1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lnet/selftest/framework.c
38 * Author: Isaac Huang <isaac@clusterfs.com>
39 * Author: Liang Zhen <liangzhen@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_LNET
46 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
48 int brw_inject_errors = 0;
49 CFS_MODULE_PARM(brw_inject_errors, "i", int, 0644,
50 "# data errors to inject randomly, zero by default");
52 static int session_timeout = 100;
53 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
54 "test session timeout in seconds (100 by default, 0 == never)");
56 static int rpc_timeout = 64;
57 CFS_MODULE_PARM(rpc_timeout, "i", int, 0644,
58 "rpc timeout in seconds (64 by default, 0 == never)");
60 #define SFW_TEST_CONCURRENCY 1792
61 #define SFW_EXTRA_TEST_BUFFERS 8 /* tolerate buggy peers with extra buffers */
63 #define sfw_test_buffers(tsi) ((tsi)->tsi_loop + SFW_EXTRA_TEST_BUFFERS)
65 #define sfw_unpack_id(id) \
67 __swab64s(&(id).nid); \
68 __swab32s(&(id).pid); \
71 #define sfw_unpack_sid(sid) \
73 __swab64s(&(sid).ses_nid); \
74 __swab64s(&(sid).ses_stamp); \
77 #define sfw_unpack_fw_counters(fc) \
79 __swab32s(&(fc).brw_errors); \
80 __swab32s(&(fc).ping_errors); \
81 __swab32s(&(fc).active_tests); \
82 __swab32s(&(fc).active_batches); \
83 __swab32s(&(fc).zombie_sessions); \
86 #define sfw_unpack_rpc_counters(rc) \
88 __swab32s(&(rc).errors); \
89 __swab32s(&(rc).rpcs_sent); \
90 __swab32s(&(rc).rpcs_rcvd); \
91 __swab32s(&(rc).rpcs_dropped); \
92 __swab32s(&(rc).rpcs_expired); \
93 __swab64s(&(rc).bulk_get); \
94 __swab64s(&(rc).bulk_put); \
97 #define sfw_unpack_lnet_counters(lc) \
99 __swab32s(&(lc).errors); \
100 __swab32s(&(lc).msgs_max); \
101 __swab32s(&(lc).msgs_alloc); \
102 __swab32s(&(lc).send_count); \
103 __swab32s(&(lc).recv_count); \
104 __swab32s(&(lc).drop_count); \
105 __swab32s(&(lc).route_count); \
106 __swab64s(&(lc).send_length); \
107 __swab64s(&(lc).recv_length); \
108 __swab64s(&(lc).drop_length); \
109 __swab64s(&(lc).route_length); \
112 #define sfw_test_active(t) (cfs_atomic_read(&(t)->tsi_nactive) != 0)
113 #define sfw_batch_active(b) (cfs_atomic_read(&(b)->bat_nactive) != 0)
115 struct smoketest_framework {
116 cfs_list_t fw_zombie_rpcs; /* RPCs to be recycled */
117 cfs_list_t fw_zombie_sessions; /* stopping sessions */
118 cfs_list_t fw_tests; /* registered test cases */
119 cfs_atomic_t fw_nzombies; /* # zombie sessions */
120 cfs_spinlock_t fw_lock; /* serialise */
121 sfw_session_t *fw_session; /* _the_ session */
122 int fw_shuttingdown; /* shutdown in progress */
123 srpc_server_rpc_t *fw_active_srpc; /* running RPC */
127 int sfw_stop_batch (sfw_batch_t *tsb, int force);
128 void sfw_destroy_session (sfw_session_t *sn);
130 static inline sfw_test_case_t *
131 sfw_find_test_case(int id)
133 sfw_test_case_t *tsc;
135 LASSERT (id <= SRPC_SERVICE_MAX_ID);
136 LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
138 cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
139 sfw_test_case_t, tsc_list) {
140 if (tsc->tsc_srv_service->sv_id == id)
148 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
150 sfw_test_case_t *tsc;
152 if (sfw_find_test_case(service->sv_id) != NULL) {
153 CERROR ("Failed to register test %s (%d)\n",
154 service->sv_name, service->sv_id);
158 LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
162 memset(tsc, 0, sizeof(sfw_test_case_t));
163 tsc->tsc_cli_ops = cliops;
164 tsc->tsc_srv_service = service;
166 cfs_list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
171 sfw_add_session_timer (void)
173 sfw_session_t *sn = sfw_data.fw_session;
174 stt_timer_t *timer = &sn->sn_timer;
176 LASSERT (!sfw_data.fw_shuttingdown);
178 if (sn == NULL || sn->sn_timeout == 0)
181 LASSERT (!sn->sn_timer_active);
183 sn->sn_timer_active = 1;
184 timer->stt_expires = cfs_time_add(sn->sn_timeout,
185 cfs_time_current_sec());
186 stt_add_timer(timer);
191 sfw_del_session_timer (void)
193 sfw_session_t *sn = sfw_data.fw_session;
195 if (sn == NULL || !sn->sn_timer_active)
198 LASSERT (sn->sn_timeout != 0);
200 if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
201 sn->sn_timer_active = 0;
206 /* Racing is impossible in single-threaded userland selftest */
209 return EBUSY; /* racing with sfw_session_expired() */
212 /* called with sfw_data.fw_lock held */
214 sfw_deactivate_session (void)
216 sfw_session_t *sn = sfw_data.fw_session;
219 sfw_test_case_t *tsc;
221 if (sn == NULL) return;
223 LASSERT (!sn->sn_timer_active);
225 sfw_data.fw_session = NULL;
226 cfs_atomic_inc(&sfw_data.fw_nzombies);
227 cfs_list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
229 cfs_spin_unlock(&sfw_data.fw_lock);
231 cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
232 sfw_test_case_t, tsc_list) {
233 srpc_abort_service(tsc->tsc_srv_service);
236 cfs_spin_lock(&sfw_data.fw_lock);
238 cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
239 sfw_batch_t, bat_list) {
240 if (sfw_batch_active(tsb)) {
242 sfw_stop_batch(tsb, 1);
247 return; /* wait for active batches to stop */
249 cfs_list_del_init(&sn->sn_list);
250 cfs_spin_unlock(&sfw_data.fw_lock);
252 sfw_destroy_session(sn);
254 cfs_spin_lock(&sfw_data.fw_lock);
261 sfw_session_removed (void)
263 return (sfw_data.fw_session == NULL) ? 1 : 0;
269 sfw_session_expired (void *data)
271 sfw_session_t *sn = data;
273 cfs_spin_lock(&sfw_data.fw_lock);
275 LASSERT (sn->sn_timer_active);
276 LASSERT (sn == sfw_data.fw_session);
278 CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
279 libcfs_nid2str(sn->sn_id.ses_nid),
280 sn->sn_id.ses_stamp, &sn->sn_name[0]);
282 sn->sn_timer_active = 0;
283 sfw_deactivate_session();
285 cfs_spin_unlock(&sfw_data.fw_lock);
290 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
292 stt_timer_t *timer = &sn->sn_timer;
294 memset(sn, 0, sizeof(sfw_session_t));
295 CFS_INIT_LIST_HEAD(&sn->sn_list);
296 CFS_INIT_LIST_HEAD(&sn->sn_batches);
297 cfs_atomic_set(&sn->sn_refcount, 1); /* +1 for caller */
298 cfs_atomic_set(&sn->sn_brw_errors, 0);
299 cfs_atomic_set(&sn->sn_ping_errors, 0);
300 strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
302 sn->sn_timer_active = 0;
304 sn->sn_timeout = session_timeout;
306 timer->stt_data = sn;
307 timer->stt_func = sfw_session_expired;
308 CFS_INIT_LIST_HEAD(&timer->stt_list);
311 /* completion handler for incoming framework RPCs */
313 sfw_server_rpc_done (srpc_server_rpc_t *rpc)
315 srpc_service_t *sv = rpc->srpc_service;
316 int status = rpc->srpc_status;
319 "Incoming framework RPC done: "
320 "service %s, peer %s, status %s:%d\n",
321 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
322 swi_state2str(rpc->srpc_wi.swi_state),
325 if (rpc->srpc_bulk != NULL)
331 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
333 LASSERT (rpc->crpc_bulk.bk_niov == 0);
334 LASSERT (cfs_list_empty(&rpc->crpc_list));
335 LASSERT (cfs_atomic_read(&rpc->crpc_refcount) == 0);
337 LASSERT (rpc->crpc_bulk.bk_pages == NULL);
341 "Outgoing framework RPC done: "
342 "service %d, peer %s, status %s:%d:%d\n",
343 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
344 swi_state2str(rpc->crpc_wi.swi_state),
345 rpc->crpc_aborted, rpc->crpc_status);
347 cfs_spin_lock(&sfw_data.fw_lock);
349 /* my callers must finish all RPCs before shutting me down */
350 LASSERT (!sfw_data.fw_shuttingdown);
351 cfs_list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
353 cfs_spin_unlock(&sfw_data.fw_lock);
358 sfw_find_batch (lst_bid_t bid)
360 sfw_session_t *sn = sfw_data.fw_session;
363 LASSERT (sn != NULL);
365 cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
366 sfw_batch_t, bat_list) {
367 if (bat->bat_id.bat_id == bid.bat_id)
375 sfw_bid2batch (lst_bid_t bid)
377 sfw_session_t *sn = sfw_data.fw_session;
380 LASSERT (sn != NULL);
382 bat = sfw_find_batch(bid);
386 LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
391 bat->bat_session = sn;
393 cfs_atomic_set(&bat->bat_nactive, 0);
394 CFS_INIT_LIST_HEAD(&bat->bat_tests);
396 cfs_list_add_tail(&bat->bat_list, &sn->sn_batches);
401 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
403 sfw_session_t *sn = sfw_data.fw_session;
404 sfw_counters_t *cnt = &reply->str_fw;
407 reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
409 if (request->str_sid.ses_nid == LNET_NID_ANY) {
410 reply->str_status = EINVAL;
414 if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
415 reply->str_status = ESRCH;
420 reply->str_lnet = the_lnet.ln_counters;
423 srpc_get_counters(&reply->str_rpc);
425 cnt->brw_errors = cfs_atomic_read(&sn->sn_brw_errors);
426 cnt->ping_errors = cfs_atomic_read(&sn->sn_ping_errors);
427 cnt->zombie_sessions = cfs_atomic_read(&sfw_data.fw_nzombies);
429 cnt->active_tests = cnt->active_batches = 0;
430 cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
431 sfw_batch_t, bat_list) {
432 int n = cfs_atomic_read(&bat->bat_nactive);
435 cnt->active_batches++;
436 cnt->active_tests += n;
440 reply->str_status = 0;
445 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
447 sfw_session_t *sn = sfw_data.fw_session;
449 if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
450 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
451 reply->mksn_status = EINVAL;
456 reply->mksn_status = 0;
457 reply->mksn_sid = sn->sn_id;
458 reply->mksn_timeout = sn->sn_timeout;
460 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
461 cfs_atomic_inc(&sn->sn_refcount);
465 if (!request->mksn_force) {
466 reply->mksn_status = EBUSY;
467 strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
472 /* brand new or create by force */
473 LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
475 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
479 sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
481 cfs_spin_lock(&sfw_data.fw_lock);
483 sfw_deactivate_session();
484 LASSERT (sfw_data.fw_session == NULL);
485 sfw_data.fw_session = sn;
487 cfs_spin_unlock(&sfw_data.fw_lock);
489 reply->mksn_status = 0;
490 reply->mksn_sid = sn->sn_id;
491 reply->mksn_timeout = sn->sn_timeout;
496 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
498 sfw_session_t *sn = sfw_data.fw_session;
500 reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
502 if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
503 reply->rmsn_status = EINVAL;
507 if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
508 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
512 if (!cfs_atomic_dec_and_test(&sn->sn_refcount)) {
513 reply->rmsn_status = 0;
517 cfs_spin_lock(&sfw_data.fw_lock);
518 sfw_deactivate_session();
519 cfs_spin_unlock(&sfw_data.fw_lock);
521 reply->rmsn_status = 0;
522 reply->rmsn_sid = LST_INVALID_SID;
523 LASSERT (sfw_data.fw_session == NULL);
528 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
530 sfw_session_t *sn = sfw_data.fw_session;
533 reply->dbg_status = ESRCH;
534 reply->dbg_sid = LST_INVALID_SID;
538 reply->dbg_status = 0;
539 reply->dbg_sid = sn->sn_id;
540 reply->dbg_timeout = sn->sn_timeout;
541 strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
547 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
549 sfw_test_unit_t *tsu = rpc->crpc_priv;
550 sfw_test_instance_t *tsi = tsu->tsu_instance;
552 /* Called with hold of tsi->tsi_lock */
553 LASSERT (cfs_list_empty(&rpc->crpc_list));
554 cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
558 sfw_load_test (sfw_test_instance_t *tsi)
560 sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
561 int nrequired = sfw_test_buffers(tsi);
564 LASSERT (tsc != NULL);
566 if (tsi->tsi_is_client) {
567 tsi->tsi_ops = tsc->tsc_cli_ops;
571 nposted = srpc_service_add_buffers(tsc->tsc_srv_service, nrequired);
572 if (nposted != nrequired) {
573 CWARN ("Failed to reserve enough buffers: "
574 "service %s, %d needed, %d reserved\n",
575 tsc->tsc_srv_service->sv_name, nrequired, nposted);
576 srpc_service_remove_buffers(tsc->tsc_srv_service, nposted);
580 CDEBUG (D_NET, "Reserved %d buffers for test %s\n",
581 nposted, tsc->tsc_srv_service->sv_name);
586 sfw_unload_test (sfw_test_instance_t *tsi)
588 sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
590 LASSERT (tsc != NULL);
592 if (!tsi->tsi_is_client)
593 srpc_service_remove_buffers(tsc->tsc_srv_service,
594 sfw_test_buffers(tsi));
599 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
601 srpc_client_rpc_t *rpc;
602 sfw_test_unit_t *tsu;
604 if (!tsi->tsi_is_client) goto clean;
606 tsi->tsi_ops->tso_fini(tsi);
608 LASSERT (!tsi->tsi_stopping);
609 LASSERT (cfs_list_empty(&tsi->tsi_active_rpcs));
610 LASSERT (!sfw_test_active(tsi));
612 while (!cfs_list_empty(&tsi->tsi_units)) {
613 tsu = cfs_list_entry(tsi->tsi_units.next,
614 sfw_test_unit_t, tsu_list);
615 cfs_list_del(&tsu->tsu_list);
616 LIBCFS_FREE(tsu, sizeof(*tsu));
619 while (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
620 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
621 srpc_client_rpc_t, crpc_list);
622 cfs_list_del(&rpc->crpc_list);
623 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
627 sfw_unload_test(tsi);
628 LIBCFS_FREE(tsi, sizeof(*tsi));
633 sfw_destroy_batch (sfw_batch_t *tsb)
635 sfw_test_instance_t *tsi;
637 LASSERT (!sfw_batch_active(tsb));
638 LASSERT (cfs_list_empty(&tsb->bat_list));
640 while (!cfs_list_empty(&tsb->bat_tests)) {
641 tsi = cfs_list_entry(tsb->bat_tests.next,
642 sfw_test_instance_t, tsi_list);
643 cfs_list_del_init(&tsi->tsi_list);
644 sfw_destroy_test_instance(tsi);
647 LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
652 sfw_destroy_session (sfw_session_t *sn)
656 LASSERT (cfs_list_empty(&sn->sn_list));
657 LASSERT (sn != sfw_data.fw_session);
659 while (!cfs_list_empty(&sn->sn_batches)) {
660 batch = cfs_list_entry(sn->sn_batches.next,
661 sfw_batch_t, bat_list);
662 cfs_list_del_init(&batch->bat_list);
663 sfw_destroy_batch(batch);
666 LIBCFS_FREE(sn, sizeof(*sn));
667 cfs_atomic_dec(&sfw_data.fw_nzombies);
672 sfw_unpack_test_req (srpc_msg_t *msg)
674 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
676 LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
677 LASSERT (req->tsr_is_client);
679 if (msg->msg_magic == SRPC_MSG_MAGIC)
680 return; /* no flipping needed */
682 LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
684 if (req->tsr_service == SRPC_SERVICE_BRW) {
685 test_bulk_req_t *bulk = &req->tsr_u.bulk;
687 __swab32s(&bulk->blk_opc);
688 __swab32s(&bulk->blk_npg);
689 __swab32s(&bulk->blk_flags);
693 if (req->tsr_service == SRPC_SERVICE_PING) {
694 test_ping_req_t *ping = &req->tsr_u.ping;
696 __swab32s(&ping->png_size);
697 __swab32s(&ping->png_flags);
706 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
708 srpc_msg_t *msg = &rpc->srpc_reqstbuf->buf_msg;
709 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
710 srpc_bulk_t *bk = rpc->srpc_bulk;
711 int ndest = req->tsr_ndest;
712 sfw_test_unit_t *tsu;
713 sfw_test_instance_t *tsi;
717 LIBCFS_ALLOC(tsi, sizeof(*tsi));
719 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
724 memset(tsi, 0, sizeof(*tsi));
725 cfs_spin_lock_init(&tsi->tsi_lock);
726 cfs_atomic_set(&tsi->tsi_nactive, 0);
727 CFS_INIT_LIST_HEAD(&tsi->tsi_units);
728 CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
729 CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
731 tsi->tsi_stopping = 0;
732 tsi->tsi_batch = tsb;
733 tsi->tsi_loop = req->tsr_loop;
734 tsi->tsi_concur = req->tsr_concur;
735 tsi->tsi_service = req->tsr_service;
736 tsi->tsi_is_client = !!(req->tsr_is_client);
737 tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
739 rc = sfw_load_test(tsi);
741 LIBCFS_FREE(tsi, sizeof(*tsi));
745 LASSERT (!sfw_batch_active(tsb));
747 if (!tsi->tsi_is_client) {
748 /* it's test server, just add it to tsb */
749 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
753 LASSERT (bk != NULL);
755 LASSERT (bk->bk_pages != NULL);
757 LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
758 LASSERT ((unsigned int)bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
760 sfw_unpack_test_req(msg);
761 memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
763 for (i = 0; i < ndest; i++) {
764 lnet_process_id_packed_t *dests;
765 lnet_process_id_packed_t id;
769 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
770 LASSERT (dests != NULL); /* my pages are within KVM always */
772 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
774 id = dests[i % SFW_ID_PER_PAGE];
775 if (msg->msg_magic != SRPC_MSG_MAGIC)
778 for (j = 0; j < tsi->tsi_concur; j++) {
779 LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
782 CERROR ("Can't allocate tsu for %d\n",
787 tsu->tsu_dest.nid = id.nid;
788 tsu->tsu_dest.pid = id.pid;
789 tsu->tsu_instance = tsi;
790 tsu->tsu_private = NULL;
791 cfs_list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
795 rc = tsi->tsi_ops->tso_init(tsi);
797 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
803 sfw_destroy_test_instance(tsi);
808 sfw_test_unit_done (sfw_test_unit_t *tsu)
810 sfw_test_instance_t *tsi = tsu->tsu_instance;
811 sfw_batch_t *tsb = tsi->tsi_batch;
812 sfw_session_t *sn = tsb->bat_session;
814 LASSERT (sfw_test_active(tsi));
816 if (!cfs_atomic_dec_and_test(&tsi->tsi_nactive))
819 /* the test instance is done */
820 cfs_spin_lock(&tsi->tsi_lock);
822 tsi->tsi_stopping = 0;
824 cfs_spin_unlock(&tsi->tsi_lock);
826 cfs_spin_lock(&sfw_data.fw_lock);
828 if (!cfs_atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
829 sn == sfw_data.fw_session) { /* sn also active */
830 cfs_spin_unlock(&sfw_data.fw_lock);
834 LASSERT (!cfs_list_empty(&sn->sn_list)); /* I'm a zombie! */
836 cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
837 sfw_batch_t, bat_list) {
838 if (sfw_batch_active(tsb)) {
839 cfs_spin_unlock(&sfw_data.fw_lock);
844 cfs_list_del_init(&sn->sn_list);
845 cfs_spin_unlock(&sfw_data.fw_lock);
847 sfw_destroy_session(sn);
852 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
854 sfw_test_unit_t *tsu = rpc->crpc_priv;
855 sfw_test_instance_t *tsi = tsu->tsu_instance;
858 tsi->tsi_ops->tso_done_rpc(tsu, rpc);
860 cfs_spin_lock(&tsi->tsi_lock);
862 LASSERT (sfw_test_active(tsi));
863 LASSERT (!cfs_list_empty(&rpc->crpc_list));
865 cfs_list_del_init(&rpc->crpc_list);
867 /* batch is stopping or loop is done or get error */
868 if (tsi->tsi_stopping ||
869 tsu->tsu_loop == 0 ||
870 (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
873 /* dec ref for poster */
874 srpc_client_rpc_decref(rpc);
876 cfs_spin_unlock(&tsi->tsi_lock);
879 swi_schedule_workitem(&tsu->tsu_worker);
883 sfw_test_unit_done(tsu);
888 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
889 int nblk, int blklen, srpc_client_rpc_t **rpcpp)
891 srpc_client_rpc_t *rpc = NULL;
892 sfw_test_instance_t *tsi = tsu->tsu_instance;
894 cfs_spin_lock(&tsi->tsi_lock);
896 LASSERT (sfw_test_active(tsi));
898 if (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
899 /* pick request from buffer */
900 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
901 srpc_client_rpc_t, crpc_list);
902 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
903 cfs_list_del_init(&rpc->crpc_list);
905 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
906 blklen, sfw_test_rpc_done,
907 sfw_test_rpc_fini, tsu);
910 cfs_spin_unlock(&tsi->tsi_lock);
913 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
914 blklen, sfw_test_rpc_done,
915 sfw_test_rpc_fini, tsu);
917 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
926 sfw_run_test (swi_workitem_t *wi)
928 sfw_test_unit_t *tsu = wi->swi_workitem.wi_data;
929 sfw_test_instance_t *tsi = tsu->tsu_instance;
930 srpc_client_rpc_t *rpc = NULL;
932 LASSERT (wi == &tsu->tsu_worker);
934 if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
935 LASSERT (rpc == NULL);
939 LASSERT (rpc != NULL);
941 cfs_spin_lock(&tsi->tsi_lock);
943 if (tsi->tsi_stopping) {
944 cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
945 cfs_spin_unlock(&tsi->tsi_lock);
949 if (tsu->tsu_loop > 0)
952 cfs_list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
953 cfs_spin_unlock(&tsi->tsi_lock);
955 rpc->crpc_timeout = rpc_timeout;
957 cfs_spin_lock(&rpc->crpc_lock);
959 cfs_spin_unlock(&rpc->crpc_lock);
964 * No one can schedule me now since:
965 * - previous RPC, if any, has done and
966 * - no new RPC is initiated.
967 * - my batch is still active; no one can run it again now.
968 * Cancel pending schedules and prevent future schedule attempts:
970 swi_kill_workitem(wi);
971 sfw_test_unit_done(tsu);
976 sfw_run_batch (sfw_batch_t *tsb)
979 sfw_test_unit_t *tsu;
980 sfw_test_instance_t *tsi;
982 if (sfw_batch_active(tsb)) {
983 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
984 tsb->bat_id.bat_id, cfs_atomic_read(&tsb->bat_nactive));
988 cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
989 sfw_test_instance_t, tsi_list) {
990 if (!tsi->tsi_is_client) /* skip server instances */
993 LASSERT (!tsi->tsi_stopping);
994 LASSERT (!sfw_test_active(tsi));
996 cfs_atomic_inc(&tsb->bat_nactive);
998 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
999 sfw_test_unit_t, tsu_list) {
1000 cfs_atomic_inc(&tsi->tsi_nactive);
1001 tsu->tsu_loop = tsi->tsi_loop;
1002 wi = &tsu->tsu_worker;
1003 swi_init_workitem(wi, tsu, sfw_run_test,
1005 swi_schedule_workitem(wi);
1013 sfw_stop_batch (sfw_batch_t *tsb, int force)
1015 sfw_test_instance_t *tsi;
1016 srpc_client_rpc_t *rpc;
1018 if (!sfw_batch_active(tsb)) {
1019 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1023 cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1024 sfw_test_instance_t, tsi_list) {
1025 cfs_spin_lock(&tsi->tsi_lock);
1027 if (!tsi->tsi_is_client ||
1028 !sfw_test_active(tsi) || tsi->tsi_stopping) {
1029 cfs_spin_unlock(&tsi->tsi_lock);
1033 tsi->tsi_stopping = 1;
1036 cfs_spin_unlock(&tsi->tsi_lock);
1040 /* abort launched rpcs in the test */
1041 cfs_list_for_each_entry_typed (rpc, &tsi->tsi_active_rpcs,
1042 srpc_client_rpc_t, crpc_list) {
1043 cfs_spin_lock(&rpc->crpc_lock);
1045 srpc_abort_rpc(rpc, -EINTR);
1047 cfs_spin_unlock(&rpc->crpc_lock);
1050 cfs_spin_unlock(&tsi->tsi_lock);
1057 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1059 sfw_test_instance_t *tsi;
1065 reply->bar_active = cfs_atomic_read(&tsb->bat_nactive);
1069 cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1070 sfw_test_instance_t, tsi_list) {
1074 reply->bar_active = cfs_atomic_read(&tsi->tsi_nactive);
1082 sfw_free_pages (srpc_server_rpc_t *rpc)
1084 srpc_free_bulk(rpc->srpc_bulk);
1085 rpc->srpc_bulk = NULL;
1089 sfw_alloc_pages (srpc_server_rpc_t *rpc, int npages, int sink)
1091 LASSERT (rpc->srpc_bulk == NULL);
1092 LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
1094 rpc->srpc_bulk = srpc_alloc_bulk(npages, sink);
1095 if (rpc->srpc_bulk == NULL) return -ENOMEM;
1101 sfw_add_test (srpc_server_rpc_t *rpc)
1103 sfw_session_t *sn = sfw_data.fw_session;
1104 srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1105 srpc_test_reqst_t *request;
1109 request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1110 reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1112 if (request->tsr_loop == 0 ||
1113 request->tsr_concur == 0 ||
1114 request->tsr_sid.ses_nid == LNET_NID_ANY ||
1115 request->tsr_ndest > SFW_MAX_NDESTS ||
1116 (request->tsr_is_client && request->tsr_ndest == 0) ||
1117 request->tsr_concur > SFW_MAX_CONCUR ||
1118 request->tsr_service > SRPC_SERVICE_MAX_ID ||
1119 request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1120 reply->tsr_status = EINVAL;
1124 if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1125 sfw_find_test_case(request->tsr_service) == NULL) {
1126 reply->tsr_status = ENOENT;
1130 bat = sfw_bid2batch(request->tsr_bid);
1132 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1133 rpc->srpc_service->sv_name,
1134 libcfs_id2str(rpc->srpc_peer));
1138 if (sfw_batch_active(bat)) {
1139 reply->tsr_status = EBUSY;
1143 if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1144 /* rpc will be resumed later in sfw_bulk_ready */
1145 return sfw_alloc_pages(rpc,
1146 sfw_id_pages(request->tsr_ndest), 1);
1149 rc = sfw_add_test_instance(bat, rpc);
1150 CDEBUG (rc == 0 ? D_NET : D_WARNING,
1151 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1152 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1153 request->tsr_is_client ? "client" : "server",
1154 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1156 reply->tsr_status = (rc < 0) ? -rc : rc;
1161 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1163 sfw_session_t *sn = sfw_data.fw_session;
1167 reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1169 if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1170 reply->bar_status = ESRCH;
1174 bat = sfw_find_batch(request->bar_bid);
1176 reply->bar_status = ENOENT;
1180 switch (request->bar_opc) {
1181 case SRPC_BATCH_OPC_RUN:
1182 rc = sfw_run_batch(bat);
1185 case SRPC_BATCH_OPC_STOP:
1186 rc = sfw_stop_batch(bat, request->bar_arg);
1189 case SRPC_BATCH_OPC_QUERY:
1190 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1194 return -EINVAL; /* drop it */
1197 reply->bar_status = (rc < 0) ? -rc : rc;
1202 sfw_handle_server_rpc (srpc_server_rpc_t *rpc)
1204 srpc_service_t *sv = rpc->srpc_service;
1205 srpc_msg_t *reply = &rpc->srpc_replymsg;
1206 srpc_msg_t *request = &rpc->srpc_reqstbuf->buf_msg;
1209 LASSERT (sfw_data.fw_active_srpc == NULL);
1210 LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1212 cfs_spin_lock(&sfw_data.fw_lock);
1214 if (sfw_data.fw_shuttingdown) {
1215 cfs_spin_unlock(&sfw_data.fw_lock);
1219 /* Remove timer to avoid racing with it or expiring active session */
1220 if (sfw_del_session_timer() != 0) {
1221 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1222 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1223 cfs_spin_unlock(&sfw_data.fw_lock);
1227 sfw_data.fw_active_srpc = rpc;
1228 cfs_spin_unlock(&sfw_data.fw_lock);
1230 sfw_unpack_message(request);
1231 LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1236 case SRPC_SERVICE_TEST:
1237 rc = sfw_add_test(rpc);
1240 case SRPC_SERVICE_BATCH:
1241 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1242 &reply->msg_body.bat_reply);
1245 case SRPC_SERVICE_QUERY_STAT:
1246 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1247 &reply->msg_body.stat_reply);
1250 case SRPC_SERVICE_DEBUG:
1251 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1252 &reply->msg_body.dbg_reply);
1255 case SRPC_SERVICE_MAKE_SESSION:
1256 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1257 &reply->msg_body.mksn_reply);
1260 case SRPC_SERVICE_REMOVE_SESSION:
1261 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1262 &reply->msg_body.rmsn_reply);
1266 rpc->srpc_done = sfw_server_rpc_done;
1267 cfs_spin_lock(&sfw_data.fw_lock);
1270 if (!sfw_data.fw_shuttingdown)
1271 sfw_add_session_timer();
1273 LASSERT (!sfw_data.fw_shuttingdown);
1274 sfw_add_session_timer();
1277 sfw_data.fw_active_srpc = NULL;
1278 cfs_spin_unlock(&sfw_data.fw_lock);
1283 sfw_bulk_ready (srpc_server_rpc_t *rpc, int status)
1285 srpc_service_t *sv = rpc->srpc_service;
1288 LASSERT (rpc->srpc_bulk != NULL);
1289 LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1290 LASSERT (sfw_data.fw_active_srpc == NULL);
1291 LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1293 cfs_spin_lock(&sfw_data.fw_lock);
1296 CERROR ("Bulk transfer failed for RPC: "
1297 "service %s, peer %s, status %d\n",
1298 sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1299 cfs_spin_unlock(&sfw_data.fw_lock);
1303 if (sfw_data.fw_shuttingdown) {
1304 cfs_spin_unlock(&sfw_data.fw_lock);
1308 if (sfw_del_session_timer() != 0) {
1309 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1310 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1311 cfs_spin_unlock(&sfw_data.fw_lock);
1315 sfw_data.fw_active_srpc = rpc;
1316 cfs_spin_unlock(&sfw_data.fw_lock);
1318 rc = sfw_add_test(rpc);
1320 cfs_spin_lock(&sfw_data.fw_lock);
1323 if (!sfw_data.fw_shuttingdown)
1324 sfw_add_session_timer();
1326 LASSERT (!sfw_data.fw_shuttingdown);
1327 sfw_add_session_timer();
1330 sfw_data.fw_active_srpc = NULL;
1331 cfs_spin_unlock(&sfw_data.fw_lock);
1336 sfw_create_rpc (lnet_process_id_t peer, int service,
1337 int nbulkiov, int bulklen,
1338 void (*done) (srpc_client_rpc_t *), void *priv)
1340 srpc_client_rpc_t *rpc;
1342 cfs_spin_lock(&sfw_data.fw_lock);
1344 LASSERT (!sfw_data.fw_shuttingdown);
1345 LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1347 if (nbulkiov == 0 && !cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1348 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1349 srpc_client_rpc_t, crpc_list);
1350 cfs_list_del(&rpc->crpc_list);
1351 cfs_spin_unlock(&sfw_data.fw_lock);
1353 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1354 done, sfw_client_rpc_fini, priv);
1358 cfs_spin_unlock(&sfw_data.fw_lock);
1360 rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1361 nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1367 sfw_unpack_message (srpc_msg_t *msg)
1369 if (msg->msg_magic == SRPC_MSG_MAGIC)
1370 return; /* no flipping needed */
1372 LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1374 __swab32s(&msg->msg_type);
1376 if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1377 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1379 __swab32s(&req->str_type);
1380 __swab64s(&req->str_rpyid);
1381 sfw_unpack_sid(req->str_sid);
1385 if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1386 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1388 __swab32s(&rep->str_status);
1389 sfw_unpack_sid(rep->str_sid);
1390 sfw_unpack_fw_counters(rep->str_fw);
1391 sfw_unpack_rpc_counters(rep->str_rpc);
1392 sfw_unpack_lnet_counters(rep->str_lnet);
1396 if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1397 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1399 __swab64s(&req->mksn_rpyid);
1400 __swab32s(&req->mksn_force);
1401 sfw_unpack_sid(req->mksn_sid);
1405 if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1406 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1408 __swab32s(&rep->mksn_status);
1409 __swab32s(&rep->mksn_timeout);
1410 sfw_unpack_sid(rep->mksn_sid);
1414 if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1415 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1417 __swab64s(&req->rmsn_rpyid);
1418 sfw_unpack_sid(req->rmsn_sid);
1422 if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1423 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1425 __swab32s(&rep->rmsn_status);
1426 sfw_unpack_sid(rep->rmsn_sid);
1430 if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1431 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1433 __swab64s(&req->dbg_rpyid);
1434 __swab32s(&req->dbg_flags);
1435 sfw_unpack_sid(req->dbg_sid);
1439 if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1440 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1442 __swab32s(&rep->dbg_nbatch);
1443 __swab32s(&rep->dbg_timeout);
1444 sfw_unpack_sid(rep->dbg_sid);
1448 if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1449 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1451 __swab32s(&req->bar_opc);
1452 __swab64s(&req->bar_rpyid);
1453 __swab32s(&req->bar_testidx);
1454 __swab32s(&req->bar_arg);
1455 sfw_unpack_sid(req->bar_sid);
1456 __swab64s(&req->bar_bid.bat_id);
1460 if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1461 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1463 __swab32s(&rep->bar_status);
1464 sfw_unpack_sid(rep->bar_sid);
1468 if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1469 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1471 __swab64s(&req->tsr_rpyid);
1472 __swab64s(&req->tsr_bulkid);
1473 __swab32s(&req->tsr_loop);
1474 __swab32s(&req->tsr_ndest);
1475 __swab32s(&req->tsr_concur);
1476 __swab32s(&req->tsr_service);
1477 sfw_unpack_sid(req->tsr_sid);
1478 __swab64s(&req->tsr_bid.bat_id);
1482 if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1483 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1485 __swab32s(&rep->tsr_status);
1486 sfw_unpack_sid(rep->tsr_sid);
1490 if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1491 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1493 __swab64s(&req->join_rpyid);
1494 sfw_unpack_sid(req->join_sid);
1498 if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1499 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1501 __swab32s(&rep->join_status);
1502 __swab32s(&rep->join_timeout);
1503 sfw_unpack_sid(rep->join_sid);
1512 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1514 LASSERT (cfs_atomic_read(&rpc->crpc_refcount) > 0);
1515 LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1517 cfs_spin_lock(&rpc->crpc_lock);
1518 srpc_abort_rpc(rpc, -EINTR);
1519 cfs_spin_unlock(&rpc->crpc_lock);
1524 sfw_post_rpc (srpc_client_rpc_t *rpc)
1526 cfs_spin_lock(&rpc->crpc_lock);
1528 LASSERT (!rpc->crpc_closed);
1529 LASSERT (!rpc->crpc_aborted);
1530 LASSERT (cfs_list_empty(&rpc->crpc_list));
1531 LASSERT (!sfw_data.fw_shuttingdown);
1533 rpc->crpc_timeout = rpc_timeout;
1536 cfs_spin_unlock(&rpc->crpc_lock);
1540 static srpc_service_t sfw_services[] =
1543 /* sv_id */ SRPC_SERVICE_DEBUG,
1544 /* sv_name */ "debug",
1548 /* sv_id */ SRPC_SERVICE_QUERY_STAT,
1549 /* sv_name */ "query stats",
1553 /* sv_id */ SRPC_SERVICE_MAKE_SESSION,
1554 /* sv_name */ "make session",
1558 /* sv_id */ SRPC_SERVICE_REMOVE_SESSION,
1559 /* sv_name */ "remove session",
1563 /* sv_id */ SRPC_SERVICE_BATCH,
1564 /* sv_name */ "batch service",
1568 /* sv_id */ SRPC_SERVICE_TEST,
1569 /* sv_name */ "test service",
1579 extern sfw_test_client_ops_t ping_test_client;
1580 extern srpc_service_t ping_test_service;
1581 extern void ping_init_test_client(void);
1582 extern void ping_init_test_service(void);
1584 extern sfw_test_client_ops_t brw_test_client;
1585 extern srpc_service_t brw_test_service;
1586 extern void brw_init_test_client(void);
1587 extern void brw_init_test_service(void);
1597 sfw_test_case_t *tsc;
1602 s = getenv("SESSION_TIMEOUT");
1603 session_timeout = s != NULL ? atoi(s) : session_timeout;
1605 s = getenv("BRW_INJECT_ERRORS");
1606 brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
1608 s = getenv("RPC_TIMEOUT");
1609 rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1612 if (session_timeout < 0) {
1613 CERROR ("Session timeout must be non-negative: %d\n",
1618 if (rpc_timeout < 0) {
1619 CERROR ("RPC timeout must be non-negative: %d\n",
1624 if (session_timeout == 0)
1625 CWARN ("Zero session_timeout specified "
1626 "- test sessions never expire.\n");
1628 if (rpc_timeout == 0)
1629 CWARN ("Zero rpc_timeout specified "
1630 "- test RPC never expire.\n");
1632 memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1634 sfw_data.fw_session = NULL;
1635 sfw_data.fw_active_srpc = NULL;
1636 cfs_spin_lock_init(&sfw_data.fw_lock);
1637 cfs_atomic_set(&sfw_data.fw_nzombies, 0);
1638 CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1639 CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1640 CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1642 brw_init_test_client();
1643 brw_init_test_service();
1644 rc = sfw_register_test(&brw_test_service, &brw_test_client);
1647 ping_init_test_client();
1648 ping_init_test_service();
1649 rc = sfw_register_test(&ping_test_service, &ping_test_client);
1653 cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1654 sfw_test_case_t, tsc_list) {
1655 sv = tsc->tsc_srv_service;
1656 sv->sv_concur = SFW_TEST_CONCURRENCY;
1658 rc = srpc_add_service(sv);
1659 LASSERT (rc != -EBUSY);
1661 CWARN ("Failed to add %s service: %d\n",
1667 for (i = 0; ; i++) {
1668 sv = &sfw_services[i];
1669 if (sv->sv_name == NULL) break;
1671 sv->sv_bulk_ready = NULL;
1672 sv->sv_handler = sfw_handle_server_rpc;
1673 sv->sv_concur = SFW_SERVICE_CONCURRENCY;
1674 if (sv->sv_id == SRPC_SERVICE_TEST)
1675 sv->sv_bulk_ready = sfw_bulk_ready;
1677 rc = srpc_add_service(sv);
1678 LASSERT (rc != -EBUSY);
1680 CWARN ("Failed to add %s service: %d\n",
1685 /* about to sfw_shutdown, no need to add buffer */
1686 if (error) continue;
1688 rc = srpc_service_add_buffers(sv, SFW_POST_BUFFERS);
1689 if (rc != SFW_POST_BUFFERS) {
1690 CWARN ("Failed to reserve enough buffers: "
1691 "service %s, %d needed, %d reserved\n",
1692 sv->sv_name, SFW_POST_BUFFERS, rc);
1706 sfw_test_case_t *tsc;
1709 cfs_spin_lock(&sfw_data.fw_lock);
1711 sfw_data.fw_shuttingdown = 1;
1713 lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1714 "waiting for active RPC to finish.\n");
1716 LASSERT (sfw_data.fw_active_srpc == NULL);
1719 if (sfw_del_session_timer() != 0)
1720 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1721 "waiting for session timer to explode.\n");
1723 sfw_deactivate_session();
1724 lst_wait_until(cfs_atomic_read(&sfw_data.fw_nzombies) == 0,
1726 "waiting for %d zombie sessions to die.\n",
1727 cfs_atomic_read(&sfw_data.fw_nzombies));
1729 cfs_spin_unlock(&sfw_data.fw_lock);
1731 for (i = 0; ; i++) {
1732 sv = &sfw_services[i];
1733 if (sv->sv_name == NULL)
1736 srpc_shutdown_service(sv);
1737 srpc_remove_service(sv);
1740 cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1741 sfw_test_case_t, tsc_list) {
1742 sv = tsc->tsc_srv_service;
1743 srpc_shutdown_service(sv);
1744 srpc_remove_service(sv);
1747 while (!cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1748 srpc_client_rpc_t *rpc;
1750 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1751 srpc_client_rpc_t, crpc_list);
1752 cfs_list_del(&rpc->crpc_list);
1754 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1757 for (i = 0; ; i++) {
1758 sv = &sfw_services[i];
1759 if (sv->sv_name == NULL)
1762 srpc_wait_service_shutdown(sv);
1765 while (!cfs_list_empty(&sfw_data.fw_tests)) {
1766 tsc = cfs_list_entry(sfw_data.fw_tests.next,
1767 sfw_test_case_t, tsc_list);
1769 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1771 cfs_list_del(&tsc->tsc_list);
1772 LIBCFS_FREE(tsc, sizeof(*tsc));