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/
31 * lnet/selftest/framework.c
33 * Author: Isaac Huang <isaac@clusterfs.com>
34 * Author: Liang Zhen <liangzhen@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_LNET
41 struct lst_sid LST_INVALID_SID = { .ses_nid = LNET_NID_ANY, .ses_stamp = -1};
43 static int session_timeout = 100;
44 module_param(session_timeout, int, 0444);
45 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
47 static int rpc_timeout = 64;
48 module_param(rpc_timeout, int, 0644);
49 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
51 #define sfw_unpack_id(id) \
53 __swab64s(&(id).nid); \
54 __swab32s(&(id).pid); \
57 #define sfw_unpack_sid(sid) \
59 __swab64s(&(sid).ses_nid); \
60 __swab64s(&(sid).ses_stamp); \
63 #define sfw_unpack_fw_counters(fc) \
65 __swab32s(&(fc).running_ms); \
66 __swab32s(&(fc).active_batches); \
67 __swab32s(&(fc).zombie_sessions); \
68 __swab32s(&(fc).brw_errors); \
69 __swab32s(&(fc).ping_errors); \
72 #define sfw_unpack_rpc_counters(rc) \
74 __swab32s(&(rc).errors); \
75 __swab32s(&(rc).rpcs_sent); \
76 __swab32s(&(rc).rpcs_rcvd); \
77 __swab32s(&(rc).rpcs_dropped); \
78 __swab32s(&(rc).rpcs_expired); \
79 __swab64s(&(rc).bulk_get); \
80 __swab64s(&(rc).bulk_put); \
83 #define sfw_unpack_lnet_counters(lc) \
85 __swab32s(&(lc).lcc_errors); \
86 __swab32s(&(lc).lcc_msgs_max); \
87 __swab32s(&(lc).lcc_msgs_alloc); \
88 __swab32s(&(lc).lcc_send_count); \
89 __swab32s(&(lc).lcc_recv_count); \
90 __swab32s(&(lc).lcc_drop_count); \
91 __swab32s(&(lc).lcc_route_count); \
92 __swab64s(&(lc).lcc_send_length); \
93 __swab64s(&(lc).lcc_recv_length); \
94 __swab64s(&(lc).lcc_drop_length); \
95 __swab64s(&(lc).lcc_route_length); \
98 #define sfw_test_active(t) (atomic_read(&(t)->tsi_nactive) != 0)
99 #define sfw_batch_active(b) (atomic_read(&(b)->bat_nactive) != 0)
101 static struct smoketest_framework {
102 /* RPCs to be recycled */
103 struct list_head fw_zombie_rpcs;
104 /* stopping sessions */
105 struct list_head fw_zombie_sessions;
106 /* registered test cases */
107 struct list_head fw_tests;
108 /* # zombie sessions */
109 atomic_t fw_nzombies;
113 struct sfw_session *fw_session;
114 /* shutdown in progress */
117 struct srpc_server_rpc *fw_active_srpc;
121 static int sfw_stop_batch(struct sfw_batch *tsb, int force);
122 static void sfw_destroy_session(struct sfw_session *sn);
124 static inline struct sfw_test_case *
125 sfw_find_test_case(int id)
127 struct sfw_test_case *tsc;
129 LASSERT(id <= SRPC_SERVICE_MAX_ID);
130 LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
132 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
133 if (tsc->tsc_srv_service->sv_id == id)
141 sfw_register_test(struct srpc_service *service,
142 struct sfw_test_client_ops *cliops)
144 struct sfw_test_case *tsc;
146 if (sfw_find_test_case(service->sv_id) != NULL) {
147 CERROR ("Failed to register test %s (%d)\n",
148 service->sv_name, service->sv_id);
152 LIBCFS_ALLOC(tsc, sizeof(*tsc));
156 tsc->tsc_cli_ops = cliops;
157 tsc->tsc_srv_service = service;
159 list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
164 sfw_add_session_timer (void)
166 struct sfw_session *sn = sfw_data.fw_session;
167 struct stt_timer *timer = &sn->sn_timer;
169 LASSERT (!sfw_data.fw_shuttingdown);
171 if (sn == NULL || sn->sn_timeout == 0)
174 LASSERT (!sn->sn_timer_active);
176 sn->sn_timer_active = 1;
177 timer->stt_expires = ktime_get_real_seconds()+ sn->sn_timeout;
178 stt_add_timer(timer);
182 sfw_del_session_timer (void)
184 struct sfw_session *sn = sfw_data.fw_session;
186 if (sn == NULL || !sn->sn_timer_active)
189 LASSERT (sn->sn_timeout != 0);
191 if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
192 sn->sn_timer_active = 0;
196 return EBUSY; /* racing with sfw_session_expired() */
199 /* called with sfw_data.fw_lock held */
201 sfw_deactivate_session (void)
202 __must_hold(&sfw_data.fw_lock)
204 struct sfw_session *sn = sfw_data.fw_session;
206 struct sfw_batch *tsb;
207 struct sfw_test_case *tsc;
209 if (sn == NULL) return;
211 LASSERT(!sn->sn_timer_active);
213 sfw_data.fw_session = NULL;
214 atomic_inc(&sfw_data.fw_nzombies);
215 list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
217 spin_unlock(&sfw_data.fw_lock);
219 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
220 srpc_abort_service(tsc->tsc_srv_service);
223 spin_lock(&sfw_data.fw_lock);
225 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
226 if (sfw_batch_active(tsb)) {
228 sfw_stop_batch(tsb, 1);
233 return; /* wait for active batches to stop */
235 list_del_init(&sn->sn_list);
236 spin_unlock(&sfw_data.fw_lock);
238 sfw_destroy_session(sn);
240 spin_lock(&sfw_data.fw_lock);
245 sfw_session_expired (void *data)
247 struct sfw_session *sn = data;
249 spin_lock(&sfw_data.fw_lock);
251 LASSERT (sn->sn_timer_active);
252 LASSERT (sn == sfw_data.fw_session);
254 CWARN ("Session expired! sid: %s-%llu, name: %s\n",
255 libcfs_nid2str(sn->sn_id.ses_nid),
256 sn->sn_id.ses_stamp, &sn->sn_name[0]);
258 sn->sn_timer_active = 0;
259 sfw_deactivate_session();
261 spin_unlock(&sfw_data.fw_lock);
265 sfw_init_session(struct sfw_session *sn, struct lst_sid sid,
266 unsigned features, const char *name)
268 struct stt_timer *timer = &sn->sn_timer;
270 memset(sn, 0, sizeof(struct sfw_session));
271 INIT_LIST_HEAD(&sn->sn_list);
272 INIT_LIST_HEAD(&sn->sn_batches);
273 atomic_set(&sn->sn_refcount, 1); /* +1 for caller */
274 atomic_set(&sn->sn_brw_errors, 0);
275 atomic_set(&sn->sn_ping_errors, 0);
276 strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
278 sn->sn_timer_active = 0;
280 sn->sn_features = features;
281 sn->sn_timeout = session_timeout;
282 sn->sn_started = ktime_get();
284 timer->stt_data = sn;
285 timer->stt_func = sfw_session_expired;
286 INIT_LIST_HEAD(&timer->stt_list);
289 /* completion handler for incoming framework RPCs */
291 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
293 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
294 int status = rpc->srpc_status;
297 "Incoming framework RPC done: "
298 "service %s, peer %s, status %s:%d\n",
299 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
300 swi_state2str(rpc->srpc_wi.swi_state),
303 if (rpc->srpc_bulk != NULL)
308 sfw_client_rpc_fini(struct srpc_client_rpc *rpc)
310 LASSERT(rpc->crpc_bulk.bk_niov == 0);
311 LASSERT(list_empty(&rpc->crpc_list));
312 LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
314 CDEBUG(D_NET, "Outgoing framework RPC done: "
315 "service %d, peer %s, status %s:%d:%d\n",
316 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
317 swi_state2str(rpc->crpc_wi.swi_state),
318 rpc->crpc_aborted, rpc->crpc_status);
320 spin_lock(&sfw_data.fw_lock);
322 /* my callers must finish all RPCs before shutting me down */
323 LASSERT(!sfw_data.fw_shuttingdown);
324 list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
326 spin_unlock(&sfw_data.fw_lock);
329 static struct sfw_batch *
330 sfw_find_batch(struct lst_bid bid)
332 struct sfw_session *sn = sfw_data.fw_session;
333 struct sfw_batch *bat;
337 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
338 if (bat->bat_id.bat_id == bid.bat_id)
345 static struct sfw_batch *
346 sfw_bid2batch(struct lst_bid bid)
348 struct sfw_session *sn = sfw_data.fw_session;
349 struct sfw_batch *bat;
351 LASSERT (sn != NULL);
353 bat = sfw_find_batch(bid);
357 LIBCFS_ALLOC(bat, sizeof(*bat));
362 bat->bat_session = sn;
364 atomic_set(&bat->bat_nactive, 0);
365 INIT_LIST_HEAD(&bat->bat_tests);
367 list_add_tail(&bat->bat_list, &sn->sn_batches);
372 sfw_get_stats(struct srpc_stat_reqst *request, struct srpc_stat_reply *reply)
374 struct sfw_session *sn = sfw_data.fw_session;
375 struct sfw_counters *cnt = &reply->str_fw;
376 struct sfw_batch *bat;
378 reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
380 if (request->str_sid.ses_nid == LNET_NID_ANY) {
381 reply->str_status = EINVAL;
385 if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
386 reply->str_status = ESRCH;
390 lnet_counters_get_common(&reply->str_lnet);
391 srpc_get_counters(&reply->str_rpc);
393 /* send over the msecs since the session was started
394 - with 32 bits to send, this is ~49 days */
395 cnt->running_ms = ktime_ms_delta(ktime_get(), sn->sn_started);
396 cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
397 cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
398 cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
400 cnt->active_batches = 0;
401 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
402 if (atomic_read(&bat->bat_nactive) > 0)
403 cnt->active_batches++;
406 reply->str_status = 0;
411 sfw_make_session(struct srpc_mksn_reqst *request, struct srpc_mksn_reply *reply)
413 struct sfw_session *sn = sfw_data.fw_session;
414 struct srpc_msg *msg = container_of(request, struct srpc_msg,
415 msg_body.mksn_reqst);
418 if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
419 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
420 reply->mksn_status = EINVAL;
425 reply->mksn_status = 0;
426 reply->mksn_sid = sn->sn_id;
427 reply->mksn_timeout = sn->sn_timeout;
429 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
430 atomic_inc(&sn->sn_refcount);
434 if (!request->mksn_force) {
435 reply->mksn_status = EBUSY;
436 cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
437 sizeof(reply->mksn_name));
438 if (cplen >= sizeof(reply->mksn_name))
444 /* reject the request if it requires unknown features
445 * NB: old version will always accept all features because it's not
446 * aware of struct srpc_msg::msg_ses_feats, it's a defect but it's also
447 * harmless because it will return zero feature to console, and it's
448 * console's responsibility to make sure all nodes in a session have
449 * same feature mask. */
450 if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
451 reply->mksn_status = EPROTO;
455 /* brand new or create by force */
456 LIBCFS_ALLOC(sn, sizeof(*sn));
458 CERROR("dropping RPC mksn under memory pressure\n");
462 sfw_init_session(sn, request->mksn_sid,
463 msg->msg_ses_feats, &request->mksn_name[0]);
465 spin_lock(&sfw_data.fw_lock);
467 sfw_deactivate_session();
468 LASSERT(sfw_data.fw_session == NULL);
469 sfw_data.fw_session = sn;
471 spin_unlock(&sfw_data.fw_lock);
473 reply->mksn_status = 0;
474 reply->mksn_sid = sn->sn_id;
475 reply->mksn_timeout = sn->sn_timeout;
480 sfw_remove_session(struct srpc_rmsn_reqst *request,
481 struct srpc_rmsn_reply *reply)
483 struct sfw_session *sn = sfw_data.fw_session;
485 reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
487 if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
488 reply->rmsn_status = EINVAL;
492 if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
493 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
497 if (!atomic_dec_and_test(&sn->sn_refcount)) {
498 reply->rmsn_status = 0;
502 spin_lock(&sfw_data.fw_lock);
503 sfw_deactivate_session();
504 spin_unlock(&sfw_data.fw_lock);
506 reply->rmsn_status = 0;
507 reply->rmsn_sid = LST_INVALID_SID;
508 LASSERT(sfw_data.fw_session == NULL);
513 sfw_debug_session(struct srpc_debug_reqst *request,
514 struct srpc_debug_reply *reply)
516 struct sfw_session *sn = sfw_data.fw_session;
519 reply->dbg_status = ESRCH;
520 reply->dbg_sid = LST_INVALID_SID;
524 reply->dbg_status = 0;
525 reply->dbg_sid = sn->sn_id;
526 reply->dbg_timeout = sn->sn_timeout;
527 if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
528 >= sizeof(reply->dbg_name))
535 sfw_test_rpc_fini(struct srpc_client_rpc *rpc)
537 struct sfw_test_unit *tsu = rpc->crpc_priv;
538 struct sfw_test_instance *tsi = tsu->tsu_instance;
540 /* Called with hold of tsi->tsi_lock */
541 LASSERT(list_empty(&rpc->crpc_list));
542 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
546 sfw_test_buffers(struct sfw_test_instance *tsi)
548 struct sfw_test_case *tsc;
549 struct srpc_service *svc;
552 LASSERT(tsi != NULL);
553 tsc = sfw_find_test_case(tsi->tsi_service);
554 LASSERT(tsc != NULL);
555 svc = tsc->tsc_srv_service;
556 LASSERT(svc != NULL);
558 nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
559 return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
563 sfw_load_test(struct sfw_test_instance *tsi)
565 struct sfw_test_case *tsc;
566 struct srpc_service *svc;
570 LASSERT(tsi != NULL);
571 tsc = sfw_find_test_case(tsi->tsi_service);
572 nbuf = sfw_test_buffers(tsi);
573 LASSERT(tsc != NULL);
574 svc = tsc->tsc_srv_service;
576 if (tsi->tsi_is_client) {
577 tsi->tsi_ops = tsc->tsc_cli_ops;
581 rc = srpc_service_add_buffers(svc, nbuf);
583 CWARN("Failed to reserve enough buffers: "
584 "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
585 /* NB: this error handler is not strictly correct, because
586 * it may release more buffers than already allocated,
587 * but it doesn't matter because request portal should
588 * be lazy portal and will grow buffers if necessary. */
589 srpc_service_remove_buffers(svc, nbuf);
593 CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
594 nbuf * (srpc_serv_is_framework(svc) ?
595 1 : cfs_cpt_number(cfs_cpt_tab)), svc->sv_name);
600 sfw_unload_test(struct sfw_test_instance *tsi)
602 struct sfw_test_case *tsc;
604 LASSERT(tsi != NULL);
605 tsc = sfw_find_test_case(tsi->tsi_service);
606 LASSERT(tsc != NULL);
608 if (tsi->tsi_is_client)
611 /* shrink buffers, because request portal is lazy portal
612 * which can grow buffers at runtime so we may leave
613 * some buffers behind, but never mind... */
614 srpc_service_remove_buffers(tsc->tsc_srv_service,
615 sfw_test_buffers(tsi));
619 sfw_destroy_test_instance(struct sfw_test_instance *tsi)
621 struct srpc_client_rpc *rpc;
622 struct sfw_test_unit *tsu;
624 if (!tsi->tsi_is_client) goto clean;
626 tsi->tsi_ops->tso_fini(tsi);
628 LASSERT(!tsi->tsi_stopping);
629 LASSERT(list_empty(&tsi->tsi_active_rpcs));
630 LASSERT(!sfw_test_active(tsi));
632 while (!list_empty(&tsi->tsi_units)) {
633 tsu = list_entry(tsi->tsi_units.next,
634 struct sfw_test_unit, tsu_list);
635 list_del(&tsu->tsu_list);
636 LIBCFS_FREE(tsu, sizeof(*tsu));
639 while (!list_empty(&tsi->tsi_free_rpcs)) {
640 rpc = list_entry(tsi->tsi_free_rpcs.next,
641 struct srpc_client_rpc, crpc_list);
642 list_del(&rpc->crpc_list);
643 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
647 sfw_unload_test(tsi);
648 LIBCFS_FREE(tsi, sizeof(*tsi));
652 sfw_destroy_batch(struct sfw_batch *tsb)
654 struct sfw_test_instance *tsi;
656 LASSERT(!sfw_batch_active(tsb));
657 LASSERT(list_empty(&tsb->bat_list));
659 while (!list_empty(&tsb->bat_tests)) {
660 tsi = list_entry(tsb->bat_tests.next,
661 struct sfw_test_instance, tsi_list);
662 list_del_init(&tsi->tsi_list);
663 sfw_destroy_test_instance(tsi);
666 LIBCFS_FREE(tsb, sizeof(*tsb));
670 sfw_destroy_session(struct sfw_session *sn)
672 struct sfw_batch *batch;
674 LASSERT(list_empty(&sn->sn_list));
675 LASSERT(sn != sfw_data.fw_session);
677 while (!list_empty(&sn->sn_batches)) {
678 batch = list_entry(sn->sn_batches.next,
679 struct sfw_batch, bat_list);
680 list_del_init(&batch->bat_list);
681 sfw_destroy_batch(batch);
684 LIBCFS_FREE(sn, sizeof(*sn));
685 atomic_dec(&sfw_data.fw_nzombies);
689 sfw_unpack_addtest_req(struct srpc_msg *msg)
691 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
693 LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
694 LASSERT (req->tsr_is_client);
696 if (msg->msg_magic == SRPC_MSG_MAGIC)
697 return; /* no flipping needed */
699 LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
701 if (req->tsr_service == SRPC_SERVICE_BRW) {
702 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
703 struct test_bulk_req *bulk = &req->tsr_u.bulk_v0;
705 __swab32s(&bulk->blk_opc);
706 __swab32s(&bulk->blk_npg);
707 __swab32s(&bulk->blk_flags);
710 struct test_bulk_req_v1 *bulk = &req->tsr_u.bulk_v1;
712 __swab16s(&bulk->blk_opc);
713 __swab16s(&bulk->blk_flags);
714 __swab32s(&bulk->blk_offset);
715 __swab32s(&bulk->blk_len);
721 if (req->tsr_service == SRPC_SERVICE_PING) {
722 struct test_ping_req *ping = &req->tsr_u.ping;
724 __swab32s(&ping->png_size);
725 __swab32s(&ping->png_flags);
733 sfw_add_test_instance(struct sfw_batch *tsb, struct srpc_server_rpc *rpc)
735 struct srpc_msg *msg = &rpc->srpc_reqstbuf->buf_msg;
736 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
737 struct srpc_bulk *bk = rpc->srpc_bulk;
738 int ndest = req->tsr_ndest;
739 struct sfw_test_unit *tsu;
740 struct sfw_test_instance *tsi;
744 LIBCFS_ALLOC(tsi, sizeof(*tsi));
746 CERROR ("Can't allocate test instance for batch: %llu\n",
751 spin_lock_init(&tsi->tsi_lock);
752 atomic_set(&tsi->tsi_nactive, 0);
753 INIT_LIST_HEAD(&tsi->tsi_units);
754 INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
755 INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
757 tsi->tsi_stopping = 0;
758 tsi->tsi_batch = tsb;
759 tsi->tsi_loop = req->tsr_loop;
760 tsi->tsi_concur = req->tsr_concur;
761 tsi->tsi_service = req->tsr_service;
762 tsi->tsi_is_client = !!(req->tsr_is_client);
763 tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
765 rc = sfw_load_test(tsi);
767 LIBCFS_FREE(tsi, sizeof(*tsi));
771 LASSERT (!sfw_batch_active(tsb));
773 if (!tsi->tsi_is_client) {
774 /* it's test server, just add it to tsb */
775 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
779 LASSERT (bk != NULL);
780 LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
781 LASSERT((unsigned int)bk->bk_len >=
782 sizeof(struct lnet_process_id_packed) * ndest);
784 sfw_unpack_addtest_req(msg);
785 memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
787 for (i = 0; i < ndest; i++) {
788 struct lnet_process_id_packed *dests;
789 struct lnet_process_id_packed id;
792 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].bv_page);
793 LASSERT (dests != NULL); /* my pages are within KVM always */
794 id = dests[i % SFW_ID_PER_PAGE];
795 if (msg->msg_magic != SRPC_MSG_MAGIC)
798 for (j = 0; j < tsi->tsi_concur; j++) {
799 LIBCFS_ALLOC(tsu, sizeof(*tsu));
802 CERROR ("Can't allocate tsu for %d\n",
807 tsu->tsu_dest.nid = id.nid;
808 tsu->tsu_dest.pid = id.pid;
809 tsu->tsu_instance = tsi;
810 tsu->tsu_private = NULL;
811 list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
815 rc = tsi->tsi_ops->tso_init(tsi);
817 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
823 sfw_destroy_test_instance(tsi);
828 sfw_test_unit_done(struct sfw_test_unit *tsu)
830 struct sfw_test_instance *tsi = tsu->tsu_instance;
831 struct sfw_batch *tsb = tsi->tsi_batch;
832 struct sfw_session *sn = tsb->bat_session;
834 LASSERT (sfw_test_active(tsi));
836 if (!atomic_dec_and_test(&tsi->tsi_nactive))
839 /* the test instance is done */
840 spin_lock(&tsi->tsi_lock);
842 tsi->tsi_stopping = 0;
844 spin_unlock(&tsi->tsi_lock);
846 spin_lock(&sfw_data.fw_lock);
848 if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
849 sn == sfw_data.fw_session) { /* sn also active */
850 spin_unlock(&sfw_data.fw_lock);
854 LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
856 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
857 if (sfw_batch_active(tsb)) {
858 spin_unlock(&sfw_data.fw_lock);
863 list_del_init(&sn->sn_list);
864 spin_unlock(&sfw_data.fw_lock);
866 sfw_destroy_session(sn);
870 sfw_test_rpc_done(struct srpc_client_rpc *rpc)
872 struct sfw_test_unit *tsu = rpc->crpc_priv;
873 struct sfw_test_instance *tsi = tsu->tsu_instance;
876 tsi->tsi_ops->tso_done_rpc(tsu, rpc);
878 spin_lock(&tsi->tsi_lock);
880 LASSERT(sfw_test_active(tsi));
881 LASSERT(!list_empty(&rpc->crpc_list));
883 list_del_init(&rpc->crpc_list);
885 /* batch is stopping or loop is done or get error */
886 if (tsi->tsi_stopping ||
887 tsu->tsu_loop == 0 ||
888 (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
891 /* dec ref for poster */
892 srpc_client_rpc_decref(rpc);
894 spin_unlock(&tsi->tsi_lock);
897 swi_schedule_workitem(&tsu->tsu_worker);
901 sfw_test_unit_done(tsu);
905 sfw_create_test_rpc(struct sfw_test_unit *tsu, struct lnet_process_id peer,
906 unsigned features, int nblk, int blklen,
907 struct srpc_client_rpc **rpcpp)
909 struct srpc_client_rpc *rpc = NULL;
910 struct sfw_test_instance *tsi = tsu->tsu_instance;
912 spin_lock(&tsi->tsi_lock);
914 LASSERT (sfw_test_active(tsi));
916 if (!list_empty(&tsi->tsi_free_rpcs)) {
917 /* pick request from buffer */
918 rpc = list_entry(tsi->tsi_free_rpcs.next,
919 struct srpc_client_rpc, crpc_list);
920 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
921 list_del_init(&rpc->crpc_list);
924 spin_unlock(&tsi->tsi_lock);
927 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
928 blklen, sfw_test_rpc_done,
929 sfw_test_rpc_fini, tsu);
931 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
932 blklen, sfw_test_rpc_done,
933 sfw_test_rpc_fini, tsu);
937 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
941 rpc->crpc_reqstmsg.msg_ses_feats = features;
948 sfw_run_test(struct swi_workitem *wi)
950 struct sfw_test_unit *tsu = container_of(wi, struct sfw_test_unit, tsu_worker);
951 struct sfw_test_instance *tsi = tsu->tsu_instance;
952 struct srpc_client_rpc *rpc = NULL;
954 LASSERT (wi == &tsu->tsu_worker);
956 if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
957 LASSERT (rpc == NULL);
961 LASSERT (rpc != NULL);
963 spin_lock(&tsi->tsi_lock);
965 if (tsi->tsi_stopping) {
966 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
967 spin_unlock(&tsi->tsi_lock);
971 if (tsu->tsu_loop > 0)
974 list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
975 spin_unlock(&tsi->tsi_lock);
977 spin_lock(&rpc->crpc_lock);
978 rpc->crpc_timeout = rpc_timeout;
980 spin_unlock(&rpc->crpc_lock);
985 * No one can schedule me now since:
986 * - previous RPC, if any, has done and
987 * - no new RPC is initiated.
988 * - my batch is still active; no one can run it again now.
989 * Cancel pending schedules and prevent future schedule attempts:
991 swi_exit_workitem(wi);
992 sfw_test_unit_done(tsu);
997 sfw_run_batch(struct sfw_batch *tsb)
999 struct swi_workitem *wi;
1000 struct sfw_test_unit *tsu;
1001 struct sfw_test_instance *tsi;
1003 if (sfw_batch_active(tsb)) {
1004 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1005 tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1009 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1010 if (!tsi->tsi_is_client) /* skip server instances */
1013 LASSERT(!tsi->tsi_stopping);
1014 LASSERT(!sfw_test_active(tsi));
1016 atomic_inc(&tsb->bat_nactive);
1018 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1019 atomic_inc(&tsi->tsi_nactive);
1020 tsu->tsu_loop = tsi->tsi_loop;
1021 wi = &tsu->tsu_worker;
1022 swi_init_workitem(wi, sfw_run_test,
1023 lst_sched_test[lnet_cpt_of_nid(tsu->tsu_dest.nid, NULL)]);
1024 swi_schedule_workitem(wi);
1032 sfw_stop_batch(struct sfw_batch *tsb, int force)
1034 struct sfw_test_instance *tsi;
1035 struct srpc_client_rpc *rpc;
1037 if (!sfw_batch_active(tsb)) {
1038 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1042 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1043 spin_lock(&tsi->tsi_lock);
1045 if (!tsi->tsi_is_client ||
1046 !sfw_test_active(tsi) || tsi->tsi_stopping) {
1047 spin_unlock(&tsi->tsi_lock);
1051 tsi->tsi_stopping = 1;
1054 spin_unlock(&tsi->tsi_lock);
1058 /* abort launched rpcs in the test */
1059 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1060 spin_lock(&rpc->crpc_lock);
1062 srpc_abort_rpc(rpc, -EINTR);
1064 spin_unlock(&rpc->crpc_lock);
1067 spin_unlock(&tsi->tsi_lock);
1074 sfw_query_batch(struct sfw_batch *tsb, int testidx,
1075 struct srpc_batch_reply *reply)
1077 struct sfw_test_instance *tsi;
1083 reply->bar_active = atomic_read(&tsb->bat_nactive);
1087 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1091 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1099 sfw_free_pages(struct srpc_server_rpc *rpc)
1101 srpc_free_bulk(rpc->srpc_bulk);
1102 rpc->srpc_bulk = NULL;
1106 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1109 LASSERT(rpc->srpc_bulk == NULL);
1110 LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1112 rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1113 if (rpc->srpc_bulk == NULL)
1120 sfw_add_test(struct srpc_server_rpc *rpc)
1122 struct sfw_session *sn = sfw_data.fw_session;
1123 struct srpc_test_reply *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1124 struct srpc_test_reqst *request;
1126 struct sfw_batch *bat;
1128 request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1129 reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1131 if (request->tsr_loop == 0 ||
1132 request->tsr_concur == 0 ||
1133 request->tsr_sid.ses_nid == LNET_NID_ANY ||
1134 request->tsr_ndest > SFW_MAX_NDESTS ||
1135 (request->tsr_is_client && request->tsr_ndest == 0) ||
1136 request->tsr_concur > SFW_MAX_CONCUR ||
1137 request->tsr_service > SRPC_SERVICE_MAX_ID ||
1138 request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1139 reply->tsr_status = EINVAL;
1143 if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1144 sfw_find_test_case(request->tsr_service) == NULL) {
1145 reply->tsr_status = ENOENT;
1149 bat = sfw_bid2batch(request->tsr_bid);
1151 CERROR("dropping RPC %s from %s under memory pressure\n",
1152 rpc->srpc_scd->scd_svc->sv_name,
1153 libcfs_id2str(rpc->srpc_peer));
1157 if (sfw_batch_active(bat)) {
1158 reply->tsr_status = EBUSY;
1162 if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1163 /* rpc will be resumed later in sfw_bulk_ready */
1164 int npg = sfw_id_pages(request->tsr_ndest);
1167 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1168 len = npg * PAGE_SIZE;
1171 len = sizeof(struct lnet_process_id_packed) *
1175 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1178 rc = sfw_add_test_instance(bat, rpc);
1179 CDEBUG (rc == 0 ? D_NET : D_WARNING,
1180 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1181 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1182 request->tsr_is_client ? "client" : "server",
1183 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1185 reply->tsr_status = (rc < 0) ? -rc : rc;
1190 sfw_control_batch(struct srpc_batch_reqst *request,
1191 struct srpc_batch_reply *reply)
1193 struct sfw_session *sn = sfw_data.fw_session;
1195 struct sfw_batch *bat;
1197 reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1199 if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1200 reply->bar_status = ESRCH;
1204 bat = sfw_find_batch(request->bar_bid);
1206 reply->bar_status = ENOENT;
1210 switch (request->bar_opc) {
1211 case SRPC_BATCH_OPC_RUN:
1212 rc = sfw_run_batch(bat);
1215 case SRPC_BATCH_OPC_STOP:
1216 rc = sfw_stop_batch(bat, request->bar_arg);
1219 case SRPC_BATCH_OPC_QUERY:
1220 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1224 return -EINVAL; /* drop it */
1227 reply->bar_status = (rc < 0) ? -rc : rc;
1232 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1234 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1235 struct srpc_msg *reply = &rpc->srpc_replymsg;
1236 struct srpc_msg *request = &rpc->srpc_reqstbuf->buf_msg;
1237 unsigned features = LST_FEATS_MASK;
1240 LASSERT(sfw_data.fw_active_srpc == NULL);
1241 LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1243 spin_lock(&sfw_data.fw_lock);
1245 if (sfw_data.fw_shuttingdown) {
1246 spin_unlock(&sfw_data.fw_lock);
1250 /* Remove timer to avoid racing with it or expiring active session */
1251 if (sfw_del_session_timer() != 0) {
1252 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1253 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1254 spin_unlock(&sfw_data.fw_lock);
1258 sfw_data.fw_active_srpc = rpc;
1259 spin_unlock(&sfw_data.fw_lock);
1261 sfw_unpack_message(request);
1262 LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1264 /* rpc module should have checked this */
1265 LASSERT(request->msg_version == SRPC_MSG_VERSION);
1267 if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1268 sv->sv_id != SRPC_SERVICE_DEBUG) {
1269 struct sfw_session *sn = sfw_data.fw_session;
1272 sn->sn_features != request->msg_ses_feats) {
1273 CNETERR("Features of framework RPC don't match "
1274 "features of current session: %x/%x\n",
1275 request->msg_ses_feats, sn->sn_features);
1276 reply->msg_body.reply.status = EPROTO;
1277 reply->msg_body.reply.sid = sn->sn_id;
1281 } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1282 /* NB: at this point, old version will ignore features and
1283 * create new session anyway, so console should be able
1285 reply->msg_body.reply.status = EPROTO;
1292 case SRPC_SERVICE_TEST:
1293 rc = sfw_add_test(rpc);
1296 case SRPC_SERVICE_BATCH:
1297 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1298 &reply->msg_body.bat_reply);
1301 case SRPC_SERVICE_QUERY_STAT:
1302 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1303 &reply->msg_body.stat_reply);
1306 case SRPC_SERVICE_DEBUG:
1307 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1308 &reply->msg_body.dbg_reply);
1311 case SRPC_SERVICE_MAKE_SESSION:
1312 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1313 &reply->msg_body.mksn_reply);
1316 case SRPC_SERVICE_REMOVE_SESSION:
1317 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1318 &reply->msg_body.rmsn_reply);
1322 if (sfw_data.fw_session != NULL)
1323 features = sfw_data.fw_session->sn_features;
1325 reply->msg_ses_feats = features;
1326 rpc->srpc_done = sfw_server_rpc_done;
1327 spin_lock(&sfw_data.fw_lock);
1329 if (!sfw_data.fw_shuttingdown)
1330 sfw_add_session_timer();
1332 sfw_data.fw_active_srpc = NULL;
1333 spin_unlock(&sfw_data.fw_lock);
1338 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1340 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1343 LASSERT(rpc->srpc_bulk != NULL);
1344 LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1345 LASSERT(sfw_data.fw_active_srpc == NULL);
1346 LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1348 spin_lock(&sfw_data.fw_lock);
1351 CERROR("Bulk transfer failed for RPC: "
1352 "service %s, peer %s, status %d\n",
1353 sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1354 spin_unlock(&sfw_data.fw_lock);
1358 if (sfw_data.fw_shuttingdown) {
1359 spin_unlock(&sfw_data.fw_lock);
1363 if (sfw_del_session_timer() != 0) {
1364 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1365 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1366 spin_unlock(&sfw_data.fw_lock);
1370 sfw_data.fw_active_srpc = rpc;
1371 spin_unlock(&sfw_data.fw_lock);
1373 rc = sfw_add_test(rpc);
1375 spin_lock(&sfw_data.fw_lock);
1377 if (!sfw_data.fw_shuttingdown)
1378 sfw_add_session_timer();
1380 sfw_data.fw_active_srpc = NULL;
1381 spin_unlock(&sfw_data.fw_lock);
1385 struct srpc_client_rpc *
1386 sfw_create_rpc(struct lnet_process_id peer, int service,
1387 unsigned features, int nbulkiov, int bulklen,
1388 void (*done)(struct srpc_client_rpc *), void *priv)
1390 struct srpc_client_rpc *rpc = NULL;
1392 spin_lock(&sfw_data.fw_lock);
1394 LASSERT (!sfw_data.fw_shuttingdown);
1395 LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1397 if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1398 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1399 struct srpc_client_rpc, crpc_list);
1400 list_del(&rpc->crpc_list);
1402 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1403 done, sfw_client_rpc_fini, priv);
1406 spin_unlock(&sfw_data.fw_lock);
1409 rpc = srpc_create_client_rpc(peer, service,
1410 nbulkiov, bulklen, done,
1411 nbulkiov != 0 ? NULL :
1412 sfw_client_rpc_fini,
1416 if (rpc != NULL) /* "session" is concept in framework */
1417 rpc->crpc_reqstmsg.msg_ses_feats = features;
1423 sfw_unpack_message(struct srpc_msg *msg)
1425 if (msg->msg_magic == SRPC_MSG_MAGIC)
1426 return; /* no flipping needed */
1428 /* srpc module should guarantee I wouldn't get crap */
1429 LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1431 if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1432 struct srpc_stat_reqst *req = &msg->msg_body.stat_reqst;
1434 __swab32s(&req->str_type);
1435 __swab64s(&req->str_rpyid);
1436 sfw_unpack_sid(req->str_sid);
1440 if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1441 struct srpc_stat_reply *rep = &msg->msg_body.stat_reply;
1443 __swab32s(&rep->str_status);
1444 sfw_unpack_sid(rep->str_sid);
1445 sfw_unpack_fw_counters(rep->str_fw);
1446 sfw_unpack_rpc_counters(rep->str_rpc);
1447 sfw_unpack_lnet_counters(rep->str_lnet);
1451 if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1452 struct srpc_mksn_reqst *req = &msg->msg_body.mksn_reqst;
1454 __swab64s(&req->mksn_rpyid);
1455 __swab32s(&req->mksn_force);
1456 sfw_unpack_sid(req->mksn_sid);
1460 if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1461 struct srpc_mksn_reply *rep = &msg->msg_body.mksn_reply;
1463 __swab32s(&rep->mksn_status);
1464 __swab32s(&rep->mksn_timeout);
1465 sfw_unpack_sid(rep->mksn_sid);
1469 if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1470 struct srpc_rmsn_reqst *req = &msg->msg_body.rmsn_reqst;
1472 __swab64s(&req->rmsn_rpyid);
1473 sfw_unpack_sid(req->rmsn_sid);
1477 if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1478 struct srpc_rmsn_reply *rep = &msg->msg_body.rmsn_reply;
1480 __swab32s(&rep->rmsn_status);
1481 sfw_unpack_sid(rep->rmsn_sid);
1485 if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1486 struct srpc_debug_reqst *req = &msg->msg_body.dbg_reqst;
1488 __swab64s(&req->dbg_rpyid);
1489 __swab32s(&req->dbg_flags);
1490 sfw_unpack_sid(req->dbg_sid);
1494 if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1495 struct srpc_debug_reply *rep = &msg->msg_body.dbg_reply;
1497 __swab32s(&rep->dbg_nbatch);
1498 __swab32s(&rep->dbg_timeout);
1499 sfw_unpack_sid(rep->dbg_sid);
1503 if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1504 struct srpc_batch_reqst *req = &msg->msg_body.bat_reqst;
1506 __swab32s(&req->bar_opc);
1507 __swab64s(&req->bar_rpyid);
1508 __swab32s(&req->bar_testidx);
1509 __swab32s(&req->bar_arg);
1510 sfw_unpack_sid(req->bar_sid);
1511 __swab64s(&req->bar_bid.bat_id);
1515 if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1516 struct srpc_batch_reply *rep = &msg->msg_body.bat_reply;
1518 __swab32s(&rep->bar_status);
1519 sfw_unpack_sid(rep->bar_sid);
1523 if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1524 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
1526 __swab64s(&req->tsr_rpyid);
1527 __swab64s(&req->tsr_bulkid);
1528 __swab32s(&req->tsr_loop);
1529 __swab32s(&req->tsr_ndest);
1530 __swab32s(&req->tsr_concur);
1531 __swab32s(&req->tsr_service);
1532 sfw_unpack_sid(req->tsr_sid);
1533 __swab64s(&req->tsr_bid.bat_id);
1537 if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1538 struct srpc_test_reply *rep = &msg->msg_body.tes_reply;
1540 __swab32s(&rep->tsr_status);
1541 sfw_unpack_sid(rep->tsr_sid);
1545 if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1546 struct srpc_join_reqst *req = &msg->msg_body.join_reqst;
1548 __swab64s(&req->join_rpyid);
1549 sfw_unpack_sid(req->join_sid);
1553 if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1554 struct srpc_join_reply *rep = &msg->msg_body.join_reply;
1556 __swab32s(&rep->join_status);
1557 __swab32s(&rep->join_timeout);
1558 sfw_unpack_sid(rep->join_sid);
1566 sfw_abort_rpc(struct srpc_client_rpc *rpc)
1568 LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1569 LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1571 spin_lock(&rpc->crpc_lock);
1572 srpc_abort_rpc(rpc, -EINTR);
1573 spin_unlock(&rpc->crpc_lock);
1577 sfw_post_rpc(struct srpc_client_rpc *rpc)
1579 spin_lock(&rpc->crpc_lock);
1581 LASSERT(!rpc->crpc_closed);
1582 LASSERT(!rpc->crpc_aborted);
1583 LASSERT(list_empty(&rpc->crpc_list));
1584 LASSERT(!sfw_data.fw_shuttingdown);
1586 rpc->crpc_timeout = rpc_timeout;
1589 spin_unlock(&rpc->crpc_lock);
1592 static struct srpc_service sfw_services[] = {
1593 { .sv_id = SRPC_SERVICE_DEBUG, .sv_name = "debug", },
1594 { .sv_id = SRPC_SERVICE_QUERY_STAT, .sv_name = "query stats", },
1595 { .sv_id = SRPC_SERVICE_MAKE_SESSION, .sv_name = "make session", },
1596 { .sv_id = SRPC_SERVICE_REMOVE_SESSION, .sv_name = "remove session", },
1597 { .sv_id = SRPC_SERVICE_BATCH, .sv_name = "batch service", },
1598 { .sv_id = SRPC_SERVICE_TEST, .sv_name = "test service", },
1607 struct srpc_service *sv;
1608 struct sfw_test_case *tsc;
1611 if (session_timeout < 0) {
1612 CERROR ("Session timeout must be non-negative: %d\n",
1617 if (rpc_timeout < 0) {
1618 CERROR ("RPC timeout must be non-negative: %d\n",
1623 if (session_timeout == 0)
1624 CWARN ("Zero session_timeout specified "
1625 "- test sessions never expire.\n");
1627 if (rpc_timeout == 0)
1628 CWARN ("Zero rpc_timeout specified "
1629 "- test RPC never expire.\n");
1631 memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1633 sfw_data.fw_session = NULL;
1634 sfw_data.fw_active_srpc = NULL;
1635 spin_lock_init(&sfw_data.fw_lock);
1636 atomic_set(&sfw_data.fw_nzombies, 0);
1637 INIT_LIST_HEAD(&sfw_data.fw_tests);
1638 INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1639 INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1641 brw_init_test_client();
1642 brw_init_test_service();
1643 rc = sfw_register_test(&brw_test_service, &brw_test_client);
1646 ping_init_test_client();
1647 ping_init_test_service();
1648 rc = sfw_register_test(&ping_test_service, &ping_test_client);
1652 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1653 sv = tsc->tsc_srv_service;
1655 rc = srpc_add_service(sv);
1656 LASSERT(rc != -EBUSY);
1658 CWARN("Failed to add %s service: %d\n",
1664 for (i = 0; ; i++) {
1665 sv = &sfw_services[i];
1666 if (sv->sv_name == NULL) break;
1668 sv->sv_bulk_ready = NULL;
1669 sv->sv_handler = sfw_handle_server_rpc;
1670 sv->sv_wi_total = SFW_FRWK_WI_MAX;
1671 if (sv->sv_id == SRPC_SERVICE_TEST)
1672 sv->sv_bulk_ready = sfw_bulk_ready;
1674 rc = srpc_add_service(sv);
1675 LASSERT (rc != -EBUSY);
1677 CWARN ("Failed to add %s service: %d\n",
1682 /* about to sfw_shutdown, no need to add buffer */
1683 if (error) continue;
1685 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1687 CWARN("Failed to reserve enough buffers: "
1688 "service %s, %d needed: %d\n",
1689 sv->sv_name, sv->sv_wi_total, rc);
1702 struct srpc_service *sv;
1703 struct sfw_test_case *tsc;
1706 spin_lock(&sfw_data.fw_lock);
1708 sfw_data.fw_shuttingdown = 1;
1709 lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1710 "waiting for active RPC to finish.\n");
1712 if (sfw_del_session_timer() != 0)
1713 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1714 "waiting for session timer to explode.\n");
1716 sfw_deactivate_session();
1717 lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1719 "waiting for %d zombie sessions to die.\n",
1720 atomic_read(&sfw_data.fw_nzombies));
1722 spin_unlock(&sfw_data.fw_lock);
1724 for (i = 0; ; i++) {
1725 sv = &sfw_services[i];
1726 if (sv->sv_name == NULL)
1729 srpc_shutdown_service(sv);
1730 srpc_remove_service(sv);
1733 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1734 sv = tsc->tsc_srv_service;
1735 srpc_shutdown_service(sv);
1736 srpc_remove_service(sv);
1739 while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1740 struct srpc_client_rpc *rpc;
1742 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1743 struct srpc_client_rpc, crpc_list);
1744 list_del(&rpc->crpc_list);
1746 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1749 for (i = 0; ; i++) {
1750 sv = &sfw_services[i];
1751 if (sv->sv_name == NULL)
1754 srpc_wait_service_shutdown(sv);
1757 while (!list_empty(&sfw_data.fw_tests)) {
1758 tsc = list_entry(sfw_data.fw_tests.next,
1759 struct sfw_test_case, tsc_list);
1761 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1763 list_del(&tsc->tsc_list);
1764 LIBCFS_FREE(tsc, sizeof(*tsc));