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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lnet/selftest/framework.c
34 * Author: Isaac Huang <isaac@clusterfs.com>
35 * Author: Liang Zhen <liangzhen@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_LNET
42 struct lst_sid LST_INVALID_SID = { .ses_nid = LNET_NID_ANY, .ses_stamp = -1};
44 static int session_timeout = 100;
45 module_param(session_timeout, int, 0444);
46 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
48 static int rpc_timeout = 64;
49 module_param(rpc_timeout, int, 0644);
50 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
52 #define sfw_unpack_id(id) \
54 __swab64s(&(id).nid); \
55 __swab32s(&(id).pid); \
58 #define sfw_unpack_sid(sid) \
60 __swab64s(&(sid).ses_nid); \
61 __swab64s(&(sid).ses_stamp); \
64 #define sfw_unpack_fw_counters(fc) \
66 __swab32s(&(fc).running_ms); \
67 __swab32s(&(fc).active_batches); \
68 __swab32s(&(fc).zombie_sessions); \
69 __swab32s(&(fc).brw_errors); \
70 __swab32s(&(fc).ping_errors); \
73 #define sfw_unpack_rpc_counters(rc) \
75 __swab32s(&(rc).errors); \
76 __swab32s(&(rc).rpcs_sent); \
77 __swab32s(&(rc).rpcs_rcvd); \
78 __swab32s(&(rc).rpcs_dropped); \
79 __swab32s(&(rc).rpcs_expired); \
80 __swab64s(&(rc).bulk_get); \
81 __swab64s(&(rc).bulk_put); \
84 #define sfw_unpack_lnet_counters(lc) \
86 __swab32s(&(lc).lcc_errors); \
87 __swab32s(&(lc).lcc_msgs_max); \
88 __swab32s(&(lc).lcc_msgs_alloc); \
89 __swab32s(&(lc).lcc_send_count); \
90 __swab32s(&(lc).lcc_recv_count); \
91 __swab32s(&(lc).lcc_drop_count); \
92 __swab32s(&(lc).lcc_route_count); \
93 __swab64s(&(lc).lcc_send_length); \
94 __swab64s(&(lc).lcc_recv_length); \
95 __swab64s(&(lc).lcc_drop_length); \
96 __swab64s(&(lc).lcc_route_length); \
99 #define sfw_test_active(t) (atomic_read(&(t)->tsi_nactive) != 0)
100 #define sfw_batch_active(b) (atomic_read(&(b)->bat_nactive) != 0)
102 static struct smoketest_framework {
103 /* RPCs to be recycled */
104 struct list_head fw_zombie_rpcs;
105 /* stopping sessions */
106 struct list_head fw_zombie_sessions;
107 /* registered test cases */
108 struct list_head fw_tests;
109 /* # zombie sessions */
110 atomic_t fw_nzombies;
114 struct sfw_session *fw_session;
115 /* shutdown in progress */
118 struct srpc_server_rpc *fw_active_srpc;
122 static int sfw_stop_batch(struct sfw_batch *tsb, int force);
123 static void sfw_destroy_session(struct sfw_session *sn);
125 static inline struct sfw_test_case *
126 sfw_find_test_case(int id)
128 struct sfw_test_case *tsc;
130 LASSERT(id <= SRPC_SERVICE_MAX_ID);
131 LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
133 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
134 if (tsc->tsc_srv_service->sv_id == id)
142 sfw_register_test(struct srpc_service *service,
143 struct sfw_test_client_ops *cliops)
145 struct sfw_test_case *tsc;
147 if (sfw_find_test_case(service->sv_id) != NULL) {
148 CERROR ("Failed to register test %s (%d)\n",
149 service->sv_name, service->sv_id);
153 LIBCFS_ALLOC(tsc, sizeof(*tsc));
157 tsc->tsc_cli_ops = cliops;
158 tsc->tsc_srv_service = service;
160 list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
165 sfw_add_session_timer (void)
167 struct sfw_session *sn = sfw_data.fw_session;
168 struct stt_timer *timer = &sn->sn_timer;
170 LASSERT (!sfw_data.fw_shuttingdown);
172 if (sn == NULL || sn->sn_timeout == 0)
175 LASSERT (!sn->sn_timer_active);
177 sn->sn_timer_active = 1;
178 timer->stt_expires = ktime_get_real_seconds()+ sn->sn_timeout;
179 stt_add_timer(timer);
184 sfw_del_session_timer (void)
186 struct sfw_session *sn = sfw_data.fw_session;
188 if (sn == NULL || !sn->sn_timer_active)
191 LASSERT (sn->sn_timeout != 0);
193 if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
194 sn->sn_timer_active = 0;
198 return EBUSY; /* racing with sfw_session_expired() */
201 /* called with sfw_data.fw_lock held */
203 sfw_deactivate_session (void)
204 __must_hold(&sfw_data.fw_lock)
206 struct sfw_session *sn = sfw_data.fw_session;
208 struct sfw_batch *tsb;
209 struct sfw_test_case *tsc;
211 if (sn == NULL) return;
213 LASSERT(!sn->sn_timer_active);
215 sfw_data.fw_session = NULL;
216 atomic_inc(&sfw_data.fw_nzombies);
217 list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
219 spin_unlock(&sfw_data.fw_lock);
221 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
222 srpc_abort_service(tsc->tsc_srv_service);
225 spin_lock(&sfw_data.fw_lock);
227 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
228 if (sfw_batch_active(tsb)) {
230 sfw_stop_batch(tsb, 1);
235 return; /* wait for active batches to stop */
237 list_del_init(&sn->sn_list);
238 spin_unlock(&sfw_data.fw_lock);
240 sfw_destroy_session(sn);
242 spin_lock(&sfw_data.fw_lock);
247 sfw_session_expired (void *data)
249 struct sfw_session *sn = data;
251 spin_lock(&sfw_data.fw_lock);
253 LASSERT (sn->sn_timer_active);
254 LASSERT (sn == sfw_data.fw_session);
256 CWARN ("Session expired! sid: %s-%llu, name: %s\n",
257 libcfs_nid2str(sn->sn_id.ses_nid),
258 sn->sn_id.ses_stamp, &sn->sn_name[0]);
260 sn->sn_timer_active = 0;
261 sfw_deactivate_session();
263 spin_unlock(&sfw_data.fw_lock);
267 sfw_init_session(struct sfw_session *sn, struct lst_sid sid,
268 unsigned features, const char *name)
270 struct stt_timer *timer = &sn->sn_timer;
272 memset(sn, 0, sizeof(struct sfw_session));
273 INIT_LIST_HEAD(&sn->sn_list);
274 INIT_LIST_HEAD(&sn->sn_batches);
275 atomic_set(&sn->sn_refcount, 1); /* +1 for caller */
276 atomic_set(&sn->sn_brw_errors, 0);
277 atomic_set(&sn->sn_ping_errors, 0);
278 strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
280 sn->sn_timer_active = 0;
282 sn->sn_features = features;
283 sn->sn_timeout = session_timeout;
284 sn->sn_started = ktime_get();
286 timer->stt_data = sn;
287 timer->stt_func = sfw_session_expired;
288 INIT_LIST_HEAD(&timer->stt_list);
291 /* completion handler for incoming framework RPCs */
293 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
295 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
296 int status = rpc->srpc_status;
299 "Incoming framework RPC done: "
300 "service %s, peer %s, status %s:%d\n",
301 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
302 swi_state2str(rpc->srpc_wi.swi_state),
305 if (rpc->srpc_bulk != NULL)
311 sfw_client_rpc_fini(struct srpc_client_rpc *rpc)
313 LASSERT(rpc->crpc_bulk.bk_niov == 0);
314 LASSERT(list_empty(&rpc->crpc_list));
315 LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
317 CDEBUG(D_NET, "Outgoing framework RPC done: "
318 "service %d, peer %s, status %s:%d:%d\n",
319 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
320 swi_state2str(rpc->crpc_wi.swi_state),
321 rpc->crpc_aborted, rpc->crpc_status);
323 spin_lock(&sfw_data.fw_lock);
325 /* my callers must finish all RPCs before shutting me down */
326 LASSERT(!sfw_data.fw_shuttingdown);
327 list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
329 spin_unlock(&sfw_data.fw_lock);
332 static struct sfw_batch *
333 sfw_find_batch(struct lst_bid bid)
335 struct sfw_session *sn = sfw_data.fw_session;
336 struct sfw_batch *bat;
340 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
341 if (bat->bat_id.bat_id == bid.bat_id)
348 static struct sfw_batch *
349 sfw_bid2batch(struct lst_bid bid)
351 struct sfw_session *sn = sfw_data.fw_session;
352 struct sfw_batch *bat;
354 LASSERT (sn != NULL);
356 bat = sfw_find_batch(bid);
360 LIBCFS_ALLOC(bat, sizeof(*bat));
365 bat->bat_session = sn;
367 atomic_set(&bat->bat_nactive, 0);
368 INIT_LIST_HEAD(&bat->bat_tests);
370 list_add_tail(&bat->bat_list, &sn->sn_batches);
375 sfw_get_stats(struct srpc_stat_reqst *request, struct srpc_stat_reply *reply)
377 struct sfw_session *sn = sfw_data.fw_session;
378 struct sfw_counters *cnt = &reply->str_fw;
379 struct sfw_batch *bat;
381 reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
383 if (request->str_sid.ses_nid == LNET_NID_ANY) {
384 reply->str_status = EINVAL;
388 if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
389 reply->str_status = ESRCH;
393 lnet_counters_get_common(&reply->str_lnet);
394 srpc_get_counters(&reply->str_rpc);
396 /* send over the msecs since the session was started
397 - with 32 bits to send, this is ~49 days */
398 cnt->running_ms = ktime_ms_delta(ktime_get(), sn->sn_started);
399 cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
400 cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
401 cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
403 cnt->active_batches = 0;
404 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
405 if (atomic_read(&bat->bat_nactive) > 0)
406 cnt->active_batches++;
409 reply->str_status = 0;
414 sfw_make_session(struct srpc_mksn_reqst *request, struct srpc_mksn_reply *reply)
416 struct sfw_session *sn = sfw_data.fw_session;
417 struct srpc_msg *msg = container_of(request, struct srpc_msg,
418 msg_body.mksn_reqst);
421 if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
422 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
423 reply->mksn_status = EINVAL;
428 reply->mksn_status = 0;
429 reply->mksn_sid = sn->sn_id;
430 reply->mksn_timeout = sn->sn_timeout;
432 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
433 atomic_inc(&sn->sn_refcount);
437 if (!request->mksn_force) {
438 reply->mksn_status = EBUSY;
439 cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
440 sizeof(reply->mksn_name));
441 if (cplen >= sizeof(reply->mksn_name))
447 /* reject the request if it requires unknown features
448 * NB: old version will always accept all features because it's not
449 * aware of struct srpc_msg::msg_ses_feats, it's a defect but it's also
450 * harmless because it will return zero feature to console, and it's
451 * console's responsibility to make sure all nodes in a session have
452 * same feature mask. */
453 if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
454 reply->mksn_status = EPROTO;
458 /* brand new or create by force */
459 LIBCFS_ALLOC(sn, sizeof(*sn));
461 CERROR("dropping RPC mksn under memory pressure\n");
465 sfw_init_session(sn, request->mksn_sid,
466 msg->msg_ses_feats, &request->mksn_name[0]);
468 spin_lock(&sfw_data.fw_lock);
470 sfw_deactivate_session();
471 LASSERT(sfw_data.fw_session == NULL);
472 sfw_data.fw_session = sn;
474 spin_unlock(&sfw_data.fw_lock);
476 reply->mksn_status = 0;
477 reply->mksn_sid = sn->sn_id;
478 reply->mksn_timeout = sn->sn_timeout;
483 sfw_remove_session(struct srpc_rmsn_reqst *request,
484 struct srpc_rmsn_reply *reply)
486 struct sfw_session *sn = sfw_data.fw_session;
488 reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
490 if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
491 reply->rmsn_status = EINVAL;
495 if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
496 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
500 if (!atomic_dec_and_test(&sn->sn_refcount)) {
501 reply->rmsn_status = 0;
505 spin_lock(&sfw_data.fw_lock);
506 sfw_deactivate_session();
507 spin_unlock(&sfw_data.fw_lock);
509 reply->rmsn_status = 0;
510 reply->rmsn_sid = LST_INVALID_SID;
511 LASSERT(sfw_data.fw_session == NULL);
516 sfw_debug_session(struct srpc_debug_reqst *request,
517 struct srpc_debug_reply *reply)
519 struct sfw_session *sn = sfw_data.fw_session;
522 reply->dbg_status = ESRCH;
523 reply->dbg_sid = LST_INVALID_SID;
527 reply->dbg_status = 0;
528 reply->dbg_sid = sn->sn_id;
529 reply->dbg_timeout = sn->sn_timeout;
530 if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
531 >= sizeof(reply->dbg_name))
538 sfw_test_rpc_fini(struct srpc_client_rpc *rpc)
540 struct sfw_test_unit *tsu = rpc->crpc_priv;
541 struct sfw_test_instance *tsi = tsu->tsu_instance;
543 /* Called with hold of tsi->tsi_lock */
544 LASSERT(list_empty(&rpc->crpc_list));
545 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
549 sfw_test_buffers(struct sfw_test_instance *tsi)
551 struct sfw_test_case *tsc;
552 struct srpc_service *svc;
555 LASSERT(tsi != NULL);
556 tsc = sfw_find_test_case(tsi->tsi_service);
557 LASSERT(tsc != NULL);
558 svc = tsc->tsc_srv_service;
559 LASSERT(svc != NULL);
561 nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
562 return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
566 sfw_load_test(struct sfw_test_instance *tsi)
568 struct sfw_test_case *tsc;
569 struct srpc_service *svc;
573 LASSERT(tsi != NULL);
574 tsc = sfw_find_test_case(tsi->tsi_service);
575 nbuf = sfw_test_buffers(tsi);
576 LASSERT(tsc != NULL);
577 svc = tsc->tsc_srv_service;
579 if (tsi->tsi_is_client) {
580 tsi->tsi_ops = tsc->tsc_cli_ops;
584 rc = srpc_service_add_buffers(svc, nbuf);
586 CWARN("Failed to reserve enough buffers: "
587 "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
588 /* NB: this error handler is not strictly correct, because
589 * it may release more buffers than already allocated,
590 * but it doesn't matter because request portal should
591 * be lazy portal and will grow buffers if necessary. */
592 srpc_service_remove_buffers(svc, nbuf);
596 CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
597 nbuf * (srpc_serv_is_framework(svc) ?
598 1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
603 sfw_unload_test(struct sfw_test_instance *tsi)
605 struct sfw_test_case *tsc;
607 LASSERT(tsi != NULL);
608 tsc = sfw_find_test_case(tsi->tsi_service);
609 LASSERT(tsc != NULL);
611 if (tsi->tsi_is_client)
614 /* shrink buffers, because request portal is lazy portal
615 * which can grow buffers at runtime so we may leave
616 * some buffers behind, but never mind... */
617 srpc_service_remove_buffers(tsc->tsc_srv_service,
618 sfw_test_buffers(tsi));
623 sfw_destroy_test_instance(struct sfw_test_instance *tsi)
625 struct srpc_client_rpc *rpc;
626 struct sfw_test_unit *tsu;
628 if (!tsi->tsi_is_client) goto clean;
630 tsi->tsi_ops->tso_fini(tsi);
632 LASSERT(!tsi->tsi_stopping);
633 LASSERT(list_empty(&tsi->tsi_active_rpcs));
634 LASSERT(!sfw_test_active(tsi));
636 while (!list_empty(&tsi->tsi_units)) {
637 tsu = list_entry(tsi->tsi_units.next,
638 struct sfw_test_unit, tsu_list);
639 list_del(&tsu->tsu_list);
640 LIBCFS_FREE(tsu, sizeof(*tsu));
643 while (!list_empty(&tsi->tsi_free_rpcs)) {
644 rpc = list_entry(tsi->tsi_free_rpcs.next,
645 struct srpc_client_rpc, crpc_list);
646 list_del(&rpc->crpc_list);
647 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
651 sfw_unload_test(tsi);
652 LIBCFS_FREE(tsi, sizeof(*tsi));
657 sfw_destroy_batch(struct sfw_batch *tsb)
659 struct sfw_test_instance *tsi;
661 LASSERT(!sfw_batch_active(tsb));
662 LASSERT(list_empty(&tsb->bat_list));
664 while (!list_empty(&tsb->bat_tests)) {
665 tsi = list_entry(tsb->bat_tests.next,
666 struct sfw_test_instance, tsi_list);
667 list_del_init(&tsi->tsi_list);
668 sfw_destroy_test_instance(tsi);
671 LIBCFS_FREE(tsb, sizeof(*tsb));
676 sfw_destroy_session(struct sfw_session *sn)
678 struct sfw_batch *batch;
680 LASSERT(list_empty(&sn->sn_list));
681 LASSERT(sn != sfw_data.fw_session);
683 while (!list_empty(&sn->sn_batches)) {
684 batch = list_entry(sn->sn_batches.next,
685 struct sfw_batch, bat_list);
686 list_del_init(&batch->bat_list);
687 sfw_destroy_batch(batch);
690 LIBCFS_FREE(sn, sizeof(*sn));
691 atomic_dec(&sfw_data.fw_nzombies);
696 sfw_unpack_addtest_req(struct srpc_msg *msg)
698 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
700 LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
701 LASSERT (req->tsr_is_client);
703 if (msg->msg_magic == SRPC_MSG_MAGIC)
704 return; /* no flipping needed */
706 LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
708 if (req->tsr_service == SRPC_SERVICE_BRW) {
709 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
710 struct test_bulk_req *bulk = &req->tsr_u.bulk_v0;
712 __swab32s(&bulk->blk_opc);
713 __swab32s(&bulk->blk_npg);
714 __swab32s(&bulk->blk_flags);
717 struct test_bulk_req_v1 *bulk = &req->tsr_u.bulk_v1;
719 __swab16s(&bulk->blk_opc);
720 __swab16s(&bulk->blk_flags);
721 __swab32s(&bulk->blk_offset);
722 __swab32s(&bulk->blk_len);
728 if (req->tsr_service == SRPC_SERVICE_PING) {
729 struct test_ping_req *ping = &req->tsr_u.ping;
731 __swab32s(&ping->png_size);
732 __swab32s(&ping->png_flags);
741 sfw_add_test_instance(struct sfw_batch *tsb, struct srpc_server_rpc *rpc)
743 struct srpc_msg *msg = &rpc->srpc_reqstbuf->buf_msg;
744 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
745 struct srpc_bulk *bk = rpc->srpc_bulk;
746 int ndest = req->tsr_ndest;
747 struct sfw_test_unit *tsu;
748 struct sfw_test_instance *tsi;
752 LIBCFS_ALLOC(tsi, sizeof(*tsi));
754 CERROR ("Can't allocate test instance for batch: %llu\n",
759 spin_lock_init(&tsi->tsi_lock);
760 atomic_set(&tsi->tsi_nactive, 0);
761 INIT_LIST_HEAD(&tsi->tsi_units);
762 INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
763 INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
765 tsi->tsi_stopping = 0;
766 tsi->tsi_batch = tsb;
767 tsi->tsi_loop = req->tsr_loop;
768 tsi->tsi_concur = req->tsr_concur;
769 tsi->tsi_service = req->tsr_service;
770 tsi->tsi_is_client = !!(req->tsr_is_client);
771 tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
773 rc = sfw_load_test(tsi);
775 LIBCFS_FREE(tsi, sizeof(*tsi));
779 LASSERT (!sfw_batch_active(tsb));
781 if (!tsi->tsi_is_client) {
782 /* it's test server, just add it to tsb */
783 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
787 LASSERT (bk != NULL);
788 LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
789 LASSERT((unsigned int)bk->bk_len >=
790 sizeof(struct lnet_process_id_packed) * ndest);
792 sfw_unpack_addtest_req(msg);
793 memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
795 for (i = 0; i < ndest; i++) {
796 struct lnet_process_id_packed *dests;
797 struct lnet_process_id_packed id;
800 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
801 LASSERT (dests != NULL); /* my pages are within KVM always */
802 id = dests[i % SFW_ID_PER_PAGE];
803 if (msg->msg_magic != SRPC_MSG_MAGIC)
806 for (j = 0; j < tsi->tsi_concur; j++) {
807 LIBCFS_ALLOC(tsu, sizeof(*tsu));
810 CERROR ("Can't allocate tsu for %d\n",
815 tsu->tsu_dest.nid = id.nid;
816 tsu->tsu_dest.pid = id.pid;
817 tsu->tsu_instance = tsi;
818 tsu->tsu_private = NULL;
819 list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
823 rc = tsi->tsi_ops->tso_init(tsi);
825 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
831 sfw_destroy_test_instance(tsi);
836 sfw_test_unit_done(struct sfw_test_unit *tsu)
838 struct sfw_test_instance *tsi = tsu->tsu_instance;
839 struct sfw_batch *tsb = tsi->tsi_batch;
840 struct sfw_session *sn = tsb->bat_session;
842 LASSERT (sfw_test_active(tsi));
844 if (!atomic_dec_and_test(&tsi->tsi_nactive))
847 /* the test instance is done */
848 spin_lock(&tsi->tsi_lock);
850 tsi->tsi_stopping = 0;
852 spin_unlock(&tsi->tsi_lock);
854 spin_lock(&sfw_data.fw_lock);
856 if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
857 sn == sfw_data.fw_session) { /* sn also active */
858 spin_unlock(&sfw_data.fw_lock);
862 LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
864 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
865 if (sfw_batch_active(tsb)) {
866 spin_unlock(&sfw_data.fw_lock);
871 list_del_init(&sn->sn_list);
872 spin_unlock(&sfw_data.fw_lock);
874 sfw_destroy_session(sn);
879 sfw_test_rpc_done(struct srpc_client_rpc *rpc)
881 struct sfw_test_unit *tsu = rpc->crpc_priv;
882 struct sfw_test_instance *tsi = tsu->tsu_instance;
885 tsi->tsi_ops->tso_done_rpc(tsu, rpc);
887 spin_lock(&tsi->tsi_lock);
889 LASSERT(sfw_test_active(tsi));
890 LASSERT(!list_empty(&rpc->crpc_list));
892 list_del_init(&rpc->crpc_list);
894 /* batch is stopping or loop is done or get error */
895 if (tsi->tsi_stopping ||
896 tsu->tsu_loop == 0 ||
897 (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
900 /* dec ref for poster */
901 srpc_client_rpc_decref(rpc);
903 spin_unlock(&tsi->tsi_lock);
906 swi_schedule_workitem(&tsu->tsu_worker);
910 sfw_test_unit_done(tsu);
915 sfw_create_test_rpc(struct sfw_test_unit *tsu, struct lnet_process_id peer,
916 unsigned features, int nblk, int blklen,
917 struct srpc_client_rpc **rpcpp)
919 struct srpc_client_rpc *rpc = NULL;
920 struct sfw_test_instance *tsi = tsu->tsu_instance;
922 spin_lock(&tsi->tsi_lock);
924 LASSERT (sfw_test_active(tsi));
926 if (!list_empty(&tsi->tsi_free_rpcs)) {
927 /* pick request from buffer */
928 rpc = list_entry(tsi->tsi_free_rpcs.next,
929 struct srpc_client_rpc, crpc_list);
930 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
931 list_del_init(&rpc->crpc_list);
934 spin_unlock(&tsi->tsi_lock);
937 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
938 blklen, sfw_test_rpc_done,
939 sfw_test_rpc_fini, tsu);
941 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
942 blklen, sfw_test_rpc_done,
943 sfw_test_rpc_fini, tsu);
947 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
951 rpc->crpc_reqstmsg.msg_ses_feats = features;
958 sfw_run_test(struct swi_workitem *wi)
960 struct sfw_test_unit *tsu = container_of(wi, struct sfw_test_unit, tsu_worker);
961 struct sfw_test_instance *tsi = tsu->tsu_instance;
962 struct srpc_client_rpc *rpc = NULL;
964 LASSERT (wi == &tsu->tsu_worker);
966 if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
967 LASSERT (rpc == NULL);
971 LASSERT (rpc != NULL);
973 spin_lock(&tsi->tsi_lock);
975 if (tsi->tsi_stopping) {
976 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
977 spin_unlock(&tsi->tsi_lock);
981 if (tsu->tsu_loop > 0)
984 list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
985 spin_unlock(&tsi->tsi_lock);
987 spin_lock(&rpc->crpc_lock);
988 rpc->crpc_timeout = rpc_timeout;
990 spin_unlock(&rpc->crpc_lock);
995 * No one can schedule me now since:
996 * - previous RPC, if any, has done and
997 * - no new RPC is initiated.
998 * - my batch is still active; no one can run it again now.
999 * Cancel pending schedules and prevent future schedule attempts:
1001 swi_exit_workitem(wi);
1002 sfw_test_unit_done(tsu);
1007 sfw_run_batch(struct sfw_batch *tsb)
1009 struct swi_workitem *wi;
1010 struct sfw_test_unit *tsu;
1011 struct sfw_test_instance *tsi;
1013 if (sfw_batch_active(tsb)) {
1014 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1015 tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1019 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1020 if (!tsi->tsi_is_client) /* skip server instances */
1023 LASSERT(!tsi->tsi_stopping);
1024 LASSERT(!sfw_test_active(tsi));
1026 atomic_inc(&tsb->bat_nactive);
1028 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1029 atomic_inc(&tsi->tsi_nactive);
1030 tsu->tsu_loop = tsi->tsi_loop;
1031 wi = &tsu->tsu_worker;
1032 swi_init_workitem(wi, sfw_run_test,
1033 lst_sched_test[lnet_cpt_of_nid(tsu->tsu_dest.nid, NULL)]);
1034 swi_schedule_workitem(wi);
1042 sfw_stop_batch(struct sfw_batch *tsb, int force)
1044 struct sfw_test_instance *tsi;
1045 struct srpc_client_rpc *rpc;
1047 if (!sfw_batch_active(tsb)) {
1048 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1052 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1053 spin_lock(&tsi->tsi_lock);
1055 if (!tsi->tsi_is_client ||
1056 !sfw_test_active(tsi) || tsi->tsi_stopping) {
1057 spin_unlock(&tsi->tsi_lock);
1061 tsi->tsi_stopping = 1;
1064 spin_unlock(&tsi->tsi_lock);
1068 /* abort launched rpcs in the test */
1069 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1070 spin_lock(&rpc->crpc_lock);
1072 srpc_abort_rpc(rpc, -EINTR);
1074 spin_unlock(&rpc->crpc_lock);
1077 spin_unlock(&tsi->tsi_lock);
1084 sfw_query_batch(struct sfw_batch *tsb, int testidx,
1085 struct srpc_batch_reply *reply)
1087 struct sfw_test_instance *tsi;
1093 reply->bar_active = atomic_read(&tsb->bat_nactive);
1097 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1101 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1109 sfw_free_pages(struct srpc_server_rpc *rpc)
1111 srpc_free_bulk(rpc->srpc_bulk);
1112 rpc->srpc_bulk = NULL;
1116 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1119 LASSERT(rpc->srpc_bulk == NULL);
1120 LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1122 rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1123 if (rpc->srpc_bulk == NULL)
1130 sfw_add_test(struct srpc_server_rpc *rpc)
1132 struct sfw_session *sn = sfw_data.fw_session;
1133 struct srpc_test_reply *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1134 struct srpc_test_reqst *request;
1136 struct sfw_batch *bat;
1138 request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1139 reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1141 if (request->tsr_loop == 0 ||
1142 request->tsr_concur == 0 ||
1143 request->tsr_sid.ses_nid == LNET_NID_ANY ||
1144 request->tsr_ndest > SFW_MAX_NDESTS ||
1145 (request->tsr_is_client && request->tsr_ndest == 0) ||
1146 request->tsr_concur > SFW_MAX_CONCUR ||
1147 request->tsr_service > SRPC_SERVICE_MAX_ID ||
1148 request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1149 reply->tsr_status = EINVAL;
1153 if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1154 sfw_find_test_case(request->tsr_service) == NULL) {
1155 reply->tsr_status = ENOENT;
1159 bat = sfw_bid2batch(request->tsr_bid);
1161 CERROR("dropping RPC %s from %s under memory pressure\n",
1162 rpc->srpc_scd->scd_svc->sv_name,
1163 libcfs_id2str(rpc->srpc_peer));
1167 if (sfw_batch_active(bat)) {
1168 reply->tsr_status = EBUSY;
1172 if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1173 /* rpc will be resumed later in sfw_bulk_ready */
1174 int npg = sfw_id_pages(request->tsr_ndest);
1177 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1178 len = npg * PAGE_SIZE;
1181 len = sizeof(struct lnet_process_id_packed) *
1185 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1188 rc = sfw_add_test_instance(bat, rpc);
1189 CDEBUG (rc == 0 ? D_NET : D_WARNING,
1190 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1191 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1192 request->tsr_is_client ? "client" : "server",
1193 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1195 reply->tsr_status = (rc < 0) ? -rc : rc;
1200 sfw_control_batch(struct srpc_batch_reqst *request,
1201 struct srpc_batch_reply *reply)
1203 struct sfw_session *sn = sfw_data.fw_session;
1205 struct sfw_batch *bat;
1207 reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1209 if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1210 reply->bar_status = ESRCH;
1214 bat = sfw_find_batch(request->bar_bid);
1216 reply->bar_status = ENOENT;
1220 switch (request->bar_opc) {
1221 case SRPC_BATCH_OPC_RUN:
1222 rc = sfw_run_batch(bat);
1225 case SRPC_BATCH_OPC_STOP:
1226 rc = sfw_stop_batch(bat, request->bar_arg);
1229 case SRPC_BATCH_OPC_QUERY:
1230 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1234 return -EINVAL; /* drop it */
1237 reply->bar_status = (rc < 0) ? -rc : rc;
1242 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1244 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1245 struct srpc_msg *reply = &rpc->srpc_replymsg;
1246 struct srpc_msg *request = &rpc->srpc_reqstbuf->buf_msg;
1247 unsigned features = LST_FEATS_MASK;
1250 LASSERT(sfw_data.fw_active_srpc == NULL);
1251 LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1253 spin_lock(&sfw_data.fw_lock);
1255 if (sfw_data.fw_shuttingdown) {
1256 spin_unlock(&sfw_data.fw_lock);
1260 /* Remove timer to avoid racing with it or expiring active session */
1261 if (sfw_del_session_timer() != 0) {
1262 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1263 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1264 spin_unlock(&sfw_data.fw_lock);
1268 sfw_data.fw_active_srpc = rpc;
1269 spin_unlock(&sfw_data.fw_lock);
1271 sfw_unpack_message(request);
1272 LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1274 /* rpc module should have checked this */
1275 LASSERT(request->msg_version == SRPC_MSG_VERSION);
1277 if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1278 sv->sv_id != SRPC_SERVICE_DEBUG) {
1279 struct sfw_session *sn = sfw_data.fw_session;
1282 sn->sn_features != request->msg_ses_feats) {
1283 CNETERR("Features of framework RPC don't match "
1284 "features of current session: %x/%x\n",
1285 request->msg_ses_feats, sn->sn_features);
1286 reply->msg_body.reply.status = EPROTO;
1287 reply->msg_body.reply.sid = sn->sn_id;
1291 } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1292 /* NB: at this point, old version will ignore features and
1293 * create new session anyway, so console should be able
1295 reply->msg_body.reply.status = EPROTO;
1302 case SRPC_SERVICE_TEST:
1303 rc = sfw_add_test(rpc);
1306 case SRPC_SERVICE_BATCH:
1307 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1308 &reply->msg_body.bat_reply);
1311 case SRPC_SERVICE_QUERY_STAT:
1312 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1313 &reply->msg_body.stat_reply);
1316 case SRPC_SERVICE_DEBUG:
1317 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1318 &reply->msg_body.dbg_reply);
1321 case SRPC_SERVICE_MAKE_SESSION:
1322 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1323 &reply->msg_body.mksn_reply);
1326 case SRPC_SERVICE_REMOVE_SESSION:
1327 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1328 &reply->msg_body.rmsn_reply);
1332 if (sfw_data.fw_session != NULL)
1333 features = sfw_data.fw_session->sn_features;
1335 reply->msg_ses_feats = features;
1336 rpc->srpc_done = sfw_server_rpc_done;
1337 spin_lock(&sfw_data.fw_lock);
1339 if (!sfw_data.fw_shuttingdown)
1340 sfw_add_session_timer();
1342 sfw_data.fw_active_srpc = NULL;
1343 spin_unlock(&sfw_data.fw_lock);
1348 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1350 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1353 LASSERT(rpc->srpc_bulk != NULL);
1354 LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1355 LASSERT(sfw_data.fw_active_srpc == NULL);
1356 LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1358 spin_lock(&sfw_data.fw_lock);
1361 CERROR("Bulk transfer failed for RPC: "
1362 "service %s, peer %s, status %d\n",
1363 sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1364 spin_unlock(&sfw_data.fw_lock);
1368 if (sfw_data.fw_shuttingdown) {
1369 spin_unlock(&sfw_data.fw_lock);
1373 if (sfw_del_session_timer() != 0) {
1374 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1375 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1376 spin_unlock(&sfw_data.fw_lock);
1380 sfw_data.fw_active_srpc = rpc;
1381 spin_unlock(&sfw_data.fw_lock);
1383 rc = sfw_add_test(rpc);
1385 spin_lock(&sfw_data.fw_lock);
1387 if (!sfw_data.fw_shuttingdown)
1388 sfw_add_session_timer();
1390 sfw_data.fw_active_srpc = NULL;
1391 spin_unlock(&sfw_data.fw_lock);
1395 struct srpc_client_rpc *
1396 sfw_create_rpc(struct lnet_process_id peer, int service,
1397 unsigned features, int nbulkiov, int bulklen,
1398 void (*done)(struct srpc_client_rpc *), void *priv)
1400 struct srpc_client_rpc *rpc = NULL;
1402 spin_lock(&sfw_data.fw_lock);
1404 LASSERT (!sfw_data.fw_shuttingdown);
1405 LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1407 if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1408 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1409 struct srpc_client_rpc, crpc_list);
1410 list_del(&rpc->crpc_list);
1412 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1413 done, sfw_client_rpc_fini, priv);
1416 spin_unlock(&sfw_data.fw_lock);
1419 rpc = srpc_create_client_rpc(peer, service,
1420 nbulkiov, bulklen, done,
1421 nbulkiov != 0 ? NULL :
1422 sfw_client_rpc_fini,
1426 if (rpc != NULL) /* "session" is concept in framework */
1427 rpc->crpc_reqstmsg.msg_ses_feats = features;
1433 sfw_unpack_message(struct srpc_msg *msg)
1435 if (msg->msg_magic == SRPC_MSG_MAGIC)
1436 return; /* no flipping needed */
1438 /* srpc module should guarantee I wouldn't get crap */
1439 LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1441 if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1442 struct srpc_stat_reqst *req = &msg->msg_body.stat_reqst;
1444 __swab32s(&req->str_type);
1445 __swab64s(&req->str_rpyid);
1446 sfw_unpack_sid(req->str_sid);
1450 if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1451 struct srpc_stat_reply *rep = &msg->msg_body.stat_reply;
1453 __swab32s(&rep->str_status);
1454 sfw_unpack_sid(rep->str_sid);
1455 sfw_unpack_fw_counters(rep->str_fw);
1456 sfw_unpack_rpc_counters(rep->str_rpc);
1457 sfw_unpack_lnet_counters(rep->str_lnet);
1461 if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1462 struct srpc_mksn_reqst *req = &msg->msg_body.mksn_reqst;
1464 __swab64s(&req->mksn_rpyid);
1465 __swab32s(&req->mksn_force);
1466 sfw_unpack_sid(req->mksn_sid);
1470 if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1471 struct srpc_mksn_reply *rep = &msg->msg_body.mksn_reply;
1473 __swab32s(&rep->mksn_status);
1474 __swab32s(&rep->mksn_timeout);
1475 sfw_unpack_sid(rep->mksn_sid);
1479 if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1480 struct srpc_rmsn_reqst *req = &msg->msg_body.rmsn_reqst;
1482 __swab64s(&req->rmsn_rpyid);
1483 sfw_unpack_sid(req->rmsn_sid);
1487 if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1488 struct srpc_rmsn_reply *rep = &msg->msg_body.rmsn_reply;
1490 __swab32s(&rep->rmsn_status);
1491 sfw_unpack_sid(rep->rmsn_sid);
1495 if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1496 struct srpc_debug_reqst *req = &msg->msg_body.dbg_reqst;
1498 __swab64s(&req->dbg_rpyid);
1499 __swab32s(&req->dbg_flags);
1500 sfw_unpack_sid(req->dbg_sid);
1504 if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1505 struct srpc_debug_reply *rep = &msg->msg_body.dbg_reply;
1507 __swab32s(&rep->dbg_nbatch);
1508 __swab32s(&rep->dbg_timeout);
1509 sfw_unpack_sid(rep->dbg_sid);
1513 if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1514 struct srpc_batch_reqst *req = &msg->msg_body.bat_reqst;
1516 __swab32s(&req->bar_opc);
1517 __swab64s(&req->bar_rpyid);
1518 __swab32s(&req->bar_testidx);
1519 __swab32s(&req->bar_arg);
1520 sfw_unpack_sid(req->bar_sid);
1521 __swab64s(&req->bar_bid.bat_id);
1525 if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1526 struct srpc_batch_reply *rep = &msg->msg_body.bat_reply;
1528 __swab32s(&rep->bar_status);
1529 sfw_unpack_sid(rep->bar_sid);
1533 if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1534 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
1536 __swab64s(&req->tsr_rpyid);
1537 __swab64s(&req->tsr_bulkid);
1538 __swab32s(&req->tsr_loop);
1539 __swab32s(&req->tsr_ndest);
1540 __swab32s(&req->tsr_concur);
1541 __swab32s(&req->tsr_service);
1542 sfw_unpack_sid(req->tsr_sid);
1543 __swab64s(&req->tsr_bid.bat_id);
1547 if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1548 struct srpc_test_reply *rep = &msg->msg_body.tes_reply;
1550 __swab32s(&rep->tsr_status);
1551 sfw_unpack_sid(rep->tsr_sid);
1555 if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1556 struct srpc_join_reqst *req = &msg->msg_body.join_reqst;
1558 __swab64s(&req->join_rpyid);
1559 sfw_unpack_sid(req->join_sid);
1563 if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1564 struct srpc_join_reply *rep = &msg->msg_body.join_reply;
1566 __swab32s(&rep->join_status);
1567 __swab32s(&rep->join_timeout);
1568 sfw_unpack_sid(rep->join_sid);
1577 sfw_abort_rpc(struct srpc_client_rpc *rpc)
1579 LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1580 LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1582 spin_lock(&rpc->crpc_lock);
1583 srpc_abort_rpc(rpc, -EINTR);
1584 spin_unlock(&rpc->crpc_lock);
1589 sfw_post_rpc(struct srpc_client_rpc *rpc)
1591 spin_lock(&rpc->crpc_lock);
1593 LASSERT(!rpc->crpc_closed);
1594 LASSERT(!rpc->crpc_aborted);
1595 LASSERT(list_empty(&rpc->crpc_list));
1596 LASSERT(!sfw_data.fw_shuttingdown);
1598 rpc->crpc_timeout = rpc_timeout;
1601 spin_unlock(&rpc->crpc_lock);
1605 static struct srpc_service sfw_services[] = {
1606 { .sv_id = SRPC_SERVICE_DEBUG, .sv_name = "debug", },
1607 { .sv_id = SRPC_SERVICE_QUERY_STAT, .sv_name = "query stats", },
1608 { .sv_id = SRPC_SERVICE_MAKE_SESSION, .sv_name = "make session", },
1609 { .sv_id = SRPC_SERVICE_REMOVE_SESSION, .sv_name = "remove session", },
1610 { .sv_id = SRPC_SERVICE_BATCH, .sv_name = "batch service", },
1611 { .sv_id = SRPC_SERVICE_TEST, .sv_name = "test service", },
1620 struct srpc_service *sv;
1621 struct sfw_test_case *tsc;
1624 if (session_timeout < 0) {
1625 CERROR ("Session timeout must be non-negative: %d\n",
1630 if (rpc_timeout < 0) {
1631 CERROR ("RPC timeout must be non-negative: %d\n",
1636 if (session_timeout == 0)
1637 CWARN ("Zero session_timeout specified "
1638 "- test sessions never expire.\n");
1640 if (rpc_timeout == 0)
1641 CWARN ("Zero rpc_timeout specified "
1642 "- test RPC never expire.\n");
1644 memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1646 sfw_data.fw_session = NULL;
1647 sfw_data.fw_active_srpc = NULL;
1648 spin_lock_init(&sfw_data.fw_lock);
1649 atomic_set(&sfw_data.fw_nzombies, 0);
1650 INIT_LIST_HEAD(&sfw_data.fw_tests);
1651 INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1652 INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1654 brw_init_test_client();
1655 brw_init_test_service();
1656 rc = sfw_register_test(&brw_test_service, &brw_test_client);
1659 ping_init_test_client();
1660 ping_init_test_service();
1661 rc = sfw_register_test(&ping_test_service, &ping_test_client);
1665 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1666 sv = tsc->tsc_srv_service;
1668 rc = srpc_add_service(sv);
1669 LASSERT(rc != -EBUSY);
1671 CWARN("Failed to add %s service: %d\n",
1677 for (i = 0; ; i++) {
1678 sv = &sfw_services[i];
1679 if (sv->sv_name == NULL) break;
1681 sv->sv_bulk_ready = NULL;
1682 sv->sv_handler = sfw_handle_server_rpc;
1683 sv->sv_wi_total = SFW_FRWK_WI_MAX;
1684 if (sv->sv_id == SRPC_SERVICE_TEST)
1685 sv->sv_bulk_ready = sfw_bulk_ready;
1687 rc = srpc_add_service(sv);
1688 LASSERT (rc != -EBUSY);
1690 CWARN ("Failed to add %s service: %d\n",
1695 /* about to sfw_shutdown, no need to add buffer */
1696 if (error) continue;
1698 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1700 CWARN("Failed to reserve enough buffers: "
1701 "service %s, %d needed: %d\n",
1702 sv->sv_name, sv->sv_wi_total, rc);
1715 struct srpc_service *sv;
1716 struct sfw_test_case *tsc;
1719 spin_lock(&sfw_data.fw_lock);
1721 sfw_data.fw_shuttingdown = 1;
1722 lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1723 "waiting for active RPC to finish.\n");
1725 if (sfw_del_session_timer() != 0)
1726 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1727 "waiting for session timer to explode.\n");
1729 sfw_deactivate_session();
1730 lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1732 "waiting for %d zombie sessions to die.\n",
1733 atomic_read(&sfw_data.fw_nzombies));
1735 spin_unlock(&sfw_data.fw_lock);
1737 for (i = 0; ; i++) {
1738 sv = &sfw_services[i];
1739 if (sv->sv_name == NULL)
1742 srpc_shutdown_service(sv);
1743 srpc_remove_service(sv);
1746 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1747 sv = tsc->tsc_srv_service;
1748 srpc_shutdown_service(sv);
1749 srpc_remove_service(sv);
1752 while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1753 struct srpc_client_rpc *rpc;
1755 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1756 struct srpc_client_rpc, crpc_list);
1757 list_del(&rpc->crpc_list);
1759 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1762 for (i = 0; ; i++) {
1763 sv = &sfw_services[i];
1764 if (sv->sv_name == NULL)
1767 srpc_wait_service_shutdown(sv);
1770 while (!list_empty(&sfw_data.fw_tests)) {
1771 tsc = list_entry(sfw_data.fw_tests.next,
1772 struct sfw_test_case, tsc_list);
1774 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1776 list_del(&tsc->tsc_list);
1777 LIBCFS_FREE(tsc, sizeof(*tsc));