projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging/lustre: use 64-bit time for selftest
[deliverable/linux.git] / drivers / staging / lustre / lnet / selftest / framework.c
CommitLineData
d7e09d03
PT		 1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * lnet/selftest/framework.c
37 *
38 * Author: Isaac Huang <isaac@clusterfs.com>
39 * Author: Liang Zhen <liangzhen@clusterfs.com>
40 */
41
42#define DEBUG_SUBSYSTEM S_LNET
43
44#include "selftest.h"
45
46lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
47
48static int session_timeout = 100;
8cc7b4b9
PT		 49module_param(session_timeout, int, 0444);
50MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
d7e09d03
PT		 51
52static int rpc_timeout = 64;
8cc7b4b9
PT		 53module_param(rpc_timeout, int, 0644);
54MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
d7e09d03 55
74d68011
MS		 56#define sfw_unpack_id(id) \
57do { \
d7e09d03
PT		 58 __swab64s(&(id).nid); \
59 __swab32s(&(id).pid); \
60} while (0)
61
74d68011
MS		 62#define sfw_unpack_sid(sid) \
63do { \
d7e09d03
PT		 64 __swab64s(&(sid).ses_nid); \
65 __swab64s(&(sid).ses_stamp); \
66} while (0)
67
74d68011
MS		 68#define sfw_unpack_fw_counters(fc) \
69do { \
d7e09d03
PT		 70 __swab32s(&(fc).running_ms); \
71 __swab32s(&(fc).active_batches); \
72 __swab32s(&(fc).zombie_sessions); \
73 __swab32s(&(fc).brw_errors); \
74 __swab32s(&(fc).ping_errors); \
75} while (0)
76
77#define sfw_unpack_rpc_counters(rc) \
74d68011 78do { \
d7e09d03
PT		 79 __swab32s(&(rc).errors); \
80 __swab32s(&(rc).rpcs_sent); \
81 __swab32s(&(rc).rpcs_rcvd); \
82 __swab32s(&(rc).rpcs_dropped); \
83 __swab32s(&(rc).rpcs_expired); \
84 __swab64s(&(rc).bulk_get); \
85 __swab64s(&(rc).bulk_put); \
86} while (0)
87
88#define sfw_unpack_lnet_counters(lc) \
74d68011 89do { \
d7e09d03
PT		 90 __swab32s(&(lc).errors); \
91 __swab32s(&(lc).msgs_max); \
92 __swab32s(&(lc).msgs_alloc); \
93 __swab32s(&(lc).send_count); \
94 __swab32s(&(lc).recv_count); \
95 __swab32s(&(lc).drop_count); \
96 __swab32s(&(lc).route_count); \
97 __swab64s(&(lc).send_length); \
98 __swab64s(&(lc).recv_length); \
99 __swab64s(&(lc).drop_length); \
100 __swab64s(&(lc).route_length); \
101} while (0)
102
103#define sfw_test_active(t) (atomic_read(&(t)->tsi_nactive) != 0)
104#define sfw_batch_active(b) (atomic_read(&(b)->bat_nactive) != 0)
105
62366bf1 106static struct smoketest_framework {
74d68011
MS		 107 struct list_head fw_zombie_rpcs; /* RPCs to be recycled */
108 struct list_head fw_zombie_sessions; /* stopping sessions */
109 struct list_head fw_tests; /* registered test cases */
110 atomic_t fw_nzombies; /* # zombie sessions */
111 spinlock_t fw_lock; /* serialise */
112 sfw_session_t *fw_session; /* _the_ session */
113 int fw_shuttingdown; /* shutdown in progress */
114 srpc_server_rpc_t *fw_active_srpc; /* running RPC */
d7e09d03
PT		 115} sfw_data;
116
117/* forward ref's */
63418983
MY		 118int sfw_stop_batch(sfw_batch_t *tsb, int force);
119void sfw_destroy_session(sfw_session_t *sn);
d7e09d03
PT		 120
121static inline sfw_test_case_t *
122sfw_find_test_case(int id)
123{
124 sfw_test_case_t *tsc;
125
63418983
MY		 126 LASSERT(id <= SRPC_SERVICE_MAX_ID);
127 LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
d7e09d03 128
63418983 129 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
d7e09d03
PT		 130 if (tsc->tsc_srv_service->sv_id == id)
131 return tsc;
132 }
133
134 return NULL;
135}
136
137static int
63418983 138sfw_register_test(srpc_service_t *service, sfw_test_client_ops_t *cliops)
d7e09d03
PT		 139{
140 sfw_test_case_t *tsc;
141
142 if (sfw_find_test_case(service->sv_id) != NULL) {
63418983 143 CERROR("Failed to register test %s (%d)\n",
d7e09d03
PT		 144 service->sv_name, service->sv_id);
145 return -EEXIST;
146 }
147
148 LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
149 if (tsc == NULL)
150 return -ENOMEM;
151
d7e09d03
PT		 152 tsc->tsc_cli_ops = cliops;
153 tsc->tsc_srv_service = service;
154
155 list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
156 return 0;
157}
158
8d94b6d2 159static void
63418983 160sfw_add_session_timer(void)
d7e09d03
PT		 161{
162 sfw_session_t *sn = sfw_data.fw_session;
74d68011 163 stt_timer_t *timer = &sn->sn_timer;
d7e09d03 164
63418983 165 LASSERT(!sfw_data.fw_shuttingdown);
d7e09d03
PT		 166
167 if (sn == NULL || sn->sn_timeout == 0)
168 return;
169
63418983 170 LASSERT(!sn->sn_timer_active);
d7e09d03
PT		 171
172 sn->sn_timer_active = 1;
d9f79e6b 173 timer->stt_expires = ktime_get_real_seconds() + sn->sn_timeout;
d7e09d03
PT		 174 stt_add_timer(timer);
175 return;
176}
177
8d94b6d2 178static int
63418983 179sfw_del_session_timer(void)
d7e09d03
PT		 180{
181 sfw_session_t *sn = sfw_data.fw_session;
182
183 if (sn == NULL || !sn->sn_timer_active)
184 return 0;
185
63418983 186 LASSERT(sn->sn_timeout != 0);
d7e09d03
PT		 187
188 if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
189 sn->sn_timer_active = 0;
190 return 0;
191 }
192
193 return EBUSY; /* racing with sfw_session_expired() */
194}
195
d7e09d03 196static void
63418983 197sfw_deactivate_session(void)
8a136d11 198 __must_hold(&sfw_data.fw_lock)
d7e09d03
PT		 199{
200 sfw_session_t *sn = sfw_data.fw_session;
74d68011
MS		 201 int nactive = 0;
202 sfw_batch_t *tsb;
d7e09d03
PT		 203 sfw_test_case_t *tsc;
204
550d35d8
DM		 205 if (sn == NULL)
206 return;
d7e09d03 207
63418983 208 LASSERT(!sn->sn_timer_active);
d7e09d03
PT		 209
210 sfw_data.fw_session = NULL;
211 atomic_inc(&sfw_data.fw_nzombies);
212 list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
213
214 spin_unlock(&sfw_data.fw_lock);
215
216 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
217 srpc_abort_service(tsc->tsc_srv_service);
218 }
219
220 spin_lock(&sfw_data.fw_lock);
221
63418983 222 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
d7e09d03
PT		 223 if (sfw_batch_active(tsb)) {
224 nactive++;
225 sfw_stop_batch(tsb, 1);
226 }
227 }
228
229 if (nactive != 0)
230 return; /* wait for active batches to stop */
231
232 list_del_init(&sn->sn_list);
233 spin_unlock(&sfw_data.fw_lock);
234
235 sfw_destroy_session(sn);
236
237 spin_lock(&sfw_data.fw_lock);
238}
239
240
8d94b6d2 241static void
63418983 242sfw_session_expired(void *data)
d7e09d03
PT		 243{
244 sfw_session_t *sn = data;
245
246 spin_lock(&sfw_data.fw_lock);
247
63418983
MY		 248 LASSERT(sn->sn_timer_active);
249 LASSERT(sn == sfw_data.fw_session);
d7e09d03 250
63418983 251 CWARN("Session expired! sid: %s-%llu, name: %s\n",
d7e09d03
PT		 252 libcfs_nid2str(sn->sn_id.ses_nid),
253 sn->sn_id.ses_stamp, &sn->sn_name[0]);
254
255 sn->sn_timer_active = 0;
256 sfw_deactivate_session();
257
258 spin_unlock(&sfw_data.fw_lock);
259}
260
261static inline void
262sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
263 unsigned features, const char *name)
264{
265 stt_timer_t *timer = &sn->sn_timer;
266
267 memset(sn, 0, sizeof(sfw_session_t));
268 INIT_LIST_HEAD(&sn->sn_list);
269 INIT_LIST_HEAD(&sn->sn_batches);
270 atomic_set(&sn->sn_refcount, 1); /* +1 for caller */
271 atomic_set(&sn->sn_brw_errors, 0);
272 atomic_set(&sn->sn_ping_errors, 0);
273 strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
274
275 sn->sn_timer_active = 0;
74d68011 276 sn->sn_id = sid;
d7e09d03
PT		 277 sn->sn_features = features;
278 sn->sn_timeout = session_timeout;
279 sn->sn_started = cfs_time_current();
280
281 timer->stt_data = sn;
282 timer->stt_func = sfw_session_expired;
283 INIT_LIST_HEAD(&timer->stt_list);
284}
285
286/* completion handler for incoming framework RPCs */
8d94b6d2 287static void
d7e09d03
PT		 288sfw_server_rpc_done(struct srpc_server_rpc *rpc)
289{
74d68011
MS		 290 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
291 int status = rpc->srpc_status;
d7e09d03 292
63418983 293 CDEBUG(D_NET,
2d00bd17 294 "Incoming framework RPC done: service %s, peer %s, status %s:%d\n",
d7e09d03
PT		 295 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
296 swi_state2str(rpc->srpc_wi.swi_state),
297 status);
298
299 if (rpc->srpc_bulk != NULL)
300 sfw_free_pages(rpc);
301 return;
302}
303
8d94b6d2 304static void
63418983 305sfw_client_rpc_fini(srpc_client_rpc_t *rpc)
d7e09d03 306{
63418983
MY		 307 LASSERT(rpc->crpc_bulk.bk_niov == 0);
308 LASSERT(list_empty(&rpc->crpc_list));
309 LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
d7e09d03 310
63418983 311 CDEBUG(D_NET,
2d00bd17 312 "Outgoing framework RPC done: service %d, peer %s, status %s:%d:%d\n",
d7e09d03
PT		 313 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
314 swi_state2str(rpc->crpc_wi.swi_state),
315 rpc->crpc_aborted, rpc->crpc_status);
316
317 spin_lock(&sfw_data.fw_lock);
318
319 /* my callers must finish all RPCs before shutting me down */
320 LASSERT(!sfw_data.fw_shuttingdown);
321 list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
322
323 spin_unlock(&sfw_data.fw_lock);
324}
325
8d94b6d2 326static sfw_batch_t *
63418983 327sfw_find_batch(lst_bid_t bid)
d7e09d03
PT		 328{
329 sfw_session_t *sn = sfw_data.fw_session;
74d68011 330 sfw_batch_t *bat;
d7e09d03 331
63418983 332 LASSERT(sn != NULL);
d7e09d03 333
63418983 334 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
d7e09d03
PT		 335 if (bat->bat_id.bat_id == bid.bat_id)
336 return bat;
337 }
338
339 return NULL;
340}
341
8d94b6d2 342static sfw_batch_t *
63418983 343sfw_bid2batch(lst_bid_t bid)
d7e09d03
PT		 344{
345 sfw_session_t *sn = sfw_data.fw_session;
74d68011 346 sfw_batch_t *bat;
d7e09d03 347
63418983 348 LASSERT(sn != NULL);
d7e09d03
PT		 349
350 bat = sfw_find_batch(bid);
351 if (bat != NULL)
352 return bat;
353
354 LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
355 if (bat == NULL)
356 return NULL;
357
358 bat->bat_error = 0;
359 bat->bat_session = sn;
360 bat->bat_id = bid;
361 atomic_set(&bat->bat_nactive, 0);
362 INIT_LIST_HEAD(&bat->bat_tests);
363
364 list_add_tail(&bat->bat_list, &sn->sn_batches);
365 return bat;
366}
367
8d94b6d2 368static int
63418983 369sfw_get_stats(srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
d7e09d03 370{
74d68011 371 sfw_session_t *sn = sfw_data.fw_session;
d7e09d03 372 sfw_counters_t *cnt = &reply->str_fw;
74d68011 373 sfw_batch_t *bat;
d7e09d03
PT		 374
375 reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
376
377 if (request->str_sid.ses_nid == LNET_NID_ANY) {
378 reply->str_status = EINVAL;
379 return 0;
380 }
381
382 if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
383 reply->str_status = ESRCH;
384 return 0;
385 }
386
387 lnet_counters_get(&reply->str_lnet);
388 srpc_get_counters(&reply->str_rpc);
389
390 /* send over the msecs since the session was started
391 - with 32 bits to send, this is ~49 days */
70513c5d 392 cnt->running_ms = jiffies_to_msecs(jiffies - sn->sn_started);
d7e09d03
PT		 393 cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
394 cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
395 cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
396
397 cnt->active_batches = 0;
63418983 398 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
d7e09d03
PT		 399 if (atomic_read(&bat->bat_nactive) > 0)
400 cnt->active_batches++;
401 }
402
403 reply->str_status = 0;
404 return 0;
405}
406
407int
408sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
409{
410 sfw_session_t *sn = sfw_data.fw_session;
74d68011 411 srpc_msg_t *msg = container_of(request, srpc_msg_t,
d7e09d03 412 msg_body.mksn_reqst);
74d68011 413 int cplen = 0;
d7e09d03
PT		 414
415 if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
416 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
417 reply->mksn_status = EINVAL;
418 return 0;
419 }
420
421 if (sn != NULL) {
422 reply->mksn_status = 0;
423 reply->mksn_sid = sn->sn_id;
424 reply->mksn_timeout = sn->sn_timeout;
425
426 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
427 atomic_inc(&sn->sn_refcount);
428 return 0;
429 }
430
431 if (!request->mksn_force) {
432 reply->mksn_status = EBUSY;
433 cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
434 sizeof(reply->mksn_name));
435 if (cplen >= sizeof(reply->mksn_name))
436 return -E2BIG;
437 return 0;
438 }
439 }
440
441 /* reject the request if it requires unknown features
442 * NB: old version will always accept all features because it's not
443 * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
444 * harmless because it will return zero feature to console, and it's
445 * console's responsibility to make sure all nodes in a session have
446 * same feature mask. */
447 if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
448 reply->mksn_status = EPROTO;
449 return 0;
450 }
451
452 /* brand new or create by force */
453 LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
454 if (sn == NULL) {
63418983 455 CERROR("Dropping RPC (mksn) under memory pressure.\n");
d7e09d03
PT		 456 return -ENOMEM;
457 }
458
459 sfw_init_session(sn, request->mksn_sid,
460 msg->msg_ses_feats, &request->mksn_name[0]);
461
462 spin_lock(&sfw_data.fw_lock);
463
464 sfw_deactivate_session();
465 LASSERT(sfw_data.fw_session == NULL);
466 sfw_data.fw_session = sn;
467
468 spin_unlock(&sfw_data.fw_lock);
469
470 reply->mksn_status = 0;
471 reply->mksn_sid = sn->sn_id;
472 reply->mksn_timeout = sn->sn_timeout;
473 return 0;
474}
475
8d94b6d2 476static int
63418983 477sfw_remove_session(srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
d7e09d03
PT		 478{
479 sfw_session_t *sn = sfw_data.fw_session;
480
481 reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
482
483 if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
484 reply->rmsn_status = EINVAL;
485 return 0;
486 }
487
488 if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
489 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
490 return 0;
491 }
492
493 if (!atomic_dec_and_test(&sn->sn_refcount)) {
494 reply->rmsn_status = 0;
495 return 0;
496 }
497
498 spin_lock(&sfw_data.fw_lock);
499 sfw_deactivate_session();
500 spin_unlock(&sfw_data.fw_lock);
501
502 reply->rmsn_status = 0;
503 reply->rmsn_sid = LST_INVALID_SID;
504 LASSERT(sfw_data.fw_session == NULL);
505 return 0;
506}
507
8d94b6d2 508static int
63418983 509sfw_debug_session(srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
d7e09d03
PT		 510{
511 sfw_session_t *sn = sfw_data.fw_session;
512
513 if (sn == NULL) {
514 reply->dbg_status = ESRCH;
515 reply->dbg_sid = LST_INVALID_SID;
516 return 0;
517 }
518
519 reply->dbg_status = 0;
520 reply->dbg_sid = sn->sn_id;
521 reply->dbg_timeout = sn->sn_timeout;
522 if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
523 >= sizeof(reply->dbg_name))
524 return -E2BIG;
525
526 return 0;
527}
528
8d94b6d2 529static void
63418983 530sfw_test_rpc_fini(srpc_client_rpc_t *rpc)
d7e09d03 531{
74d68011 532 sfw_test_unit_t *tsu = rpc->crpc_priv;
d7e09d03
PT		 533 sfw_test_instance_t *tsi = tsu->tsu_instance;
534
535 /* Called with hold of tsi->tsi_lock */
63418983 536 LASSERT(list_empty(&rpc->crpc_list));
d7e09d03
PT		 537 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
538}
539
540static inline int
541sfw_test_buffers(sfw_test_instance_t *tsi)
542{
74d68011
MS		 543 struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
544 struct srpc_service *svc = tsc->tsc_srv_service;
545 int nbuf;
d7e09d03
PT		 546
547 nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
548 return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
549}
550
8d94b6d2 551static int
d7e09d03
PT		 552sfw_load_test(struct sfw_test_instance *tsi)
553{
74d68011
MS		 554 struct sfw_test_case *tsc;
555 struct srpc_service *svc;
556 int nbuf;
557 int rc;
d7e09d03
PT		 558
559 LASSERT(tsi != NULL);
560 tsc = sfw_find_test_case(tsi->tsi_service);
561 nbuf = sfw_test_buffers(tsi);
562 LASSERT(tsc != NULL);
563 svc = tsc->tsc_srv_service;
564
565 if (tsi->tsi_is_client) {
566 tsi->tsi_ops = tsc->tsc_cli_ops;
567 return 0;
568 }
569
570 rc = srpc_service_add_buffers(svc, nbuf);
571 if (rc != 0) {
2d00bd17
JP		 572 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
573 svc->sv_name, nbuf, rc);
d7e09d03
PT		 574 /* NB: this error handler is not strictly correct, because
575 * it may release more buffers than already allocated,
576 * but it doesn't matter because request portal should
577 * be lazy portal and will grow buffers if necessary. */
578 srpc_service_remove_buffers(svc, nbuf);
579 return -ENOMEM;
580 }
581
582 CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
583 nbuf * (srpc_serv_is_framework(svc) ?
584 1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
585 return 0;
586}
587
8d94b6d2 588static void
d7e09d03
PT		 589sfw_unload_test(struct sfw_test_instance *tsi)
590{
591 struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
592
593 LASSERT(tsc != NULL);
594
595 if (tsi->tsi_is_client)
596 return;
597
598 /* shrink buffers, because request portal is lazy portal
599 * which can grow buffers at runtime so we may leave
600 * some buffers behind, but never mind... */
601 srpc_service_remove_buffers(tsc->tsc_srv_service,
602 sfw_test_buffers(tsi));
603 return;
604}
605
8d94b6d2 606static void
63418983 607sfw_destroy_test_instance(sfw_test_instance_t *tsi)
d7e09d03
PT		 608{
609 srpc_client_rpc_t *rpc;
74d68011 610 sfw_test_unit_t *tsu;
d7e09d03 611
550d35d8
DM		 612 if (!tsi->tsi_is_client)
613 goto clean;
d7e09d03
PT		 614
615 tsi->tsi_ops->tso_fini(tsi);
616
63418983
MY		 617 LASSERT(!tsi->tsi_stopping);
618 LASSERT(list_empty(&tsi->tsi_active_rpcs));
619 LASSERT(!sfw_test_active(tsi));
d7e09d03
PT		 620
621 while (!list_empty(&tsi->tsi_units)) {
622 tsu = list_entry(tsi->tsi_units.next,
623 sfw_test_unit_t, tsu_list);
624 list_del(&tsu->tsu_list);
625 LIBCFS_FREE(tsu, sizeof(*tsu));
626 }
627
628 while (!list_empty(&tsi->tsi_free_rpcs)) {
629 rpc = list_entry(tsi->tsi_free_rpcs.next,
630 srpc_client_rpc_t, crpc_list);
631 list_del(&rpc->crpc_list);
632 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
633 }
634
635clean:
636 sfw_unload_test(tsi);
637 LIBCFS_FREE(tsi, sizeof(*tsi));
638 return;
639}
640
8d94b6d2 641static void
63418983 642sfw_destroy_batch(sfw_batch_t *tsb)
d7e09d03
PT		 643{
644 sfw_test_instance_t *tsi;
645
63418983
MY		 646 LASSERT(!sfw_batch_active(tsb));
647 LASSERT(list_empty(&tsb->bat_list));
d7e09d03
PT		 648
649 while (!list_empty(&tsb->bat_tests)) {
650 tsi = list_entry(tsb->bat_tests.next,
651 sfw_test_instance_t, tsi_list);
652 list_del_init(&tsi->tsi_list);
653 sfw_destroy_test_instance(tsi);
654 }
655
656 LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
657 return;
658}
659
660void
63418983 661sfw_destroy_session(sfw_session_t *sn)
d7e09d03
PT		 662{
663 sfw_batch_t *batch;
664
63418983
MY		 665 LASSERT(list_empty(&sn->sn_list));
666 LASSERT(sn != sfw_data.fw_session);
d7e09d03
PT		 667
668 while (!list_empty(&sn->sn_batches)) {
669 batch = list_entry(sn->sn_batches.next,
670 sfw_batch_t, bat_list);
671 list_del_init(&batch->bat_list);
672 sfw_destroy_batch(batch);
673 }
674
675 LIBCFS_FREE(sn, sizeof(*sn));
676 atomic_dec(&sfw_data.fw_nzombies);
677 return;
678}
679
8d94b6d2 680static void
d7e09d03
PT		 681sfw_unpack_addtest_req(srpc_msg_t *msg)
682{
683 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
684
63418983
MY		 685 LASSERT(msg->msg_type == SRPC_MSG_TEST_REQST);
686 LASSERT(req->tsr_is_client);
d7e09d03
PT		 687
688 if (msg->msg_magic == SRPC_MSG_MAGIC)
689 return; /* no flipping needed */
690
63418983 691 LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
d7e09d03
PT		 692
693 if (req->tsr_service == SRPC_SERVICE_BRW) {
694 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
695 test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
696
697 __swab32s(&bulk->blk_opc);
698 __swab32s(&bulk->blk_npg);
699 __swab32s(&bulk->blk_flags);
700
701 } else {
702 test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
703
704 __swab16s(&bulk->blk_opc);
705 __swab16s(&bulk->blk_flags);
706 __swab32s(&bulk->blk_offset);
707 __swab32s(&bulk->blk_len);
708 }
709
710 return;
711 }
712
713 if (req->tsr_service == SRPC_SERVICE_PING) {
714 test_ping_req_t *ping = &req->tsr_u.ping;
715
716 __swab32s(&ping->png_size);
717 __swab32s(&ping->png_flags);
718 return;
719 }
720
63418983 721 LBUG();
d7e09d03
PT		 722 return;
723}
724
8d94b6d2 725static int
63418983 726sfw_add_test_instance(sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
d7e09d03 727{
74d68011
MS		 728 srpc_msg_t *msg = &rpc->srpc_reqstbuf->buf_msg;
729 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
730 srpc_bulk_t *bk = rpc->srpc_bulk;
731 int ndest = req->tsr_ndest;
732 sfw_test_unit_t *tsu;
d7e09d03 733 sfw_test_instance_t *tsi;
74d68011
MS		 734 int i;
735 int rc;
d7e09d03
PT		 736
737 LIBCFS_ALLOC(tsi, sizeof(*tsi));
738 if (tsi == NULL) {
63418983 739 CERROR("Can't allocate test instance for batch: %llu\n",
d7e09d03
PT		 740 tsb->bat_id.bat_id);
741 return -ENOMEM;
742 }
743
d7e09d03
PT		 744 spin_lock_init(&tsi->tsi_lock);
745 atomic_set(&tsi->tsi_nactive, 0);
746 INIT_LIST_HEAD(&tsi->tsi_units);
747 INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
748 INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
749
750 tsi->tsi_stopping = 0;
74d68011
MS		 751 tsi->tsi_batch = tsb;
752 tsi->tsi_loop = req->tsr_loop;
753 tsi->tsi_concur = req->tsr_concur;
d7e09d03
PT		 754 tsi->tsi_service = req->tsr_service;
755 tsi->tsi_is_client = !!(req->tsr_is_client);
756 tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
757
758 rc = sfw_load_test(tsi);
759 if (rc != 0) {
760 LIBCFS_FREE(tsi, sizeof(*tsi));
761 return rc;
762 }
763
63418983 764 LASSERT(!sfw_batch_active(tsb));
d7e09d03
PT		 765
766 if (!tsi->tsi_is_client) {
767 /* it's test server, just add it to tsb */
768 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
769 return 0;
770 }
771
63418983
MY		 772 LASSERT(bk != NULL);
773 LASSERT(bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
d7e09d03
PT		 774 LASSERT((unsigned int)bk->bk_len >=
775 sizeof(lnet_process_id_packed_t) * ndest);
776
777 sfw_unpack_addtest_req(msg);
778 memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
779
780 for (i = 0; i < ndest; i++) {
781 lnet_process_id_packed_t *dests;
74d68011
MS		 782 lnet_process_id_packed_t id;
783 int j;
d7e09d03
PT		 784
785 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
63418983 786 LASSERT(dests != NULL); /* my pages are within KVM always */
d7e09d03
PT		 787 id = dests[i % SFW_ID_PER_PAGE];
788 if (msg->msg_magic != SRPC_MSG_MAGIC)
789 sfw_unpack_id(id);
790
791 for (j = 0; j < tsi->tsi_concur; j++) {
792 LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
793 if (tsu == NULL) {
794 rc = -ENOMEM;
63418983 795 CERROR("Can't allocate tsu for %d\n",
d7e09d03
PT		 796 tsi->tsi_service);
797 goto error;
798 }
799
800 tsu->tsu_dest.nid = id.nid;
801 tsu->tsu_dest.pid = id.pid;
802 tsu->tsu_instance = tsi;
803 tsu->tsu_private = NULL;
804 list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
805 }
806 }
807
808 rc = tsi->tsi_ops->tso_init(tsi);
809 if (rc == 0) {
810 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
811 return 0;
812 }
813
814error:
63418983 815 LASSERT(rc != 0);
d7e09d03
PT		 816 sfw_destroy_test_instance(tsi);
817 return rc;
818}
819
820static void
63418983 821sfw_test_unit_done(sfw_test_unit_t *tsu)
d7e09d03
PT		 822{
823 sfw_test_instance_t *tsi = tsu->tsu_instance;
74d68011
MS		 824 sfw_batch_t *tsb = tsi->tsi_batch;
825 sfw_session_t *sn = tsb->bat_session;
d7e09d03 826
63418983 827 LASSERT(sfw_test_active(tsi));
d7e09d03
PT		 828
829 if (!atomic_dec_and_test(&tsi->tsi_nactive))
830 return;
831
832 /* the test instance is done */
833 spin_lock(&tsi->tsi_lock);
834
835 tsi->tsi_stopping = 0;
836
837 spin_unlock(&tsi->tsi_lock);
838
839 spin_lock(&sfw_data.fw_lock);
840
841 if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
842 sn == sfw_data.fw_session) { /* sn also active */
843 spin_unlock(&sfw_data.fw_lock);
844 return;
845 }
846
63418983 847 LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
d7e09d03 848
63418983 849 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
d7e09d03
PT		 850 if (sfw_batch_active(tsb)) {
851 spin_unlock(&sfw_data.fw_lock);
852 return;
853 }
854 }
855
856 list_del_init(&sn->sn_list);
857 spin_unlock(&sfw_data.fw_lock);
858
859 sfw_destroy_session(sn);
860 return;
861}
862
8d94b6d2 863static void
63418983 864sfw_test_rpc_done(srpc_client_rpc_t *rpc)
d7e09d03 865{
74d68011 866 sfw_test_unit_t *tsu = rpc->crpc_priv;
d7e09d03 867 sfw_test_instance_t *tsi = tsu->tsu_instance;
74d68011 868 int done = 0;
d7e09d03
PT		 869
870 tsi->tsi_ops->tso_done_rpc(tsu, rpc);
871
872 spin_lock(&tsi->tsi_lock);
873
63418983
MY		 874 LASSERT(sfw_test_active(tsi));
875 LASSERT(!list_empty(&rpc->crpc_list));
d7e09d03
PT		 876
877 list_del_init(&rpc->crpc_list);
878
879 /* batch is stopping or loop is done or get error */
880 if (tsi->tsi_stopping ||
881 tsu->tsu_loop == 0 ||
882 (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
883 done = 1;
884
885 /* dec ref for poster */
886 srpc_client_rpc_decref(rpc);
887
888 spin_unlock(&tsi->tsi_lock);
889
890 if (!done) {
891 swi_schedule_workitem(&tsu->tsu_worker);
892 return;
893 }
894
895 sfw_test_unit_done(tsu);
896 return;
897}
898
899int
900sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
901 unsigned features, int nblk, int blklen,
902 srpc_client_rpc_t **rpcpp)
903{
74d68011 904 srpc_client_rpc_t *rpc = NULL;
d7e09d03
PT		 905 sfw_test_instance_t *tsi = tsu->tsu_instance;
906
907 spin_lock(&tsi->tsi_lock);
908
63418983 909 LASSERT(sfw_test_active(tsi));
d7e09d03
PT		 910
911 if (!list_empty(&tsi->tsi_free_rpcs)) {
912 /* pick request from buffer */
913 rpc = list_entry(tsi->tsi_free_rpcs.next,
914 srpc_client_rpc_t, crpc_list);
63418983 915 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
d7e09d03
PT		 916 list_del_init(&rpc->crpc_list);
917 }
918
919 spin_unlock(&tsi->tsi_lock);
920
921 if (rpc == NULL) {
922 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
923 blklen, sfw_test_rpc_done,
924 sfw_test_rpc_fini, tsu);
925 } else {
926 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
927 blklen, sfw_test_rpc_done,
928 sfw_test_rpc_fini, tsu);
929 }
930
931 if (rpc == NULL) {
932 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
933 return -ENOMEM;
934 }
935
936 rpc->crpc_reqstmsg.msg_ses_feats = features;
937 *rpcpp = rpc;
938
939 return 0;
940}
941
8d94b6d2 942static int
63418983 943sfw_run_test(swi_workitem_t *wi)
d7e09d03 944{
74d68011 945 sfw_test_unit_t *tsu = wi->swi_workitem.wi_data;
d7e09d03 946 sfw_test_instance_t *tsi = tsu->tsu_instance;
74d68011 947 srpc_client_rpc_t *rpc = NULL;
d7e09d03 948
63418983 949 LASSERT(wi == &tsu->tsu_worker);
d7e09d03
PT		 950
951 if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
63418983 952 LASSERT(rpc == NULL);
d7e09d03
PT		 953 goto test_done;
954 }
955
63418983 956 LASSERT(rpc != NULL);
d7e09d03
PT		 957
958 spin_lock(&tsi->tsi_lock);
959
960 if (tsi->tsi_stopping) {
961 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
962 spin_unlock(&tsi->tsi_lock);
963 goto test_done;
964 }
965
966 if (tsu->tsu_loop > 0)
967 tsu->tsu_loop--;
968
969 list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
970 spin_unlock(&tsi->tsi_lock);
971
972 rpc->crpc_timeout = rpc_timeout;
973
974 spin_lock(&rpc->crpc_lock);
975 srpc_post_rpc(rpc);
976 spin_unlock(&rpc->crpc_lock);
977 return 0;
978
979test_done:
980 /*
981 * No one can schedule me now since:
982 * - previous RPC, if any, has done and
983 * - no new RPC is initiated.
984 * - my batch is still active; no one can run it again now.
985 * Cancel pending schedules and prevent future schedule attempts:
986 */
987 swi_exit_workitem(wi);
988 sfw_test_unit_done(tsu);
989 return 1;
990}
991
8d94b6d2 992static int
63418983 993sfw_run_batch(sfw_batch_t *tsb)
d7e09d03 994{
74d68011
MS		 995 swi_workitem_t *wi;
996 sfw_test_unit_t *tsu;
d7e09d03
PT		 997 sfw_test_instance_t *tsi;
998
999 if (sfw_batch_active(tsb)) {
b0f5aad5 1000 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
d7e09d03
PT		 1001 tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1002 return 0;
1003 }
1004
63418983 1005 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
d7e09d03
PT		 1006 if (!tsi->tsi_is_client) /* skip server instances */
1007 continue;
1008
63418983
MY		 1009 LASSERT(!tsi->tsi_stopping);
1010 LASSERT(!sfw_test_active(tsi));
d7e09d03
PT		 1011
1012 atomic_inc(&tsb->bat_nactive);
1013
63418983 1014 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
d7e09d03
PT		 1015 atomic_inc(&tsi->tsi_nactive);
1016 tsu->tsu_loop = tsi->tsi_loop;
1017 wi = &tsu->tsu_worker;
1018 swi_init_workitem(wi, tsu, sfw_run_test,
1019 lst_sched_test[\
1020 lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1021 swi_schedule_workitem(wi);
1022 }
1023 }
1024
1025 return 0;
1026}
1027
1028int
63418983 1029sfw_stop_batch(sfw_batch_t *tsb, int force)
d7e09d03
PT		 1030{
1031 sfw_test_instance_t *tsi;
74d68011 1032 srpc_client_rpc_t *rpc;
d7e09d03
PT		 1033
1034 if (!sfw_batch_active(tsb)) {
b0f5aad5 1035 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
d7e09d03
PT		 1036 return 0;
1037 }
1038
63418983 1039 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
d7e09d03
PT		 1040 spin_lock(&tsi->tsi_lock);
1041
1042 if (!tsi->tsi_is_client ||
1043 !sfw_test_active(tsi) || tsi->tsi_stopping) {
1044 spin_unlock(&tsi->tsi_lock);
1045 continue;
1046 }
1047
1048 tsi->tsi_stopping = 1;
1049
1050 if (!force) {
1051 spin_unlock(&tsi->tsi_lock);
1052 continue;
1053 }
1054
1055 /* abort launched rpcs in the test */
1056 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1057 spin_lock(&rpc->crpc_lock);
1058
1059 srpc_abort_rpc(rpc, -EINTR);
1060
1061 spin_unlock(&rpc->crpc_lock);
1062 }
1063
1064 spin_unlock(&tsi->tsi_lock);
1065 }
1066
1067 return 0;
1068}
1069
8d94b6d2 1070static int
63418983 1071sfw_query_batch(sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
d7e09d03
PT		 1072{
1073 sfw_test_instance_t *tsi;
1074
1075 if (testidx < 0)
1076 return -EINVAL;
1077
1078 if (testidx == 0) {
1079 reply->bar_active = atomic_read(&tsb->bat_nactive);
1080 return 0;
1081 }
1082
63418983 1083 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
d7e09d03
PT		 1084 if (testidx-- > 1)
1085 continue;
1086
1087 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1088 return 0;
1089 }
1090
1091 return -ENOENT;
1092}
1093
1094void
63418983 1095sfw_free_pages(srpc_server_rpc_t *rpc)
d7e09d03
PT		 1096{
1097 srpc_free_bulk(rpc->srpc_bulk);
1098 rpc->srpc_bulk = NULL;
1099}
1100
1101int
1102sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1103 int sink)
1104{
1105 LASSERT(rpc->srpc_bulk == NULL);
1106 LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1107
1108 rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1109 if (rpc->srpc_bulk == NULL)
1110 return -ENOMEM;
1111
1112 return 0;
1113}
1114
8d94b6d2 1115static int
63418983 1116sfw_add_test(srpc_server_rpc_t *rpc)
d7e09d03 1117{
74d68011 1118 sfw_session_t *sn = sfw_data.fw_session;
d7e09d03
PT		 1119 srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1120 srpc_test_reqst_t *request;
74d68011
MS		 1121 int rc;
1122 sfw_batch_t *bat;
d7e09d03
PT		 1123
1124 request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1125 reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1126
1127 if (request->tsr_loop == 0 ||
1128 request->tsr_concur == 0 ||
1129 request->tsr_sid.ses_nid == LNET_NID_ANY ||
1130 request->tsr_ndest > SFW_MAX_NDESTS ||
1131 (request->tsr_is_client && request->tsr_ndest == 0) ||
1132 request->tsr_concur > SFW_MAX_CONCUR ||
1133 request->tsr_service > SRPC_SERVICE_MAX_ID ||
1134 request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1135 reply->tsr_status = EINVAL;
1136 return 0;
1137 }
1138
1139 if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1140 sfw_find_test_case(request->tsr_service) == NULL) {
1141 reply->tsr_status = ENOENT;
1142 return 0;
1143 }
1144
1145 bat = sfw_bid2batch(request->tsr_bid);
1146 if (bat == NULL) {
63418983 1147 CERROR("Dropping RPC (%s) from %s under memory pressure.\n",
d7e09d03
PT		 1148 rpc->srpc_scd->scd_svc->sv_name,
1149 libcfs_id2str(rpc->srpc_peer));
1150 return -ENOMEM;
1151 }
1152
1153 if (sfw_batch_active(bat)) {
1154 reply->tsr_status = EBUSY;
1155 return 0;
1156 }
1157
1158 if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1159 /* rpc will be resumed later in sfw_bulk_ready */
74d68011
MS		 1160 int npg = sfw_id_pages(request->tsr_ndest);
1161 int len;
d7e09d03
PT		 1162
1163 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1164 len = npg * PAGE_CACHE_SIZE;
1165
1166 } else {
1167 len = sizeof(lnet_process_id_packed_t) *
1168 request->tsr_ndest;
1169 }
1170
1171 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1172 }
1173
1174 rc = sfw_add_test_instance(bat, rpc);
63418983 1175 CDEBUG(rc == 0 ? D_NET : D_WARNING,
d7e09d03
PT		 1176 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1177 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1178 request->tsr_is_client ? "client" : "server",
1179 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1180
1181 reply->tsr_status = (rc < 0) ? -rc : rc;
1182 return 0;
1183}
1184
8d94b6d2 1185static int
63418983 1186sfw_control_batch(srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
d7e09d03
PT		 1187{
1188 sfw_session_t *sn = sfw_data.fw_session;
74d68011
MS		 1189 int rc = 0;
1190 sfw_batch_t *bat;
d7e09d03
PT		 1191
1192 reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1193
1194 if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1195 reply->bar_status = ESRCH;
1196 return 0;
1197 }
1198
1199 bat = sfw_find_batch(request->bar_bid);
1200 if (bat == NULL) {
1201 reply->bar_status = ENOENT;
1202 return 0;
1203 }
1204
1205 switch (request->bar_opc) {
1206 case SRPC_BATCH_OPC_RUN:
1207 rc = sfw_run_batch(bat);
1208 break;
1209
1210 case SRPC_BATCH_OPC_STOP:
1211 rc = sfw_stop_batch(bat, request->bar_arg);
1212 break;
1213
1214 case SRPC_BATCH_OPC_QUERY:
1215 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1216 break;
1217
1218 default:
1219 return -EINVAL; /* drop it */
1220 }
1221
1222 reply->bar_status = (rc < 0) ? -rc : rc;
1223 return 0;
1224}
1225
8d94b6d2 1226static int
d7e09d03
PT		 1227sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1228{
74d68011
MS		 1229 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1230 srpc_msg_t *reply = &rpc->srpc_replymsg;
1231 srpc_msg_t *request = &rpc->srpc_reqstbuf->buf_msg;
1232 unsigned features = LST_FEATS_MASK;
1233 int rc = 0;
d7e09d03
PT		 1234
1235 LASSERT(sfw_data.fw_active_srpc == NULL);
1236 LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1237
1238 spin_lock(&sfw_data.fw_lock);
1239
1240 if (sfw_data.fw_shuttingdown) {
1241 spin_unlock(&sfw_data.fw_lock);
1242 return -ESHUTDOWN;
1243 }
1244
1245 /* Remove timer to avoid racing with it or expiring active session */
1246 if (sfw_del_session_timer() != 0) {
1247 CERROR("Dropping RPC (%s) from %s: racing with expiry timer.",
1248 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1249 spin_unlock(&sfw_data.fw_lock);
1250 return -EAGAIN;
1251 }
1252
1253 sfw_data.fw_active_srpc = rpc;
1254 spin_unlock(&sfw_data.fw_lock);
1255
1256 sfw_unpack_message(request);
1257 LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1258
1259 /* rpc module should have checked this */
1260 LASSERT(request->msg_version == SRPC_MSG_VERSION);
1261
1262 if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1263 sv->sv_id != SRPC_SERVICE_DEBUG) {
1264 sfw_session_t *sn = sfw_data.fw_session;
1265
1266 if (sn != NULL &&
1267 sn->sn_features != request->msg_ses_feats) {
2d00bd17 1268 CNETERR("Features of framework RPC don't match features of current session: %x/%x\n",
d7e09d03
PT		 1269 request->msg_ses_feats, sn->sn_features);
1270 reply->msg_body.reply.status = EPROTO;
1271 reply->msg_body.reply.sid = sn->sn_id;
1272 goto out;
1273 }
1274
1275 } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1276 /* NB: at this point, old version will ignore features and
1277 * create new session anyway, so console should be able
1278 * to handle this */
1279 reply->msg_body.reply.status = EPROTO;
1280 goto out;
1281 }
1282
a58a38ac 1283 switch (sv->sv_id) {
d7e09d03 1284 default:
63418983 1285 LBUG();
d7e09d03
PT		 1286 case SRPC_SERVICE_TEST:
1287 rc = sfw_add_test(rpc);
1288 break;
1289
1290 case SRPC_SERVICE_BATCH:
1291 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1292 &reply->msg_body.bat_reply);
1293 break;
1294
1295 case SRPC_SERVICE_QUERY_STAT:
1296 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1297 &reply->msg_body.stat_reply);
1298 break;
1299
1300 case SRPC_SERVICE_DEBUG:
1301 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1302 &reply->msg_body.dbg_reply);
1303 break;
1304
1305 case SRPC_SERVICE_MAKE_SESSION:
1306 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1307 &reply->msg_body.mksn_reply);
1308 break;
1309
1310 case SRPC_SERVICE_REMOVE_SESSION:
1311 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1312 &reply->msg_body.rmsn_reply);
1313 break;
1314 }
1315
1316 if (sfw_data.fw_session != NULL)
1317 features = sfw_data.fw_session->sn_features;
1318 out:
1319 reply->msg_ses_feats = features;
1320 rpc->srpc_done = sfw_server_rpc_done;
1321 spin_lock(&sfw_data.fw_lock);
1322
1323 if (!sfw_data.fw_shuttingdown)
1324 sfw_add_session_timer();
1325
1326 sfw_data.fw_active_srpc = NULL;
1327 spin_unlock(&sfw_data.fw_lock);
1328 return rc;
1329}
1330
8d94b6d2 1331static int
d7e09d03
PT		 1332sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1333{
74d68011
MS		 1334 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1335 int rc;
d7e09d03
PT		 1336
1337 LASSERT(rpc->srpc_bulk != NULL);
1338 LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1339 LASSERT(sfw_data.fw_active_srpc == NULL);
1340 LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1341
1342 spin_lock(&sfw_data.fw_lock);
1343
1344 if (status != 0) {
2d00bd17 1345 CERROR("Bulk transfer failed for RPC: service %s, peer %s, status %d\n",
d7e09d03
PT		 1346 sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1347 spin_unlock(&sfw_data.fw_lock);
1348 return -EIO;
1349 }
1350
1351 if (sfw_data.fw_shuttingdown) {
1352 spin_unlock(&sfw_data.fw_lock);
1353 return -ESHUTDOWN;
1354 }
1355
1356 if (sfw_del_session_timer() != 0) {
1357 CERROR("Dropping RPC (%s) from %s: racing with expiry timer",
1358 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1359 spin_unlock(&sfw_data.fw_lock);
1360 return -EAGAIN;
1361 }
1362
1363 sfw_data.fw_active_srpc = rpc;
1364 spin_unlock(&sfw_data.fw_lock);
1365
1366 rc = sfw_add_test(rpc);
1367
1368 spin_lock(&sfw_data.fw_lock);
1369
1370 if (!sfw_data.fw_shuttingdown)
1371 sfw_add_session_timer();
1372
1373 sfw_data.fw_active_srpc = NULL;
1374 spin_unlock(&sfw_data.fw_lock);
1375 return rc;
1376}
1377
1378srpc_client_rpc_t *
1379sfw_create_rpc(lnet_process_id_t peer, int service,
1380 unsigned features, int nbulkiov, int bulklen,
1381 void (*done)(srpc_client_rpc_t *), void *priv)
1382{
1383 srpc_client_rpc_t *rpc = NULL;
1384
1385 spin_lock(&sfw_data.fw_lock);
1386
63418983
MY		 1387 LASSERT(!sfw_data.fw_shuttingdown);
1388 LASSERT(service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
d7e09d03
PT		 1389
1390 if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1391 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1392 srpc_client_rpc_t, crpc_list);
1393 list_del(&rpc->crpc_list);
1394
1395 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1396 done, sfw_client_rpc_fini, priv);
1397 }
1398
1399 spin_unlock(&sfw_data.fw_lock);
1400
1401 if (rpc == NULL) {
1402 rpc = srpc_create_client_rpc(peer, service,
1403 nbulkiov, bulklen, done,
1404 nbulkiov != 0 ? NULL :
1405 sfw_client_rpc_fini,
1406 priv);
1407 }
1408
1409 if (rpc != NULL) /* "session" is concept in framework */
1410 rpc->crpc_reqstmsg.msg_ses_feats = features;
1411
1412 return rpc;
1413}
1414
1415void
63418983 1416sfw_unpack_message(srpc_msg_t *msg)
d7e09d03
PT		 1417{
1418 if (msg->msg_magic == SRPC_MSG_MAGIC)
1419 return; /* no flipping needed */
1420
1421 /* srpc module should guarantee I wouldn't get crap */
63418983 1422 LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
d7e09d03
PT		 1423
1424 if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1425 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1426
1427 __swab32s(&req->str_type);
1428 __swab64s(&req->str_rpyid);
1429 sfw_unpack_sid(req->str_sid);
1430 return;
1431 }
1432
1433 if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1434 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1435
1436 __swab32s(&rep->str_status);
1437 sfw_unpack_sid(rep->str_sid);
1438 sfw_unpack_fw_counters(rep->str_fw);
1439 sfw_unpack_rpc_counters(rep->str_rpc);
1440 sfw_unpack_lnet_counters(rep->str_lnet);
1441 return;
1442 }
1443
1444 if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1445 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1446
1447 __swab64s(&req->mksn_rpyid);
1448 __swab32s(&req->mksn_force);
1449 sfw_unpack_sid(req->mksn_sid);
1450 return;
1451 }
1452
1453 if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1454 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1455
1456 __swab32s(&rep->mksn_status);
1457 __swab32s(&rep->mksn_timeout);
1458 sfw_unpack_sid(rep->mksn_sid);
1459 return;
1460 }
1461
1462 if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1463 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1464
1465 __swab64s(&req->rmsn_rpyid);
1466 sfw_unpack_sid(req->rmsn_sid);
1467 return;
1468 }
1469
1470 if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1471 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1472
1473 __swab32s(&rep->rmsn_status);
1474 sfw_unpack_sid(rep->rmsn_sid);
1475 return;
1476 }
1477
1478 if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1479 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1480
1481 __swab64s(&req->dbg_rpyid);
1482 __swab32s(&req->dbg_flags);
1483 sfw_unpack_sid(req->dbg_sid);
1484 return;
1485 }
1486
1487 if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1488 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1489
1490 __swab32s(&rep->dbg_nbatch);
1491 __swab32s(&rep->dbg_timeout);
1492 sfw_unpack_sid(rep->dbg_sid);
1493 return;
1494 }
1495
1496 if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1497 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1498
1499 __swab32s(&req->bar_opc);
1500 __swab64s(&req->bar_rpyid);
1501 __swab32s(&req->bar_testidx);
1502 __swab32s(&req->bar_arg);
1503 sfw_unpack_sid(req->bar_sid);
1504 __swab64s(&req->bar_bid.bat_id);
1505 return;
1506 }
1507
1508 if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1509 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1510
1511 __swab32s(&rep->bar_status);
1512 sfw_unpack_sid(rep->bar_sid);
1513 return;
1514 }
1515
1516 if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1517 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1518
1519 __swab64s(&req->tsr_rpyid);
1520 __swab64s(&req->tsr_bulkid);
1521 __swab32s(&req->tsr_loop);
1522 __swab32s(&req->tsr_ndest);
1523 __swab32s(&req->tsr_concur);
1524 __swab32s(&req->tsr_service);
1525 sfw_unpack_sid(req->tsr_sid);
1526 __swab64s(&req->tsr_bid.bat_id);
1527 return;
1528 }
1529
1530 if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1531 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1532
1533 __swab32s(&rep->tsr_status);
1534 sfw_unpack_sid(rep->tsr_sid);
1535 return;
1536 }
1537
1538 if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1539 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1540
1541 __swab64s(&req->join_rpyid);
1542 sfw_unpack_sid(req->join_sid);
1543 return;
1544 }
1545
1546 if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1547 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1548
1549 __swab32s(&rep->join_status);
1550 __swab32s(&rep->join_timeout);
1551 sfw_unpack_sid(rep->join_sid);
1552 return;
1553 }
1554
63418983 1555 LBUG();
d7e09d03
PT		 1556 return;
1557}
1558
1559void
63418983 1560sfw_abort_rpc(srpc_client_rpc_t *rpc)
d7e09d03
PT		 1561{
1562 LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1563 LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1564
1565 spin_lock(&rpc->crpc_lock);
1566 srpc_abort_rpc(rpc, -EINTR);
1567 spin_unlock(&rpc->crpc_lock);
1568 return;
1569}
1570
1571void
63418983 1572sfw_post_rpc(srpc_client_rpc_t *rpc)
d7e09d03
PT		 1573{
1574 spin_lock(&rpc->crpc_lock);
1575
63418983
MY		 1576 LASSERT(!rpc->crpc_closed);
1577 LASSERT(!rpc->crpc_aborted);
1578 LASSERT(list_empty(&rpc->crpc_list));
1579 LASSERT(!sfw_data.fw_shuttingdown);
d7e09d03
PT		 1580
1581 rpc->crpc_timeout = rpc_timeout;
1582 srpc_post_rpc(rpc);
1583
1584 spin_unlock(&rpc->crpc_lock);
1585 return;
1586}
1587
9d0b2b7a 1588static srpc_service_t sfw_services[] = {
d7e09d03
PT		 1589 {
1590 /* sv_id */ SRPC_SERVICE_DEBUG,
1591 /* sv_name */ "debug",
1592 0
1593 },
1594 {
1595 /* sv_id */ SRPC_SERVICE_QUERY_STAT,
1596 /* sv_name */ "query stats",
1597 0
1598 },
1599 {
1600 /* sv_id */ SRPC_SERVICE_MAKE_SESSION,
1601 /* sv_name */ "make session",
1602 0
1603 },
1604 {
1605 /* sv_id */ SRPC_SERVICE_REMOVE_SESSION,
1606 /* sv_name */ "remove session",
1607 0
1608 },
1609 {
1610 /* sv_id */ SRPC_SERVICE_BATCH,
1611 /* sv_name */ "batch service",
1612 0
1613 },
1614 {
1615 /* sv_id */ SRPC_SERVICE_TEST,
1616 /* sv_name */ "test service",
1617 0
1618 },
1619 {
1620 /* sv_id */ 0,
1621 /* sv_name */ NULL,
1622 0
1623 }
1624};
1625
1626extern sfw_test_client_ops_t ping_test_client;
1627extern srpc_service_t ping_test_service;
1628extern void ping_init_test_client(void);
1629extern void ping_init_test_service(void);
1630
1631extern sfw_test_client_ops_t brw_test_client;
1632extern srpc_service_t brw_test_service;
1633extern void brw_init_test_client(void);
1634extern void brw_init_test_service(void);
1635
1636
1637int
63418983 1638sfw_startup(void)
d7e09d03 1639{
74d68011
MS		 1640 int i;
1641 int rc;
1642 int error;
1643 srpc_service_t *sv;
d7e09d03
PT		 1644 sfw_test_case_t *tsc;
1645
1646
1647 if (session_timeout < 0) {
63418983 1648 CERROR("Session timeout must be non-negative: %d\n",
d7e09d03
PT		 1649 session_timeout);
1650 return -EINVAL;
1651 }
1652
1653 if (rpc_timeout < 0) {
63418983 1654 CERROR("RPC timeout must be non-negative: %d\n",
d7e09d03
PT		 1655 rpc_timeout);
1656 return -EINVAL;
1657 }
1658
1659 if (session_timeout == 0)
2d00bd17 1660 CWARN("Zero session_timeout specified - test sessions never expire.\n");
d7e09d03
PT		 1661
1662 if (rpc_timeout == 0)
2d00bd17 1663 CWARN("Zero rpc_timeout specified - test RPC never expire.\n");
d7e09d03
PT		 1664
1665 memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1666
1667 sfw_data.fw_session = NULL;
1668 sfw_data.fw_active_srpc = NULL;
1669 spin_lock_init(&sfw_data.fw_lock);
1670 atomic_set(&sfw_data.fw_nzombies, 0);
1671 INIT_LIST_HEAD(&sfw_data.fw_tests);
1672 INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1673 INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1674
1675 brw_init_test_client();
1676 brw_init_test_service();
1677 rc = sfw_register_test(&brw_test_service, &brw_test_client);
63418983 1678 LASSERT(rc == 0);
d7e09d03
PT		 1679
1680 ping_init_test_client();
1681 ping_init_test_service();
1682 rc = sfw_register_test(&ping_test_service, &ping_test_client);
63418983 1683 LASSERT(rc == 0);
d7e09d03
PT		 1684
1685 error = 0;
63418983 1686 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
d7e09d03
PT		 1687 sv = tsc->tsc_srv_service;
1688
1689 rc = srpc_add_service(sv);
63418983 1690 LASSERT(rc != -EBUSY);
d7e09d03 1691 if (rc != 0) {
63418983 1692 CWARN("Failed to add %s service: %d\n",
d7e09d03
PT		 1693 sv->sv_name, rc);
1694 error = rc;
1695 }
1696 }
1697
1698 for (i = 0; ; i++) {
1699 sv = &sfw_services[i];
550d35d8
DM		 1700 if (sv->sv_name == NULL)
1701 break;
d7e09d03
PT		 1702
1703 sv->sv_bulk_ready = NULL;
1704 sv->sv_handler = sfw_handle_server_rpc;
1705 sv->sv_wi_total = SFW_FRWK_WI_MAX;
1706 if (sv->sv_id == SRPC_SERVICE_TEST)
1707 sv->sv_bulk_ready = sfw_bulk_ready;
1708
1709 rc = srpc_add_service(sv);
63418983 1710 LASSERT(rc != -EBUSY);
d7e09d03 1711 if (rc != 0) {
63418983 1712 CWARN("Failed to add %s service: %d\n",
d7e09d03
PT		 1713 sv->sv_name, rc);
1714 error = rc;
1715 }
1716
1717 /* about to sfw_shutdown, no need to add buffer */
550d35d8
DM		 1718 if (error)
1719 continue;
d7e09d03
PT		 1720
1721 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1722 if (rc != 0) {
2d00bd17 1723 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
d7e09d03
PT		 1724 sv->sv_name, sv->sv_wi_total, rc);
1725 error = -ENOMEM;
1726 }
1727 }
1728
1729 if (error != 0)
1730 sfw_shutdown();
1731 return error;
1732}
1733
1734void
63418983 1735sfw_shutdown(void)
d7e09d03 1736{
74d68011 1737 srpc_service_t *sv;
d7e09d03 1738 sfw_test_case_t *tsc;
74d68011 1739 int i;
d7e09d03
PT		 1740
1741 spin_lock(&sfw_data.fw_lock);
1742
1743 sfw_data.fw_shuttingdown = 1;
1744 lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1745 "waiting for active RPC to finish.\n");
1746
1747 if (sfw_del_session_timer() != 0)
1748 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1749 "waiting for session timer to explode.\n");
1750
1751 sfw_deactivate_session();
1752 lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1753 sfw_data.fw_lock,
1754 "waiting for %d zombie sessions to die.\n",
1755 atomic_read(&sfw_data.fw_nzombies));
1756
1757 spin_unlock(&sfw_data.fw_lock);
1758
1759 for (i = 0; ; i++) {
1760 sv = &sfw_services[i];
1761 if (sv->sv_name == NULL)
1762 break;
1763
1764 srpc_shutdown_service(sv);
1765 srpc_remove_service(sv);
1766 }
1767
63418983 1768 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
d7e09d03
PT		 1769 sv = tsc->tsc_srv_service;
1770 srpc_shutdown_service(sv);
1771 srpc_remove_service(sv);
1772 }
1773
1774 while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1775 srpc_client_rpc_t *rpc;
1776
1777 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1778 srpc_client_rpc_t, crpc_list);
1779 list_del(&rpc->crpc_list);
1780
1781 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1782 }
1783
1784 for (i = 0; ; i++) {
1785 sv = &sfw_services[i];
1786 if (sv->sv_name == NULL)
1787 break;
1788
1789 srpc_wait_service_shutdown(sv);
1790 }
1791
1792 while (!list_empty(&sfw_data.fw_tests)) {
1793 tsc = list_entry(sfw_data.fw_tests.next,
1794 sfw_test_case_t, tsc_list);
1795
1796 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1797
1798 list_del(&tsc->tsc_list);
1799 LIBCFS_FREE(tsc, sizeof(*tsc));
1800 }
1801
1802 return;
1803}
Linux kernel - Deliverables
This page took 0.384953 seconds and 5 git commands to generate.