send_count",
- data.st_cntrs.send_count) == NULL)
+ if (!cYAML_create_number(stats, "send_count",
+ cntrs->lct_common.lcc_send_count))
goto out;
- if (cYAML_create_number(stats, "resend_count",
- data.st_cntrs.resend_count) == NULL)
+ if (!cYAML_create_number(stats, "resend_count",
+ cntrs->lct_health.lch_resend_count))
goto out;
- if (cYAML_create_number(stats, "response_timeout_count",
- data.st_cntrs.response_timeout_count) == NULL)
+ if (!cYAML_create_number(stats, "response_timeout_count",
+ cntrs->lct_health.lch_response_timeout_count))
goto out;
- if (cYAML_create_number(stats, "local_interrupt_count",
- data.st_cntrs.local_interrupt_count) == NULL)
+ if (!cYAML_create_number(stats, "local_interrupt_count",
+ cntrs->lct_health.lch_local_interrupt_count))
goto out;
- if (cYAML_create_number(stats, "local_dropped_count",
- data.st_cntrs.local_dropped_count) == NULL)
+ if (!cYAML_create_number(stats, "local_dropped_count",
+ cntrs->lct_health.lch_local_dropped_count))
goto out;
- if (cYAML_create_number(stats, "local_aborted_count",
- data.st_cntrs.local_aborted_count) == NULL)
+ if (!cYAML_create_number(stats, "local_aborted_count",
+ cntrs->lct_health.lch_local_aborted_count))
goto out;
- if (cYAML_create_number(stats, "local_no_route_count",
- data.st_cntrs.local_no_route_count) == NULL)
+ if (!cYAML_create_number(stats, "local_no_route_count",
+ cntrs->lct_health.lch_local_no_route_count))
goto out;
- if (cYAML_create_number(stats, "local_timeout_count",
- data.st_cntrs.local_timeout_count) == NULL)
+ if (!cYAML_create_number(stats, "local_timeout_count",
+ cntrs->lct_health.lch_local_timeout_count))
goto out;
- if (cYAML_create_number(stats, "local_error_count",
- data.st_cntrs.local_error_count) == NULL)
+ if (!cYAML_create_number(stats, "local_error_count",
+ cntrs->lct_health.lch_local_error_count))
goto out;
- if (cYAML_create_number(stats, "remote_dropped_count",
- data.st_cntrs.remote_dropped_count) == NULL)
+ if (!cYAML_create_number(stats, "remote_dropped_count",
+ cntrs->lct_health.lch_remote_dropped_count))
goto out;
- if (cYAML_create_number(stats, "remote_error_count",
- data.st_cntrs.remote_error_count) == NULL)
+ if (!cYAML_create_number(stats, "remote_error_count",
+ cntrs->lct_health.lch_remote_error_count))
goto out;
- if (cYAML_create_number(stats, "remote_timeout_count",
- data.st_cntrs.remote_timeout_count) == NULL)
+ if (!cYAML_create_number(stats, "remote_timeout_count",
+ cntrs->lct_health.lch_remote_timeout_count))
goto out;
- if (cYAML_create_number(stats, "network_timeout_count",
- data.st_cntrs.network_timeout_count) == NULL)
+ if (!cYAML_create_number(stats, "network_timeout_count",
+ cntrs->lct_health.lch_network_timeout_count))
goto out;
- if (cYAML_create_number(stats, "recv_count",
- data.st_cntrs.recv_count) == NULL)
+ if (!cYAML_create_number(stats, "recv_count",
+ cntrs->lct_common.lcc_recv_count))
goto out;
- if (cYAML_create_number(stats, "route_count",
- data.st_cntrs.route_count) == NULL)
+ if (!cYAML_create_number(stats, "route_count",
+ cntrs->lct_common.lcc_route_count))
goto out;
- if (cYAML_create_number(stats, "drop_count",
- data.st_cntrs.drop_count) == NULL)
+ if (!cYAML_create_number(stats, "drop_count",
+ cntrs->lct_common.lcc_drop_count))
goto out;
- if (cYAML_create_number(stats, "send_length",
- data.st_cntrs.send_length) == NULL)
+ if (!cYAML_create_number(stats, "send_length",
+ cntrs->lct_common.lcc_send_length))
goto out;
- if (cYAML_create_number(stats, "recv_length",
- data.st_cntrs.recv_length) == NULL)
+ if (!cYAML_create_number(stats, "recv_length",
+ cntrs->lct_common.lcc_recv_length))
goto out;
- if (cYAML_create_number(stats, "route_length",
- data.st_cntrs.route_length) == NULL)
+ if (!cYAML_create_number(stats, "route_length",
+ cntrs->lct_common.lcc_route_length))
goto out;
- if (cYAML_create_number(stats, "drop_length",
- data.st_cntrs.drop_length) == NULL)
+ if (!cYAML_create_number(stats, "drop_length",
+ cntrs->lct_common.lcc_drop_length))
goto out;
- if (show_rc == NULL)
+ if (!show_rc)
cYAML_print_tree(root);
snprintf(err_str, sizeof(err_str), "\"success\"");
static int handle_yaml_config_route(struct cYAML *tree, struct cYAML **show_rc,
struct cYAML **err_rc)
{
- struct cYAML *net, *gw, *hop, *prio, *seq_no;
+ struct cYAML *net, *gw, *hop, *prio, *sen, *seq_no;
net = cYAML_get_object_item(tree, "net");
gw = cYAML_get_object_item(tree, "gateway");
hop = cYAML_get_object_item(tree, "hop");
prio = cYAML_get_object_item(tree, "priority");
+ sen = cYAML_get_object_item(tree, "health_sensitivity");
seq_no = cYAML_get_object_item(tree, "seq_no");
return lustre_lnet_config_route((net) ? net->cy_valuestring : NULL,
(gw) ? gw->cy_valuestring : NULL,
(hop) ? hop->cy_valueint : -1,
(prio) ? prio->cy_valueint : -1,
+ (sen) ? sen->cy_valueint : -1,
(seq_no) ? seq_no->cy_valueint : -1,
err_rc);
}
if (!entry || !entry->cy_valuestring)
continue;
- nids[num] = calloc(strlen(entry->cy_valuestring) + 1, 1);
+ nids[num] = strdup(entry->cy_valuestring);
if (!nids[num]) {
rc = LUSTRE_CFG_RC_OUT_OF_MEM;
goto failed;
}
- strncpy(nids[num], entry->cy_valuestring,
- strlen(entry->cy_valuestring));
num++;
}
rc = num;
struct cYAML **err_rc)
{
struct cYAML *max_intf, *numa, *discovery, *retry, *tto, *seq_no,
- *sen;
+ *sen, *recov, *rsen, *drop_asym_route;
int rc = 0;
seq_no = cYAML_get_object_item(tree, "seq_no");
: -1,
err_rc);
+ drop_asym_route = cYAML_get_object_item(tree, "drop_asym_route");
+ if (drop_asym_route)
+ rc = lustre_lnet_config_drop_asym_route(
+ drop_asym_route->cy_valueint,
+ seq_no ? seq_no->cy_valueint : -1,
+ err_rc);
+
retry = cYAML_get_object_item(tree, "retry_count");
if (retry)
rc = lustre_lnet_config_retry_count(retry->cy_valueint,
: -1,
err_rc);
+ recov = cYAML_get_object_item(tree, "recovery_interval");
+ if (recov)
+ rc = lustre_lnet_config_recov_intrv(recov->cy_valueint,
+ seq_no ? seq_no->cy_valueint
+ : -1,
+ err_rc);
+
+ rsen = cYAML_get_object_item(tree, "router_sensitivity");
+ if (rsen)
+ rc = lustre_lnet_config_rtr_sensitivity(rsen->cy_valueint,
+ seq_no ? seq_no->cy_valueint
+ : -1,
+ err_rc);
+
return rc;
}
struct cYAML **show_rc,
struct cYAML **err_rc)
{
- struct cYAML *max_intf, *numa, *discovery, *seq_no;
+ struct cYAML *max_intf, *numa, *discovery, *seq_no, *drop_asym_route;
int rc = 0;
seq_no = cYAML_get_object_item(tree, "seq_no");
: -1,
err_rc);
+ /* asymmetrical route messages are accepted by default */
+ drop_asym_route = cYAML_get_object_item(tree, "drop_asym_route");
+ if (drop_asym_route)
+ rc = lustre_lnet_config_drop_asym_route(
+ 0, seq_no ? seq_no->cy_valueint : -1, err_rc);
+
return rc;
}
struct cYAML **err_rc)
{
struct cYAML *max_intf, *numa, *discovery, *retry, *tto, *seq_no,
- *sen;
+ *sen, *recov, *rsen, *drop_asym_route;
int rc = 0;
seq_no = cYAML_get_object_item(tree, "seq_no");
: -1,
show_rc, err_rc);
+ drop_asym_route = cYAML_get_object_item(tree, "drop_asym_route");
+ if (drop_asym_route)
+ rc = lustre_lnet_show_drop_asym_route(
+ seq_no ? seq_no->cy_valueint : -1,
+ show_rc, err_rc);
+
retry = cYAML_get_object_item(tree, "retry_count");
if (retry)
rc = lustre_lnet_show_retry_count(seq_no ? seq_no->cy_valueint
: -1,
show_rc, err_rc);
+ recov = cYAML_get_object_item(tree, "recovery_interval");
+ if (recov)
+ rc = lustre_lnet_show_recov_intrv(seq_no ? seq_no->cy_valueint
+ : -1,
+ show_rc, err_rc);
+
+ rsen = cYAML_get_object_item(tree, "router_sensitivity");
+ if (rsen)
+ rc = lustre_lnet_show_hsensitivity(seq_no ? seq_no->cy_valueint
+ : -1,
+ show_rc, err_rc);
+
return rc;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff