4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lnet/selftest/conctl.c
36 * Author: Liang Zhen <liangzhen@clusterfs.com>
41 #include <libcfs/libcfsutil.h>
42 #include <lnet/lnetctl.h>
43 #include <lnet/lnetst.h>
46 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
47 static lst_sid_t session_id;
48 static int session_key;
49 static lstcon_trans_stat_t trans_stat;
51 typedef struct list_string {
52 struct list_string *lstr_next;
58 # define offsetof(typ,memb) ((unsigned long)((char *)&(((typ *)0)->memb)))
61 static int alloc_count = 0;
62 static int alloc_nob = 0;
67 lstr_t *lstr = malloc(offsetof(lstr_t, lstr_str[sz]));
70 fprintf(stderr, "Can't allocate lstr\n");
77 lstr->lstr_str[0] = 0;
83 free_lstr(lstr_t *lstr)
86 alloc_nob -= lstr->lstr_sz;
91 free_lstrs(lstr_t **list)
95 while ((lstr = *list) != NULL) {
96 *list = lstr->lstr_next;
102 new_lstrs(lstr_t **list, char *prefix, char *postfix,
103 int lo, int hi, int stride)
105 int n1 = strlen(prefix);
106 int n2 = strlen(postfix);
107 int sz = n1 + 20 + n2 + 1;
110 lstr_t *n = alloc_lstr(sz);
112 snprintf(n->lstr_str, sz - 1, "%s%u%s",
113 prefix, lo, postfix);
115 n->lstr_next = *list;
123 expand_lstr(lstr_t **list, lstr_t *l)
125 int nob = strlen(l->lstr_str);
135 b1 = strchr(l->lstr_str, '[');
137 l->lstr_next = *list;
142 b2 = strchr(b1, ']');
143 if (b2 == NULL || b2 == b1 + 1)
151 sep = strchr(expr, ',');
157 if (sscanf(expr, "%u%n", &x, &n) >= 1 && n == nob) {
159 new_lstrs(list, l->lstr_str, b2, x, x, 1);
164 if (sscanf(expr, "%u-%u%n", &x, &y, &n) >= 2 && n == nob &&
167 new_lstrs(list, l->lstr_str, b2, x, y, 1);
172 if (sscanf(expr, "%u-%u/%u%n", &x, &y, &z, &n) >= 3 && n == nob &&
175 new_lstrs(list, l->lstr_str, b2, x, y, z);
181 } while ((expr = sep) != NULL);
189 expand_strs(char *str, lstr_t **head)
197 l = alloc_lstr(strlen(str) + 1);
198 memcpy(l->lstr_str, str, strlen(str) + 1);
206 while ((l = list) != NULL) {
209 rc = expand_lstr(&nlist, l);
211 fprintf(stderr, "Syntax error in \"%s\"\n", str);
219 /* re-order onto 'list' */
220 while ((l = nlist) != NULL) {
221 nlist = l->lstr_next;
226 } while (expanded && rc > 0);
233 while ((l = list) != NULL) {
242 lst_parse_nids(char *str, int *countp, lnet_process_id_t **idspp)
250 rc = expand_strs(str, &head);
260 *idspp = malloc(c * sizeof(lnet_process_id_t));
261 if (*idspp == NULL) {
262 fprintf(stderr, "Out of memory\n");
269 while ((l = head) != NULL) {
273 (*idspp)[i].nid = libcfs_str2nid(l->lstr_str);
274 if ((*idspp)[i].nid == LNET_NID_ANY) {
275 fprintf(stderr, "Invalid nid: %s\n",
280 (*idspp)[i].pid = LUSTRE_LNET_PID;
297 lst_node_state2str(int state)
299 if (state == LST_NODE_ACTIVE)
301 if (state == LST_NODE_BUSY)
303 if (state == LST_NODE_DOWN)
310 lst_node_str2state(char *str)
312 if (strcasecmp(str, "active") == 0)
313 return LST_NODE_ACTIVE;
314 if (strcasecmp(str, "busy") == 0)
315 return LST_NODE_BUSY;
316 if (strcasecmp(str, "down") == 0)
317 return LST_NODE_DOWN;
318 if (strcasecmp(str, "unknown") == 0)
319 return LST_NODE_UNKNOWN;
320 if (strcasecmp(str, "invalid") == 0)
321 return (LST_NODE_UNKNOWN | LST_NODE_DOWN | LST_NODE_BUSY);
327 lst_test_type2name(int type)
329 if (type == LST_TEST_PING)
331 if (type == LST_TEST_BULK)
338 lst_test_name2type(char *name)
340 if (strcasecmp(name, "ping") == 0)
341 return LST_TEST_PING;
342 if (strcasecmp(name, "brw") == 0)
343 return LST_TEST_BULK;
349 lst_print_usage(char *cmd)
351 Parser_printhelp(cmd);
355 lst_print_error(char *sub, const char *def_format, ...)
359 /* local error returned from kernel */
362 fprintf(stderr, "No session exists\n");
365 fprintf(stderr, "Session is shutting down\n");
368 fprintf(stderr, "Unmatched session key or not root\n");
371 fprintf(stderr, "Can't find %s in current session\n", sub);
374 fprintf(stderr, "Invalid parameters list in command line\n");
377 fprintf(stderr, "Bad parameter address\n");
380 fprintf(stderr, "%s already exists\n", sub);
383 va_start(ap, def_format);
384 vfprintf(stderr, def_format, ap);
392 lst_free_rpcent(cfs_list_t *head)
394 lstcon_rpc_ent_t *ent;
396 while (!cfs_list_empty(head)) {
397 ent = cfs_list_entry(head->next, lstcon_rpc_ent_t, rpe_link);
399 cfs_list_del(&ent->rpe_link);
405 lst_reset_rpcent(cfs_list_t *head)
407 lstcon_rpc_ent_t *ent;
409 cfs_list_for_each_entry_typed(ent, head, lstcon_rpc_ent_t, rpe_link) {
410 ent->rpe_sid = LST_INVALID_SID;
411 ent->rpe_peer.nid = LNET_NID_ANY;
412 ent->rpe_peer.pid = LNET_PID_ANY;
413 ent->rpe_rpc_errno = ent->rpe_fwk_errno = 0;
418 lst_alloc_rpcent(cfs_list_t *head, int count, int offset)
420 lstcon_rpc_ent_t *ent;
423 for (i = 0; i < count; i++) {
424 ent = malloc(offsetof(lstcon_rpc_ent_t, rpe_payload[offset]));
426 lst_free_rpcent(head);
430 memset(ent, 0, offsetof(lstcon_rpc_ent_t, rpe_payload[offset]));
432 ent->rpe_sid = LST_INVALID_SID;
433 ent->rpe_peer.nid = LNET_NID_ANY;
434 ent->rpe_peer.pid = LNET_PID_ANY;
435 cfs_list_add(&ent->rpe_link, head);
442 lst_print_transerr(cfs_list_t *head, char *optstr)
444 lstcon_rpc_ent_t *ent;
446 cfs_list_for_each_entry_typed(ent, head, lstcon_rpc_ent_t, rpe_link) {
447 if (ent->rpe_rpc_errno == 0 && ent->rpe_fwk_errno == 0)
450 if (ent->rpe_rpc_errno != 0) {
451 fprintf(stderr, "%s RPC failed on %s: %s\n",
452 optstr, libcfs_id2str(ent->rpe_peer),
453 strerror(ent->rpe_rpc_errno));
457 fprintf(stderr, "%s failed on %s: %s\n",
458 optstr, libcfs_id2str(ent->rpe_peer),
459 strerror(ent->rpe_fwk_errno));
463 int lst_info_batch_ioctl(char *batch, int test, int server,
464 lstcon_test_batch_ent_t *entp, int *idxp,
465 int *ndentp, lstcon_node_ent_t *dentsp);
467 int lst_info_group_ioctl(char *name, lstcon_ndlist_ent_t *gent,
468 int *idx, int *count, lstcon_node_ent_t *dents);
470 int lst_query_batch_ioctl(char *batch, int test, int server,
471 int timeout, cfs_list_t *head);
474 lst_ioctl(unsigned int opc, void *buf, int len)
476 struct libcfs_ioctl_data data;
479 LIBCFS_IOC_INIT (data);
480 data.ioc_u32[0] = opc;
481 data.ioc_plen1 = len;
482 data.ioc_pbuf1 = (char *)buf;
483 data.ioc_plen2 = sizeof(trans_stat);
484 data.ioc_pbuf2 = (char *)&trans_stat;
486 memset(&trans_stat, 0, sizeof(trans_stat));
488 rc = l_ioctl(LNET_DEV_ID, IOC_LIBCFS_LNETST, &data);
490 /* local error, no valid RPC result */
495 if (trans_stat.trs_rpc_errno != 0)
498 /* Framework error */
499 if (trans_stat.trs_fwk_errno != 0)
506 lst_new_session_ioctl (char *name, int timeout, int force, lst_sid_t *sid)
508 lstio_session_new_args_t args = {0};
510 args.lstio_ses_key = session_key;
511 args.lstio_ses_timeout = timeout;
512 args.lstio_ses_force = force;
513 args.lstio_ses_idp = sid;
514 args.lstio_ses_nmlen = strlen(name);
515 args.lstio_ses_namep = name;
517 return lst_ioctl (LSTIO_SESSION_NEW, &args, sizeof(args));
521 jt_lst_new_session(int argc, char **argv)
523 char buf[LST_NAME_SIZE];
531 static struct option session_opts[] =
533 {"timeout", required_argument, 0, 't' },
534 {"force", no_argument, 0, 'f' },
538 if (session_key == 0) {
540 "Can't find env LST_SESSION or value is not valid\n");
546 c = getopt_long(argc, argv, "ft:",
547 session_opts, &optidx);
557 timeout = atoi(optarg);
560 lst_print_usage(argv[0]);
566 fprintf(stderr, "Invalid timeout value\n");
570 if (optind == argc - 1) {
571 name = argv[optind ++];
572 if (strlen(name) >= LST_NAME_SIZE) {
573 fprintf(stderr, "Name size is limited to %d\n",
578 } else if (optind == argc) {
579 char user[LST_NAME_SIZE];
580 char host[LST_NAME_SIZE];
581 struct passwd *pw = getpwuid(getuid());
584 snprintf(user, sizeof(user), "%d", (int)getuid());
586 snprintf(user, sizeof(user), "%s", pw->pw_name);
588 rc = gethostname(host, sizeof(host));
590 snprintf(host, sizeof(host), "unknown_host");
592 snprintf(buf, LST_NAME_SIZE, "%s@%s", user, host);
596 lst_print_usage(argv[0]);
600 rc = lst_new_session_ioctl(name, timeout, force, &session_id);
603 lst_print_error("session", "Failed to create session: %s\n",
608 fprintf(stdout, "SESSION: %s TIMEOUT: %d FORCE: %s\n",
609 name, timeout, force ? "Yes": "No");
615 lst_session_info_ioctl(char *name, int len, int *key,
616 lst_sid_t *sid, lstcon_ndlist_ent_t *ndinfo)
618 lstio_session_info_args_t args = {0};
620 args.lstio_ses_idp = sid;
621 args.lstio_ses_keyp = key;
622 args.lstio_ses_ndinfo = ndinfo;
623 args.lstio_ses_nmlen = len;
624 args.lstio_ses_namep = name;
626 return lst_ioctl(LSTIO_SESSION_INFO, &args, sizeof(args));
630 jt_lst_show_session(int argc, char **argv)
632 lstcon_ndlist_ent_t ndinfo;
634 char name[LST_NAME_SIZE];
638 rc = lst_session_info_ioctl(name, LST_NAME_SIZE, &key, &sid, &ndinfo);
641 lst_print_error("session", "Failed to show session: %s\n",
646 fprintf(stdout, "%s ID: "LPU64"@%s, KEY: %d NODES: %d\n",
647 name, sid.ses_stamp, libcfs_nid2str(sid.ses_nid),
648 key, ndinfo.nle_nnode);
654 lst_end_session_ioctl(void)
656 lstio_session_end_args_t args = {0};
658 args.lstio_ses_key = session_key;
659 return lst_ioctl (LSTIO_SESSION_END, &args, sizeof(args));
663 jt_lst_end_session(int argc, char **argv)
667 if (session_key == 0) {
669 "Can't find env LST_SESSION or value is not valid\n");
673 rc = lst_end_session_ioctl();
676 fprintf(stdout, "session is ended\n");
681 lst_print_error("session", "Failed to end session: %s\n",
686 if (trans_stat.trs_rpc_errno != 0) {
688 "[RPC] Failed to send %d session RPCs: %s\n",
689 lstcon_rpc_stat_failure(&trans_stat, 0),
690 strerror(trans_stat.trs_rpc_errno));
693 if (trans_stat.trs_fwk_errno != 0) {
695 "[FWK] Failed to end session on %d nodes: %s\n",
696 lstcon_sesop_stat_failure(&trans_stat, 0),
697 strerror(trans_stat.trs_fwk_errno));
704 lst_ping_ioctl(char *str, int type, int timeout,
705 int count, lnet_process_id_t *ids, cfs_list_t *head)
707 lstio_debug_args_t args = {0};
709 args.lstio_dbg_key = session_key;
710 args.lstio_dbg_type = type;
711 args.lstio_dbg_flags = 0;
712 args.lstio_dbg_timeout = timeout;
713 args.lstio_dbg_nmlen = (str == NULL) ? 0: strlen(str);
714 args.lstio_dbg_namep = str;
715 args.lstio_dbg_count = count;
716 args.lstio_dbg_idsp = ids;
717 args.lstio_dbg_resultp = head;
719 return lst_ioctl (LSTIO_DEBUG, &args, sizeof(args));
723 lst_get_node_count(int type, char *str, int *countp, lnet_process_id_t **idspp)
725 char buf[LST_NAME_SIZE];
726 lstcon_test_batch_ent_t ent;
727 lstcon_ndlist_ent_t *entp = &ent.tbe_cli_nle;
733 case LST_OPC_SESSION:
734 rc = lst_session_info_ioctl(buf, LST_NAME_SIZE,
738 case LST_OPC_BATCHSRV:
739 entp = &ent.tbe_srv_nle;
740 case LST_OPC_BATCHCLI:
741 rc = lst_info_batch_ioctl(str, 0, 0, &ent, NULL, NULL, NULL);
745 rc = lst_info_group_ioctl(str, entp, NULL, NULL, NULL);
749 rc = lst_parse_nids(str, &entp->nle_nnode, idspp) < 0 ? -1 : 0;
758 *countp = entp->nle_nnode;
764 jt_lst_ping(int argc, char **argv)
767 lnet_process_id_t *ids = NULL;
768 lstcon_rpc_ent_t *ent = NULL;
778 static struct option ping_opts[] =
780 {"session", no_argument, 0, 's' },
781 {"server", no_argument, 0, 'v' },
782 {"batch", required_argument, 0, 'b' },
783 {"group", required_argument, 0, 'g' },
784 {"nodes", required_argument, 0, 'n' },
785 {"timeout", required_argument, 0, 't' },
789 if (session_key == 0) {
791 "Can't find env LST_SESSION or value is not valid\n");
797 c = getopt_long(argc, argv, "g:b:n:t:sv",
805 type = LST_OPC_SESSION;
809 type = LST_OPC_GROUP;
814 type = LST_OPC_BATCHCLI;
819 type = LST_OPC_NODES;
824 timeout = atoi(optarg);
832 lst_print_usage(argv[0]);
837 if (type == 0 || timeout <= 0 || optind != argc) {
838 lst_print_usage(argv[0]);
842 if (type == LST_OPC_BATCHCLI && server)
843 type = LST_OPC_BATCHSRV;
845 rc = lst_get_node_count(type, str, &count, &ids);
847 fprintf(stderr, "Failed to get count of nodes from %s: %s\n",
848 (str == NULL) ? "session" : str, strerror(errno));
852 CFS_INIT_LIST_HEAD(&head);
854 rc = lst_alloc_rpcent(&head, count, LST_NAME_SIZE);
856 fprintf(stderr, "Out of memory\n");
861 fprintf(stdout, "Target %s is empty\n",
862 (str == NULL) ? "session" : str);
866 rc = lst_ping_ioctl(str, type, timeout, count, ids, &head);
867 if (rc == -1) { /* local failure */
868 lst_print_error("debug", "Failed to ping %s: %s\n",
869 (str == NULL) ? "session" : str,
875 /* ignore RPC errors and framwork errors */
876 cfs_list_for_each_entry_typed(ent, &head, lstcon_rpc_ent_t, rpe_link) {
877 fprintf(stdout, "\t%s: %s [session: %s id: %s]\n",
878 libcfs_id2str(ent->rpe_peer),
879 lst_node_state2str(ent->rpe_state),
880 (ent->rpe_state == LST_NODE_ACTIVE ||
881 ent->rpe_state == LST_NODE_BUSY)?
882 (ent->rpe_rpc_errno == 0 ?
883 &ent->rpe_payload[0] : "Unknown") :
884 "<NULL>", libcfs_nid2str(ent->rpe_sid.ses_nid));
888 lst_free_rpcent(&head);
898 lst_add_nodes_ioctl (char *name, int count, lnet_process_id_t *ids,
901 lstio_group_nodes_args_t args = {0};
903 args.lstio_grp_key = session_key;
904 args.lstio_grp_nmlen = strlen(name);
905 args.lstio_grp_namep = name;
906 args.lstio_grp_count = count;
907 args.lstio_grp_idsp = ids;
908 args.lstio_grp_resultp = resultp;
910 return lst_ioctl(LSTIO_NODES_ADD, &args, sizeof(args));
914 lst_add_group_ioctl (char *name)
916 lstio_group_add_args_t args = {0};
918 args.lstio_grp_key = session_key;
919 args.lstio_grp_nmlen = strlen(name);
920 args.lstio_grp_namep = name;
922 return lst_ioctl(LSTIO_GROUP_ADD, &args, sizeof(args));
926 jt_lst_add_group(int argc, char **argv)
929 lnet_process_id_t *ids;
935 if (session_key == 0) {
937 "Can't find env LST_SESSION or value is not valid\n");
942 lst_print_usage(argv[0]);
947 if (strlen(name) >= LST_NAME_SIZE) {
948 fprintf(stderr, "Name length is limited to %d\n",
953 rc = lst_add_group_ioctl(name);
955 lst_print_error("group", "Failed to add group %s: %s\n",
956 name, strerror(errno));
960 CFS_INIT_LIST_HEAD(&head);
962 for (i = 2; i < argc; i++) {
963 /* parse address list */
964 rc = lst_parse_nids(argv[i], &count, &ids);
966 fprintf(stderr, "Ignore invalid id list %s\n",
974 rc = lst_alloc_rpcent(&head, count, 0);
976 fprintf(stderr, "Out of memory\n");
980 rc = lst_add_nodes_ioctl(name, count, ids, &head);
985 lst_free_rpcent(&head);
986 fprintf(stderr, "%s are added to session\n", argv[i]);
991 lst_free_rpcent(&head);
992 lst_print_error("group", "Failed to add nodes %s: %s\n",
993 argv[i], strerror(errno));
997 lst_print_transerr(&head, "create session");
998 lst_free_rpcent(&head);
1005 lst_del_group_ioctl (char *name)
1007 lstio_group_del_args_t args = {0};
1009 args.lstio_grp_key = session_key;
1010 args.lstio_grp_nmlen = strlen(name);
1011 args.lstio_grp_namep = name;
1013 return lst_ioctl(LSTIO_GROUP_DEL, &args, sizeof(args));
1017 jt_lst_del_group(int argc, char **argv)
1021 if (session_key == 0) {
1023 "Can't find env LST_SESSION or value is not valid\n");
1028 lst_print_usage(argv[0]);
1032 rc = lst_del_group_ioctl(argv[1]);
1034 fprintf(stdout, "Group is deleted\n");
1039 lst_print_error("group", "Failed to delete group: %s\n",
1044 fprintf(stderr, "Group is deleted with some errors\n");
1046 if (trans_stat.trs_rpc_errno != 0) {
1047 fprintf(stderr, "[RPC] Failed to send %d end session RPCs: %s\n",
1048 lstcon_rpc_stat_failure(&trans_stat, 0),
1049 strerror(trans_stat.trs_rpc_errno));
1052 if (trans_stat.trs_fwk_errno != 0) {
1054 "[FWK] Failed to end session on %d nodes: %s\n",
1055 lstcon_sesop_stat_failure(&trans_stat, 0),
1056 strerror(trans_stat.trs_fwk_errno));
1063 lst_update_group_ioctl(int opc, char *name, int clean, int count,
1064 lnet_process_id_t *ids, cfs_list_t *resultp)
1066 lstio_group_update_args_t args = {0};
1068 args.lstio_grp_key = session_key;
1069 args.lstio_grp_opc = opc;
1070 args.lstio_grp_args = clean;
1071 args.lstio_grp_nmlen = strlen(name);
1072 args.lstio_grp_namep = name;
1073 args.lstio_grp_count = count;
1074 args.lstio_grp_idsp = ids;
1075 args.lstio_grp_resultp = resultp;
1077 return lst_ioctl(LSTIO_GROUP_UPDATE, &args, sizeof(args));
1081 jt_lst_update_group(int argc, char **argv)
1084 lnet_process_id_t *ids = NULL;
1094 static struct option update_group_opts[] =
1096 {"refresh", no_argument, 0, 'f' },
1097 {"clean", required_argument, 0, 'c' },
1098 {"remove", required_argument, 0, 'r' },
1102 if (session_key == 0) {
1104 "Can't find env LST_SESSION or value is not valid\n");
1109 c = getopt_long(argc, argv, "fc:r:",
1110 update_group_opts, &optidx);
1112 /* Detect the end of the options. */
1119 lst_print_usage(argv[0]);
1122 opc = LST_GROUP_REFRESH;
1127 lst_print_usage(argv[0]);
1130 opc = LST_GROUP_RMND;
1135 clean = lst_node_str2state(optarg);
1136 if (opc != 0 || clean <= 0) {
1137 lst_print_usage(argv[0]);
1140 opc = LST_GROUP_CLEAN;
1144 lst_print_usage(argv[0]);
1149 /* no OPC or group is specified */
1150 if (opc == 0 || optind != argc - 1) {
1151 lst_print_usage(argv[0]);
1157 CFS_INIT_LIST_HEAD(&head);
1159 if (opc == LST_GROUP_RMND || opc == LST_GROUP_REFRESH) {
1160 rc = lst_get_node_count(opc == LST_GROUP_RMND ? LST_OPC_NODES :
1162 opc == LST_GROUP_RMND ? str : grp,
1166 fprintf(stderr, "Can't get count of nodes from %s: %s\n",
1167 opc == LST_GROUP_RMND ? str : grp,
1172 rc = lst_alloc_rpcent(&head, count, 0);
1174 fprintf(stderr, "Out of memory\n");
1181 rc = lst_update_group_ioctl(opc, grp, clean, count, ids, &head);
1187 lst_free_rpcent(&head);
1192 lst_free_rpcent(&head);
1193 lst_print_error("group", "Failed to update group: %s\n",
1198 lst_print_transerr(&head, "Updating group");
1200 lst_free_rpcent(&head);
1206 lst_list_group_ioctl(int len, char *name, int idx)
1208 lstio_group_list_args_t args = {0};
1210 args.lstio_grp_key = session_key;
1211 args.lstio_grp_idx = idx;
1212 args.lstio_grp_nmlen = len;
1213 args.lstio_grp_namep = name;
1215 return lst_ioctl(LSTIO_GROUP_LIST, &args, sizeof(args));
1219 lst_info_group_ioctl(char *name, lstcon_ndlist_ent_t *gent,
1220 int *idx, int *count, lstcon_node_ent_t *dents)
1222 lstio_group_info_args_t args = {0};
1224 args.lstio_grp_key = session_key;
1225 args.lstio_grp_nmlen = strlen(name);
1226 args.lstio_grp_namep = name;
1227 args.lstio_grp_entp = gent;
1228 args.lstio_grp_idxp = idx;
1229 args.lstio_grp_ndentp = count;
1230 args.lstio_grp_dentsp = dents;
1232 return lst_ioctl(LSTIO_GROUP_INFO, &args, sizeof(args));
1236 lst_list_group_all(void)
1238 char name[LST_NAME_SIZE];
1242 /* no group is specified, list name of all groups */
1243 for (i = 0; ; i++) {
1244 rc = lst_list_group_ioctl(LST_NAME_SIZE, name, i);
1246 fprintf(stdout, "%d) %s\n", i + 1, name);
1250 if (errno == ENOENT)
1253 lst_print_error("group", "Failed to list group: %s\n",
1258 fprintf(stdout, "Total %d groups\n", i);
1263 #define LST_NODES_TITLE "\tACTIVE\tBUSY\tDOWN\tUNKNOWN\tTOTAL\n"
1266 jt_lst_list_group(int argc, char **argv)
1268 lstcon_ndlist_ent_t gent;
1269 lstcon_node_ent_t *dents;
1284 static struct option list_group_opts[] =
1286 {"active", no_argument, 0, 'a' },
1287 {"busy", no_argument, 0, 'b' },
1288 {"down", no_argument, 0, 'd' },
1289 {"unknown", no_argument, 0, 'u' },
1290 {"all", no_argument, 0, 'l' },
1294 if (session_key == 0) {
1296 "Can't find env LST_SESSION or value is not valid\n");
1301 c = getopt_long(argc, argv, "abdul",
1302 list_group_opts, &optidx);
1309 verbose = active = 1;
1321 verbose = unknown = 1;
1328 lst_print_usage(argv[0]);
1333 if (optind == argc) {
1334 /* no group is specified, list name of all groups */
1335 rc = lst_list_group_all();
1341 fprintf(stdout, LST_NODES_TITLE);
1343 /* list nodes in specified groups */
1344 for (i = optind; i < argc; i++) {
1345 rc = lst_info_group_ioctl(argv[i], &gent, NULL, NULL, NULL);
1347 if (errno == ENOENT) {
1352 lst_print_error("group", "Failed to list group\n",
1358 fprintf(stdout, "\t%d\t%d\t%d\t%d\t%d\t%s\n",
1359 gent.nle_nactive, gent.nle_nbusy,
1360 gent.nle_ndown, gent.nle_nunknown,
1361 gent.nle_nnode, argv[i]);
1365 fprintf(stdout, "Group [ %s ]\n", argv[i]);
1367 if (gent.nle_nnode == 0) {
1368 fprintf(stdout, "No nodes found [ %s ]\n", argv[i]);
1372 count = gent.nle_nnode;
1374 dents = malloc(count * sizeof(lstcon_node_ent_t));
1375 if (dents == NULL) {
1376 fprintf(stderr, "Failed to malloc: %s\n",
1382 rc = lst_info_group_ioctl(argv[i], &gent, &index, &count, dents);
1384 lst_print_error("group", "Failed to list group: %s\n",
1390 for (j = 0, c = 0; j < count; j++) {
1392 ((active && dents[j].nde_state == LST_NODE_ACTIVE) ||
1393 (busy && dents[j].nde_state == LST_NODE_BUSY) ||
1394 (down && dents[j].nde_state == LST_NODE_DOWN) ||
1395 (unknown && dents[j].nde_state == LST_NODE_UNKNOWN))) {
1397 fprintf(stdout, "\t%s: %s\n",
1398 libcfs_id2str(dents[j].nde_id),
1399 lst_node_state2str(dents[j].nde_state));
1404 fprintf(stdout, "Total %d nodes [ %s ]\n", c, argv[i]);
1413 lst_stat_ioctl (char *name, int count, lnet_process_id_t *idsp,
1414 int timeout, cfs_list_t *resultp)
1416 lstio_stat_args_t args = {0};
1418 args.lstio_sta_key = session_key;
1419 args.lstio_sta_timeout = timeout;
1420 args.lstio_sta_nmlen = strlen(name);
1421 args.lstio_sta_namep = name;
1422 args.lstio_sta_count = count;
1423 args.lstio_sta_idsp = idsp;
1424 args.lstio_sta_resultp = resultp;
1426 return lst_ioctl (LSTIO_STAT_QUERY, &args, sizeof(args));
1430 cfs_list_t srp_link;
1433 lnet_process_id_t *srp_ids;
1434 cfs_list_t srp_result[2];
1435 } lst_stat_req_param_t;
1438 lst_stat_req_param_free(lst_stat_req_param_t *srp)
1442 for (i = 0; i < 2; i++)
1443 lst_free_rpcent(&srp->srp_result[i]);
1445 if (srp->srp_ids != NULL)
1452 lst_stat_req_param_alloc(char *name, lst_stat_req_param_t **srpp, int save_old)
1454 lst_stat_req_param_t *srp = NULL;
1455 int count = save_old ? 2 : 1;
1459 srp = malloc(sizeof(*srp));
1463 memset(srp, 0, sizeof(*srp));
1464 CFS_INIT_LIST_HEAD(&srp->srp_result[0]);
1465 CFS_INIT_LIST_HEAD(&srp->srp_result[1]);
1467 rc = lst_get_node_count(LST_OPC_GROUP, name,
1468 &srp->srp_count, NULL);
1469 if (rc != 0 && errno == ENOENT) {
1470 rc = lst_get_node_count(LST_OPC_NODES, name,
1471 &srp->srp_count, &srp->srp_ids);
1476 "Failed to get count of nodes from %s: %s\n",
1477 name, strerror(errno));
1478 lst_stat_req_param_free(srp);
1483 srp->srp_name = name;
1485 for (i = 0; i < count; i++) {
1486 rc = lst_alloc_rpcent(&srp->srp_result[i], srp->srp_count,
1487 sizeof(sfw_counters_t) +
1488 sizeof(srpc_counters_t) +
1489 sizeof(lnet_counters_t));
1491 fprintf(stderr, "Out of memory\n");
1501 lst_stat_req_param_free(srp);
1509 } lst_srpc_stat_result;
1511 #define LST_LNET_AVG 0
1512 #define LST_LNET_MIN 1
1513 #define LST_LNET_MAX 2
1516 float lnet_avg_sndrate;
1517 float lnet_min_sndrate;
1518 float lnet_max_sndrate;
1519 float lnet_total_sndrate;
1521 float lnet_avg_rcvrate;
1522 float lnet_min_rcvrate;
1523 float lnet_max_rcvrate;
1524 float lnet_total_rcvrate;
1526 float lnet_avg_sndperf;
1527 float lnet_min_sndperf;
1528 float lnet_max_sndperf;
1529 float lnet_total_sndperf;
1531 float lnet_avg_rcvperf;
1532 float lnet_min_rcvperf;
1533 float lnet_max_rcvperf;
1534 float lnet_total_rcvperf;
1536 int lnet_stat_count;
1537 } lst_lnet_stat_result_t;
1539 lst_lnet_stat_result_t lnet_stat_result;
1542 lst_lnet_stat_value(int bw, int send, int off)
1546 p = bw ? &lnet_stat_result.lnet_avg_sndperf :
1547 &lnet_stat_result.lnet_avg_sndrate;
1558 lst_cal_lnet_stat(float delta, lnet_counters_t *lnet_new,
1559 lnet_counters_t *lnet_old)
1564 perf = (float)(lnet_new->send_length -
1565 lnet_old->send_length) / (1024 * 1024) / delta;
1566 lnet_stat_result.lnet_total_sndperf += perf;
1568 if (lnet_stat_result.lnet_min_sndperf > perf ||
1569 lnet_stat_result.lnet_min_sndperf == 0)
1570 lnet_stat_result.lnet_min_sndperf = perf;
1572 if (lnet_stat_result.lnet_max_sndperf < perf)
1573 lnet_stat_result.lnet_max_sndperf = perf;
1575 perf = (float)(lnet_new->recv_length -
1576 lnet_old->recv_length) / (1024 * 1024) / delta;
1577 lnet_stat_result.lnet_total_rcvperf += perf;
1579 if (lnet_stat_result.lnet_min_rcvperf > perf ||
1580 lnet_stat_result.lnet_min_rcvperf == 0)
1581 lnet_stat_result.lnet_min_rcvperf = perf;
1583 if (lnet_stat_result.lnet_max_rcvperf < perf)
1584 lnet_stat_result.lnet_max_rcvperf = perf;
1586 rate = (lnet_new->send_count - lnet_old->send_count) / delta;
1587 lnet_stat_result.lnet_total_sndrate += rate;
1589 if (lnet_stat_result.lnet_min_sndrate > rate ||
1590 lnet_stat_result.lnet_min_sndrate == 0)
1591 lnet_stat_result.lnet_min_sndrate = rate;
1593 if (lnet_stat_result.lnet_max_sndrate < rate)
1594 lnet_stat_result.lnet_max_sndrate = rate;
1596 rate = (lnet_new->recv_count - lnet_old->recv_count) / delta;
1597 lnet_stat_result.lnet_total_rcvrate += rate;
1599 if (lnet_stat_result.lnet_min_rcvrate > rate ||
1600 lnet_stat_result.lnet_min_rcvrate == 0)
1601 lnet_stat_result.lnet_min_rcvrate = rate;
1603 if (lnet_stat_result.lnet_max_rcvrate < rate)
1604 lnet_stat_result.lnet_max_rcvrate = rate;
1606 lnet_stat_result.lnet_stat_count ++;
1608 lnet_stat_result.lnet_avg_sndrate = lnet_stat_result.lnet_total_sndrate /
1609 lnet_stat_result.lnet_stat_count;
1610 lnet_stat_result.lnet_avg_rcvrate = lnet_stat_result.lnet_total_rcvrate /
1611 lnet_stat_result.lnet_stat_count;
1613 lnet_stat_result.lnet_avg_sndperf = lnet_stat_result.lnet_total_sndperf /
1614 lnet_stat_result.lnet_stat_count;
1615 lnet_stat_result.lnet_avg_rcvperf = lnet_stat_result.lnet_total_rcvperf /
1616 lnet_stat_result.lnet_stat_count;
1621 lst_print_lnet_stat(char *name, int bwrt, int rdwr, int type)
1630 if (lnet_stat_result.lnet_stat_count == 0)
1633 if (bwrt == 1) /* bw only */
1636 if (bwrt == 2) /* rates only */
1639 if (rdwr == 1) /* recv only */
1642 if (rdwr == 2) /* send only */
1645 for (i = start1; i <= end1; i++) {
1646 fprintf(stdout, "[LNet %s of %s]\n",
1647 i == 0 ? "Rates" : "Bandwidth", name);
1649 for (j = start2; j <= end2; j++) {
1650 fprintf(stdout, "[%c] ", j == 0 ? 'R' : 'W');
1652 if ((type & 1) != 0) {
1653 fprintf(stdout, i == 0 ? "Avg: %-8.0f RPC/s " :
1654 "Avg: %-8.2f MB/s ",
1655 lst_lnet_stat_value(i, j, 0));
1658 if ((type & 2) != 0) {
1659 fprintf(stdout, i == 0 ? "Min: %-8.0f RPC/s " :
1660 "Min: %-8.2f MB/s ",
1661 lst_lnet_stat_value(i, j, 1));
1664 if ((type & 4) != 0) {
1665 fprintf(stdout, i == 0 ? "Max: %-8.0f RPC/s" :
1667 lst_lnet_stat_value(i, j, 2));
1670 fprintf(stdout, "\n");
1676 lst_print_stat(char *name, cfs_list_t *resultp,
1677 int idx, int lnet, int bwrt, int rdwr, int type)
1680 lstcon_rpc_ent_t *new;
1681 lstcon_rpc_ent_t *old;
1682 sfw_counters_t *sfwk_new;
1683 sfw_counters_t *sfwk_old;
1684 srpc_counters_t *srpc_new;
1685 srpc_counters_t *srpc_old;
1686 lnet_counters_t *lnet_new;
1687 lnet_counters_t *lnet_old;
1691 CFS_INIT_LIST_HEAD(&tmp[0]);
1692 CFS_INIT_LIST_HEAD(&tmp[1]);
1694 memset(&lnet_stat_result, 0, sizeof(lnet_stat_result));
1696 while (!cfs_list_empty(&resultp[idx])) {
1697 if (cfs_list_empty(&resultp[1 - idx])) {
1698 fprintf(stderr, "Group is changed, re-run stat\n");
1702 new = cfs_list_entry(resultp[idx].next, lstcon_rpc_ent_t,
1704 old = cfs_list_entry(resultp[1 - idx].next, lstcon_rpc_ent_t,
1707 /* first time get stats result, can't calculate diff */
1708 if (new->rpe_peer.nid == LNET_NID_ANY)
1711 if (new->rpe_peer.nid != old->rpe_peer.nid ||
1712 new->rpe_peer.pid != old->rpe_peer.pid) {
1713 /* Something wrong. i.e, somebody change the group */
1717 cfs_list_del(&new->rpe_link);
1718 cfs_list_add_tail(&new->rpe_link, &tmp[idx]);
1720 cfs_list_del(&old->rpe_link);
1721 cfs_list_add_tail(&old->rpe_link, &tmp[1 - idx]);
1723 if (new->rpe_rpc_errno != 0 || new->rpe_fwk_errno != 0 ||
1724 old->rpe_rpc_errno != 0 || old->rpe_fwk_errno != 0) {
1729 sfwk_new = (sfw_counters_t *)&new->rpe_payload[0];
1730 sfwk_old = (sfw_counters_t *)&old->rpe_payload[0];
1732 srpc_new = (srpc_counters_t *)((char *)sfwk_new + sizeof(*sfwk_new));
1733 srpc_old = (srpc_counters_t *)((char *)sfwk_old + sizeof(*sfwk_old));
1735 lnet_new = (lnet_counters_t *)((char *)srpc_new + sizeof(*srpc_new));
1736 lnet_old = (lnet_counters_t *)((char *)srpc_old + sizeof(*srpc_old));
1738 /* use the timestamp from the remote node, not our rpe_stamp
1739 * from when we copied up the data out of the kernel */
1741 delta = (float) (sfwk_new->running_ms -
1742 sfwk_old->running_ms) / 1000;
1744 if (!lnet) /* TODO */
1747 lst_cal_lnet_stat(delta, lnet_new, lnet_old);
1750 cfs_list_splice(&tmp[idx], &resultp[idx]);
1751 cfs_list_splice(&tmp[1 - idx], &resultp[1 - idx]);
1754 fprintf(stdout, "Failed to stat on %d nodes\n", errcount);
1756 if (!lnet) /* TODO */
1759 lst_print_lnet_stat(name, bwrt, rdwr, type);
1763 jt_lst_stat(int argc, char **argv)
1766 lst_stat_req_param_t *srp;
1769 int timeout = 5; /* default timeout, 5 sec */
1770 int delay = 5; /* default delay, 5 sec */
1771 int count = -1; /* run forever */
1772 int lnet = 1; /* lnet stat by default */
1780 static struct option stat_opts[] =
1782 {"timeout", required_argument, 0, 't' },
1783 {"delay" , required_argument, 0, 'd' },
1784 {"count" , required_argument, 0, 'o' },
1785 {"lnet" , no_argument, 0, 'l' },
1786 {"rpc" , no_argument, 0, 'c' },
1787 {"bw" , no_argument, 0, 'b' },
1788 {"rate" , no_argument, 0, 'a' },
1789 {"read" , no_argument, 0, 'r' },
1790 {"write" , no_argument, 0, 'w' },
1791 {"avg" , no_argument, 0, 'g' },
1792 {"min" , no_argument, 0, 'n' },
1793 {"max" , no_argument, 0, 'x' },
1797 if (session_key == 0) {
1799 "Can't find env LST_SESSION or value is not valid\n");
1804 c = getopt_long(argc, argv, "t:d:lcbarwgnx", stat_opts, &optidx);
1811 timeout = atoi(optarg);
1814 delay = atoi(optarg);
1817 count = atoi(optarg);
1859 lst_print_usage(argv[0]);
1864 if (optind == argc) {
1865 lst_print_usage(argv[0]);
1869 if (timeout <= 0 || delay <= 0) {
1870 fprintf(stderr, "Invalid timeout or delay value\n");
1875 fprintf(stderr, "Invalid count value\n");
1879 /* extra count to get first data point */
1883 CFS_INIT_LIST_HEAD(&head);
1885 while (optind < argc) {
1886 rc = lst_stat_req_param_alloc(argv[optind++], &srp, 1);
1890 cfs_list_add_tail(&srp->srp_link, &head);
1894 time_t now = time(NULL);
1896 if (now - last < delay) {
1897 sleep(delay - now + last);
1903 cfs_list_for_each_entry_typed(srp, &head, lst_stat_req_param_t,
1905 rc = lst_stat_ioctl(srp->srp_name,
1906 srp->srp_count, srp->srp_ids,
1907 timeout, &srp->srp_result[idx]);
1909 lst_print_error("stat", "Failed to stat %s: %s\n",
1910 srp->srp_name, strerror(errno));
1914 lst_print_stat(srp->srp_name, srp->srp_result,
1915 idx, lnet, bwrt, rdwr, type);
1917 lst_reset_rpcent(&srp->srp_result[1 - idx]);
1924 } while (count == -1 || count > 0);
1927 while (!cfs_list_empty(&head)) {
1928 srp = cfs_list_entry(head.next, lst_stat_req_param_t, srp_link);
1930 cfs_list_del(&srp->srp_link);
1931 lst_stat_req_param_free(srp);
1938 jt_lst_show_error(int argc, char **argv)
1941 lst_stat_req_param_t *srp;
1942 lstcon_rpc_ent_t *ent;
1943 sfw_counters_t *sfwk;
1944 srpc_counters_t *srpc;
1945 lnet_counters_t *lnet;
1952 static struct option show_error_opts[] =
1954 {"session", no_argument, 0, 's' },
1958 if (session_key == 0) {
1960 "Can't find env LST_SESSION or value is not valid\n");
1965 c = getopt_long(argc, argv, "s", show_error_opts, &optidx);
1976 lst_print_usage(argv[0]);
1981 if (optind == argc) {
1982 lst_print_usage(argv[0]);
1986 CFS_INIT_LIST_HEAD(&head);
1988 while (optind < argc) {
1989 rc = lst_stat_req_param_alloc(argv[optind++], &srp, 0);
1993 cfs_list_add_tail(&srp->srp_link, &head);
1996 cfs_list_for_each_entry_typed(srp, &head, lst_stat_req_param_t,
1998 rc = lst_stat_ioctl(srp->srp_name, srp->srp_count,
1999 srp->srp_ids, 10, &srp->srp_result[0]);
2002 lst_print_error(srp->srp_name, "Failed to show errors of %s: %s\n",
2003 srp->srp_name, strerror(errno));
2007 fprintf(stdout, "%s:\n", srp->srp_name);
2011 cfs_list_for_each_entry_typed(ent, &srp->srp_result[0],
2012 lstcon_rpc_ent_t, rpe_link) {
2013 if (ent->rpe_rpc_errno != 0) {
2015 fprintf(stderr, "RPC failure, can't show error on %s\n",
2016 libcfs_id2str(ent->rpe_peer));
2020 if (ent->rpe_fwk_errno != 0) {
2022 fprintf(stderr, "Framework failure, can't show error on %s\n",
2023 libcfs_id2str(ent->rpe_peer));
2027 sfwk = (sfw_counters_t *)&ent->rpe_payload[0];
2028 srpc = (srpc_counters_t *)((char *)sfwk + sizeof(*sfwk));
2029 lnet = (lnet_counters_t *)((char *)srpc + sizeof(*srpc));
2031 if (srpc->errors == 0 &&
2032 sfwk->brw_errors == 0 && sfwk->ping_errors == 0)
2036 sfwk->brw_errors == 0 && sfwk->ping_errors == 0)
2041 fprintf(stderr, "%s: [Session %d brw errors, %d ping errors]%c",
2042 libcfs_id2str(ent->rpe_peer),
2043 sfwk->brw_errors, sfwk->ping_errors,
2044 show_rpc ? ' ' : '\n');
2049 fprintf(stderr, "[RPC: %d errors, %d dropped, %d expired]\n",
2050 srpc->errors, srpc->rpcs_dropped, srpc->rpcs_expired);
2053 fprintf(stdout, "Total %d error nodes in %s\n", ecount, srp->srp_name);
2056 while (!cfs_list_empty(&head)) {
2057 srp = cfs_list_entry(head.next, lst_stat_req_param_t, srp_link);
2059 cfs_list_del(&srp->srp_link);
2060 lst_stat_req_param_free(srp);
2067 lst_add_batch_ioctl (char *name)
2069 lstio_batch_add_args_t args = {0};
2071 args.lstio_bat_key = session_key;
2072 args.lstio_bat_nmlen = strlen(name);
2073 args.lstio_bat_namep = name;
2075 return lst_ioctl (LSTIO_BATCH_ADD, &args, sizeof(args));
2079 jt_lst_add_batch(int argc, char **argv)
2084 if (session_key == 0) {
2086 "Can't find env LST_SESSION or value is not valid\n");
2091 lst_print_usage(argv[0]);
2096 if (strlen(name) >= LST_NAME_SIZE) {
2097 fprintf(stderr, "Name length is limited to %d\n",
2102 rc = lst_add_batch_ioctl(name);
2106 lst_print_error("batch", "Failed to create batch: %s\n",
2113 lst_start_batch_ioctl (char *name, int timeout, cfs_list_t *resultp)
2115 lstio_batch_run_args_t args = {0};
2117 args.lstio_bat_key = session_key;
2118 args.lstio_bat_timeout = timeout;
2119 args.lstio_bat_nmlen = strlen(name);
2120 args.lstio_bat_namep = name;
2121 args.lstio_bat_resultp = resultp;
2123 return lst_ioctl(LSTIO_BATCH_START, &args, sizeof(args));
2127 jt_lst_start_batch(int argc, char **argv)
2137 static struct option start_batch_opts[] =
2139 {"timeout", required_argument, 0, 't' },
2143 if (session_key == 0) {
2145 "Can't find env LST_SESSION or value is not valid\n");
2150 c = getopt_long(argc, argv, "t:",
2151 start_batch_opts, &optidx);
2153 /* Detect the end of the options. */
2159 timeout = atoi(optarg);
2162 lst_print_usage(argv[0]);
2167 if (optind == argc) {
2168 batch = LST_DEFAULT_BATCH;
2170 } else if (optind == argc - 1) {
2171 batch = argv[optind];
2174 lst_print_usage(argv[0]);
2178 rc = lst_get_node_count(LST_OPC_BATCHCLI, batch, &count, NULL);
2180 fprintf(stderr, "Failed to get count of nodes from %s: %s\n",
2181 batch, strerror(errno));
2185 CFS_INIT_LIST_HEAD(&head);
2187 rc = lst_alloc_rpcent(&head, count, 0);
2189 fprintf(stderr, "Out of memory\n");
2193 rc = lst_start_batch_ioctl(batch, timeout, &head);
2196 fprintf(stdout, "%s is running now\n", batch);
2197 lst_free_rpcent(&head);
2202 lst_print_error("batch", "Failed to start batch: %s\n",
2204 lst_free_rpcent(&head);
2208 lst_print_transerr(&head, "Run batch");
2210 lst_free_rpcent(&head);
2216 lst_stop_batch_ioctl(char *name, int force, cfs_list_t *resultp)
2218 lstio_batch_stop_args_t args = {0};
2220 args.lstio_bat_key = session_key;
2221 args.lstio_bat_force = force;
2222 args.lstio_bat_nmlen = strlen(name);
2223 args.lstio_bat_namep = name;
2224 args.lstio_bat_resultp = resultp;
2226 return lst_ioctl(LSTIO_BATCH_STOP, &args, sizeof(args));
2230 jt_lst_stop_batch(int argc, char **argv)
2240 static struct option stop_batch_opts[] =
2242 {"force", no_argument, 0, 'f' },
2246 if (session_key == 0) {
2248 "Can't find env LST_SESSION or value is not valid\n");
2253 c = getopt_long(argc, argv, "f",
2254 stop_batch_opts, &optidx);
2256 /* Detect the end of the options. */
2265 lst_print_usage(argv[0]);
2270 if (optind == argc) {
2271 batch = LST_DEFAULT_BATCH;
2273 } else if (optind == argc - 1) {
2274 batch = argv[optind];
2277 lst_print_usage(argv[0]);
2281 rc = lst_get_node_count(LST_OPC_BATCHCLI, batch, &count, NULL);
2283 fprintf(stderr, "Failed to get count of nodes from %s: %s\n",
2284 batch, strerror(errno));
2288 CFS_INIT_LIST_HEAD(&head);
2290 rc = lst_alloc_rpcent(&head, count, 0);
2292 fprintf(stderr, "Out of memory\n");
2296 rc = lst_stop_batch_ioctl(batch, force, &head);
2301 lst_reset_rpcent(&head);
2303 rc = lst_query_batch_ioctl(batch, 0, 0, 30, &head);
2307 if (lstcon_tsbqry_stat_run(&trans_stat, 0) == 0 &&
2308 lstcon_tsbqry_stat_failure(&trans_stat, 0) == 0)
2311 fprintf(stdout, "%d batch in stopping\n",
2312 lstcon_tsbqry_stat_run(&trans_stat, 0));
2316 fprintf(stdout, "Batch is stopped\n");
2317 lst_free_rpcent(&head);
2322 lst_print_error("batch", "Failed to stop batch: %s\n",
2324 lst_free_rpcent(&head);
2328 lst_print_transerr(&head, "stop batch");
2330 lst_free_rpcent(&head);
2336 lst_list_batch_ioctl(int len, char *name, int index)
2338 lstio_batch_list_args_t args = {0};
2340 args.lstio_bat_key = session_key;
2341 args.lstio_bat_idx = index;
2342 args.lstio_bat_nmlen = len;
2343 args.lstio_bat_namep = name;
2345 return lst_ioctl(LSTIO_BATCH_LIST, &args, sizeof(args));
2349 lst_info_batch_ioctl(char *batch, int test, int server,
2350 lstcon_test_batch_ent_t *entp, int *idxp,
2351 int *ndentp, lstcon_node_ent_t *dentsp)
2353 lstio_batch_info_args_t args = {0};
2355 args.lstio_bat_key = session_key;
2356 args.lstio_bat_nmlen = strlen(batch);
2357 args.lstio_bat_namep = batch;
2358 args.lstio_bat_server = server;
2359 args.lstio_bat_testidx = test;
2360 args.lstio_bat_entp = entp;
2361 args.lstio_bat_idxp = idxp;
2362 args.lstio_bat_ndentp = ndentp;
2363 args.lstio_bat_dentsp = dentsp;
2365 return lst_ioctl(LSTIO_BATCH_INFO, &args, sizeof(args));
2369 lst_list_batch_all(void)
2371 char name[LST_NAME_SIZE];
2375 for (i = 0; ; i++) {
2376 rc = lst_list_batch_ioctl(LST_NAME_SIZE, name, i);
2378 fprintf(stdout, "%d) %s\n", i + 1, name);
2382 if (errno == ENOENT)
2385 lst_print_error("batch", "Failed to list batch: %s\n",
2390 fprintf(stdout, "Total %d batches\n", i);
2396 lst_list_tsb_nodes(char *batch, int test, int server,
2397 int count, int active, int invalid)
2399 lstcon_node_ent_t *dents;
2408 /* verbose list, show nodes in batch or test */
2409 dents = malloc(count * sizeof(lstcon_node_ent_t));
2410 if (dents == NULL) {
2411 fprintf(stdout, "Can't allocate memory\n");
2415 rc = lst_info_batch_ioctl(batch, test, server,
2416 NULL, &index, &count, dents);
2419 lst_print_error((test > 0) ? "test" : "batch",
2420 (test > 0) ? "Failed to query test: %s\n" :
2421 "Failed to query batch: %s\n",
2426 for (i = 0, c = 0; i < count; i++) {
2427 if ((!active && dents[i].nde_state == LST_NODE_ACTIVE) ||
2428 (!invalid && (dents[i].nde_state == LST_NODE_BUSY ||
2429 dents[i].nde_state == LST_NODE_DOWN ||
2430 dents[i].nde_state == LST_NODE_UNKNOWN)))
2433 fprintf(stdout, "\t%s: %s\n",
2434 libcfs_id2str(dents[i].nde_id),
2435 lst_node_state2str(dents[i].nde_state));
2439 fprintf(stdout, "Total %d nodes\n", c);
2446 jt_lst_list_batch(int argc, char **argv)
2448 lstcon_test_batch_ent_t ent;
2451 int verbose = 0; /* list nodes in batch or test */
2460 static struct option list_batch_opts[] =
2462 {"test", required_argument, 0, 't' },
2463 {"invalid", no_argument, 0, 'i' },
2464 {"active", no_argument, 0, 'a' },
2465 {"all", no_argument, 0, 'l' },
2466 {"server", no_argument, 0, 's' },
2470 if (session_key == 0) {
2472 "Can't find env LST_SESSION or value is not valid\n");
2477 c = getopt_long(argc, argv, "ailst:",
2478 list_batch_opts, &optidx);
2485 verbose = active = 1;
2488 verbose = invalid = 1;
2491 verbose = active = invalid = 1;
2497 test = atoi(optarg);
2501 lst_print_usage(argv[0]);
2506 if (optind == argc) {
2507 /* list all batches */
2508 rc = lst_list_batch_all();
2512 if (ntest == 1 && test <= 0) {
2513 fprintf(stderr, "Invalid test id, test id starts from 1\n");
2517 if (optind != argc - 1) {
2518 lst_print_usage(argv[0]);
2522 batch = argv[optind];
2525 /* show detail of specified batch or test */
2526 rc = lst_info_batch_ioctl(batch, test, server,
2527 &ent, NULL, NULL, NULL);
2529 lst_print_error((test > 0) ? "test" : "batch",
2530 (test > 0) ? "Failed to query test: %s\n" :
2531 "Failed to query batch: %s\n",
2537 /* list nodes in test or batch */
2538 rc = lst_list_tsb_nodes(batch, test, server,
2539 server ? ent.tbe_srv_nle.nle_nnode :
2540 ent.tbe_cli_nle.nle_nnode,
2545 /* only show number of hosts in batch or test */
2547 fprintf(stdout, "Batch: %s Tests: %d State: %d\n",
2548 batch, ent.u.tbe_batch.bae_ntest,
2549 ent.u.tbe_batch.bae_state);
2550 ntest = ent.u.tbe_batch.bae_ntest;
2551 test = 1; /* starting from test 1 */
2555 "\tTest %d(%s) (loop: %d, concurrency: %d)\n",
2556 test, lst_test_type2name(ent.u.tbe_test.tse_type),
2557 ent.u.tbe_test.tse_loop,
2558 ent.u.tbe_test.tse_concur);
2563 fprintf(stdout, LST_NODES_TITLE);
2564 fprintf(stdout, "client\t%d\t%d\t%d\t%d\t%d\n"
2565 "server\t%d\t%d\t%d\t%d\t%d\n",
2566 ent.tbe_cli_nle.nle_nactive,
2567 ent.tbe_cli_nle.nle_nbusy,
2568 ent.tbe_cli_nle.nle_ndown,
2569 ent.tbe_cli_nle.nle_nunknown,
2570 ent.tbe_cli_nle.nle_nnode,
2571 ent.tbe_srv_nle.nle_nactive,
2572 ent.tbe_srv_nle.nle_nbusy,
2573 ent.tbe_srv_nle.nle_ndown,
2574 ent.tbe_srv_nle.nle_nunknown,
2575 ent.tbe_srv_nle.nle_nnode);
2584 lst_query_batch_ioctl(char *batch, int test, int server,
2585 int timeout, cfs_list_t *head)
2587 lstio_batch_query_args_t args = {0};
2589 args.lstio_bat_key = session_key;
2590 args.lstio_bat_testidx = test;
2591 args.lstio_bat_client = !(server);
2592 args.lstio_bat_timeout = timeout;
2593 args.lstio_bat_nmlen = strlen(batch);
2594 args.lstio_bat_namep = batch;
2595 args.lstio_bat_resultp = head;
2597 return lst_ioctl(LSTIO_BATCH_QUERY, &args, sizeof(args));
2601 lst_print_tsb_verbose(cfs_list_t *head,
2602 int active, int idle, int error)
2604 lstcon_rpc_ent_t *ent;
2606 cfs_list_for_each_entry_typed(ent, head, lstcon_rpc_ent_t, rpe_link) {
2607 if (ent->rpe_priv[0] == 0 && active)
2610 if (ent->rpe_priv[0] != 0 && idle)
2613 if (ent->rpe_fwk_errno == 0 && error)
2616 fprintf(stdout, "%s [%s]: %s\n",
2617 libcfs_id2str(ent->rpe_peer),
2618 lst_node_state2str(ent->rpe_state),
2619 ent->rpe_rpc_errno != 0 ?
2620 strerror(ent->rpe_rpc_errno) :
2621 (ent->rpe_priv[0] > 0 ? "Running" : "Idle"));
2626 jt_lst_query_batch(int argc, char **argv)
2628 lstcon_test_batch_ent_t ent;
2635 int timeout = 5; /* default 5 seconds */
2636 int delay = 5; /* default 5 seconds */
2637 int loop = 1; /* default 1 loop */
2647 static struct option query_batch_opts[] =
2649 {"timeout", required_argument, 0, 'o' },
2650 {"delay", required_argument, 0, 'd' },
2651 {"loop", required_argument, 0, 'c' },
2652 {"test", required_argument, 0, 't' },
2653 {"server", no_argument, 0, 's' },
2654 {"active", no_argument, 0, 'a' },
2655 {"idle", no_argument, 0, 'i' },
2656 {"error", no_argument, 0, 'e' },
2657 {"all", no_argument, 0, 'l' },
2661 if (session_key == 0) {
2663 "Can't find env LST_SESSION or value is not valid\n");
2668 c = getopt_long(argc, argv, "o:d:c:t:saiel",
2669 query_batch_opts, &optidx);
2671 /* Detect the end of the options. */
2677 timeout = atoi(optarg);
2680 delay = atoi(optarg);
2683 loop = atoi(optarg);
2686 test = atoi(optarg);
2692 active = verbose = 1;
2698 error = verbose = 1;
2704 lst_print_usage(argv[0]);
2709 if (test < 0 || timeout <= 0 || delay <= 0 || loop <= 0) {
2710 lst_print_usage(argv[0]);
2714 if (optind == argc) {
2715 batch = LST_DEFAULT_BATCH;
2717 } else if (optind == argc - 1) {
2718 batch = argv[optind];
2721 lst_print_usage(argv[0]);
2726 CFS_INIT_LIST_HEAD(&head);
2729 rc = lst_info_batch_ioctl(batch, test, server,
2730 &ent, NULL, NULL, NULL);
2732 fprintf(stderr, "Failed to query %s [%d]: %s\n",
2733 batch, test, strerror(errno));
2737 count = server ? ent.tbe_srv_nle.nle_nnode :
2738 ent.tbe_cli_nle.nle_nnode;
2740 fprintf(stdout, "Batch or test is empty\n");
2745 rc = lst_alloc_rpcent(&head, count, 0);
2747 fprintf(stderr, "Out of memory\n");
2751 for (i = 0; i < loop; i++) {
2752 time_t now = time(NULL);
2754 if (now - last < delay) {
2755 sleep(delay - now + last);
2761 rc = lst_query_batch_ioctl(batch, test,
2762 server, timeout, &head);
2764 fprintf(stderr, "Failed to query batch: %s\n",
2771 lst_print_tsb_verbose(&head, active, idle, error);
2775 fprintf(stdout, "%s [%d] ", batch, test);
2777 if (lstcon_rpc_stat_failure(&trans_stat, 0) != 0) {
2778 fprintf(stdout, "%d of %d nodes are unknown, ",
2779 lstcon_rpc_stat_failure(&trans_stat, 0),
2780 lstcon_rpc_stat_total(&trans_stat, 0));
2783 if (lstcon_rpc_stat_failure(&trans_stat, 0) == 0 &&
2784 lstcon_tsbqry_stat_run(&trans_stat, 0) == 0 &&
2785 lstcon_tsbqry_stat_failure(&trans_stat, 0) == 0) {
2786 fprintf(stdout, "is stopped\n");
2790 if (lstcon_rpc_stat_failure(&trans_stat, 0) == 0 &&
2791 lstcon_tsbqry_stat_idle(&trans_stat, 0) == 0 &&
2792 lstcon_tsbqry_stat_failure(&trans_stat, 0) == 0) {
2793 fprintf(stdout, "is running\n");
2797 fprintf(stdout, "stopped: %d , running: %d, failed: %d\n",
2798 lstcon_tsbqry_stat_idle(&trans_stat, 0),
2799 lstcon_tsbqry_stat_run(&trans_stat, 0),
2800 lstcon_tsbqry_stat_failure(&trans_stat, 0));
2803 lst_free_rpcent(&head);
2809 lst_parse_distribute(char *dstr, int *dist, int *span)
2815 dstr = strchr(dstr, ':');
2819 *span = atoi(dstr + 1);
2827 lst_get_bulk_param(int argc, char **argv, lst_test_bulk_param_t *bulk)
2834 bulk->blk_size = 4096;
2835 bulk->blk_opc = LST_BRW_READ;
2836 bulk->blk_flags = LST_BRW_CHECK_NONE;
2839 if (strcasestr(argv[i], "check=") == argv[i] ||
2840 strcasestr(argv[i], "c=") == argv[i]) {
2841 tok = strchr(argv[i], '=') + 1;
2843 if (strcasecmp(tok, "full") == 0) {
2844 bulk->blk_flags = LST_BRW_CHECK_FULL;
2845 } else if (strcasecmp(tok, "simple") == 0) {
2846 bulk->blk_flags = LST_BRW_CHECK_SIMPLE;
2848 fprintf(stderr, "Unknow flag %s\n", tok);
2852 } else if (strcasestr(argv[i], "size=") == argv[i] ||
2853 strcasestr(argv[i], "s=") == argv[i]) {
2854 tok = strchr(argv[i], '=') + 1;
2856 bulk->blk_size = strtol(tok, &end, 0);
2857 if (bulk->blk_size <= 0) {
2858 fprintf(stderr, "Invalid size %s\n", tok);
2865 if (*end == 'k' || *end == 'K')
2866 bulk->blk_size *= 1024;
2867 else if (*end == 'm' || *end == 'M')
2868 bulk->blk_size *= 1024 * 1024;
2870 if (bulk->blk_size > CFS_PAGE_SIZE * LNET_MAX_IOV) {
2871 fprintf(stderr, "Size exceed limitation: %d bytes\n",
2876 } else if (strcasecmp(argv[i], "read") == 0 ||
2877 strcasecmp(argv[i], "r") == 0) {
2878 bulk->blk_opc = LST_BRW_READ;
2880 } else if (strcasecmp(argv[i], "write") == 0 ||
2881 strcasecmp(argv[i], "w") == 0) {
2882 bulk->blk_opc = LST_BRW_WRITE;
2885 fprintf(stderr, "Unknow parameter: %s\n", argv[i]);
2896 lst_get_test_param(char *test, int argc, char **argv, void **param, int *plen)
2898 lst_test_bulk_param_t *bulk = NULL;
2901 type = lst_test_name2type(test);
2903 fprintf(stderr, "Unknow test name %s\n", test);
2912 bulk = malloc(sizeof(*bulk));
2914 fprintf(stderr, "Out of memory\n");
2918 memset(bulk, 0, sizeof(*bulk));
2920 if (lst_get_bulk_param(argc, argv, bulk) != 0) {
2926 *plen = sizeof(*bulk);
2934 /* TODO: parse more parameter */
2939 lst_add_test_ioctl(char *batch, int type, int loop, int concur,
2940 int dist, int span, char *sgrp, char *dgrp,
2941 void *param, int plen, int *retp, cfs_list_t *resultp)
2943 lstio_test_args_t args = {0};
2945 args.lstio_tes_key = session_key;
2946 args.lstio_tes_bat_nmlen = strlen(batch);
2947 args.lstio_tes_bat_name = batch;
2948 args.lstio_tes_type = type;
2949 args.lstio_tes_oneside = 0;
2950 args.lstio_tes_loop = loop;
2951 args.lstio_tes_concur = concur;
2952 args.lstio_tes_dist = dist;
2953 args.lstio_tes_span = span;
2954 args.lstio_tes_sgrp_nmlen = strlen(sgrp);
2955 args.lstio_tes_sgrp_name = sgrp;
2956 args.lstio_tes_dgrp_nmlen = strlen(dgrp);
2957 args.lstio_tes_dgrp_name = dgrp;
2958 args.lstio_tes_param_len = plen;
2959 args.lstio_tes_param = param;
2960 args.lstio_tes_retp = retp;
2961 args.lstio_tes_resultp = resultp;
2963 return lst_ioctl(LSTIO_TEST_ADD, &args, sizeof(args));
2967 jt_lst_add_test(int argc, char **argv)
2989 static struct option add_test_opts[] =
2991 {"batch", required_argument, 0, 'b' },
2992 {"concurrency", required_argument, 0, 'c' },
2993 {"distribute", required_argument, 0, 'd' },
2994 {"from", required_argument, 0, 'f' },
2995 {"to", required_argument, 0, 't' },
2996 {"loop", required_argument, 0, 'l' },
3000 if (session_key == 0) {
3002 "Can't find env LST_SESSION or value is not valid\n");
3007 c = getopt_long(argc, argv, "b:c:d:f:l:t:",
3008 add_test_opts, &optidx);
3010 /* Detect the end of the options. */
3019 concur = atoi(optarg);
3028 loop = atoi(optarg);
3034 lst_print_usage(argv[0]);
3039 if (optind == argc || from == NULL || to == NULL) {
3040 lst_print_usage(argv[0]);
3044 if (concur <= 0 || concur > LST_MAX_CONCUR) {
3045 fprintf(stderr, "Invalid concurrency of test: %d\n", concur);
3050 batch = LST_DEFAULT_BATCH;
3053 rc = lst_parse_distribute(dstr, &dist, &span);
3055 fprintf(stderr, "Invalid distribution: %s\n", dstr);
3060 test = argv[optind++];
3065 type = lst_get_test_param(test, argc, argv, ¶m, &plen);
3067 fprintf(stderr, "Failed to add test (%s)\n", test);
3071 CFS_INIT_LIST_HEAD(&head);
3073 rc = lst_get_node_count(LST_OPC_GROUP, from, &fcount, NULL);
3075 fprintf(stderr, "Can't get count of nodes from %s: %s\n",
3076 from, strerror(errno));
3080 rc = lst_get_node_count(LST_OPC_GROUP, to, &tcount, NULL);
3082 fprintf(stderr, "Can't get count of nodes from %s: %s\n",
3083 to, strerror(errno));
3087 rc = lst_alloc_rpcent(&head, fcount > tcount ? fcount : tcount, 0);
3089 fprintf(stderr, "Out of memory\n");
3093 rc = lst_add_test_ioctl(batch, type, loop, concur,
3094 dist, span, from, to, param, plen, &ret, &head);
3097 fprintf(stdout, "Test was added successfully\n");
3099 fprintf(stdout, "Server group contains userland test "
3100 "nodes, old version of tcplnd can't accept "
3101 "connection request\n");
3108 lst_print_error("test", "Failed to add test: %s\n",
3113 lst_print_transerr(&head, "add test");
3115 lst_free_rpcent(&head);
3123 static command_t lst_cmdlist[] = {
3124 {"new_session", jt_lst_new_session, NULL,
3125 "Usage: lst new_session [--timeout TIME] [--force] [NAME]" },
3126 {"end_session", jt_lst_end_session, NULL,
3127 "Usage: lst end_session" },
3128 {"show_session", jt_lst_show_session, NULL,
3129 "Usage: lst show_session" },
3130 {"ping", jt_lst_ping , NULL,
3131 "Usage: lst ping [--group NAME] [--batch NAME] [--session] [--nodes IDS]" },
3132 {"add_group", jt_lst_add_group, NULL,
3133 "Usage: lst group NAME IDs [IDs]..." },
3134 {"del_group", jt_lst_del_group, NULL,
3135 "Usage: lst del_group NAME" },
3136 {"update_group", jt_lst_update_group, NULL,
3137 "Usage: lst update_group NAME [--clean] [--refresh] [--remove IDs]" },
3138 {"list_group", jt_lst_list_group, NULL,
3139 "Usage: lst list_group [--active] [--busy] [--down] [--unknown] GROUP ..." },
3140 {"stat", jt_lst_stat, NULL,
3141 "Usage: lst stat [--bw] [--rate] [--read] [--write] [--max] [--min] [--avg] "
3142 " [--timeout #] [--delay #] [--count #] GROUP [GROUP]" },
3143 {"show_error", jt_lst_show_error, NULL,
3144 "Usage: lst show_error NAME | IDS ..." },
3145 {"add_batch", jt_lst_add_batch, NULL,
3146 "Usage: lst add_batch NAME" },
3147 {"run", jt_lst_start_batch, NULL,
3148 "Usage: lst run [--timeout TIME] [NAME]" },
3149 {"stop", jt_lst_stop_batch, NULL,
3150 "Usage: lst stop [--force] BATCH_NAME" },
3151 {"list_batch", jt_lst_list_batch, NULL,
3152 "Usage: lst list_batch NAME [--test ID] [--server]" },
3153 {"query", jt_lst_query_batch, NULL,
3154 "Usage: lst query [--test ID] [--server] [--timeout TIME] NAME" },
3155 {"add_test", jt_lst_add_test, NULL,
3156 "Usage: lst add_test [--batch BATCH] [--loop #] [--concurrency #] "
3157 " [--distribute #:#] [--from GROUP] [--to GROUP] TEST..." },
3158 {"help", Parser_help, 0, "help" },
3163 lst_initialize(void)
3167 key = getenv("LST_SESSION");
3174 session_key = atoi(key);
3180 main(int argc, char **argv)
3186 rc = libcfs_arch_init();
3190 rc = lst_initialize();
3194 rc = ptl_initialize(argc, argv);
3198 Parser_init("lst > ", lst_cmdlist);
3201 rc = Parser_execarg(argc - 1, argv + 1, lst_cmdlist);
3208 libcfs_arch_cleanup();