projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'powerpc-4.6-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[deliverable/linux.git] / drivers / staging / lustre / lnet / selftest / framework.c
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
46 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
47
48 static int session_timeout = 100;
49 module_param(session_timeout, int, 0444);
50 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
51
52 static int rpc_timeout = 64;
53 module_param(rpc_timeout, int, 0644);
54 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
55
56 #define sfw_unpack_id(id) \
57 do { \
58 __swab64s(&(id).nid); \
59 __swab32s(&(id).pid); \
60 } while (0)
61
62 #define sfw_unpack_sid(sid) \
63 do { \
64 __swab64s(&(sid).ses_nid); \
65 __swab64s(&(sid).ses_stamp); \
66 } while (0)
67
68 #define sfw_unpack_fw_counters(fc) \
69 do { \
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) \
78 do { \
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) \
89 do { \
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))
104 #define sfw_batch_active(b) (atomic_read(&(b)->bat_nactive))
105
106 static struct smoketest_framework {
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 struct srpc_server_rpc *fw_active_srpc;/* running RPC */
115 } sfw_data;
116
117 /* forward ref's */
118 int sfw_stop_batch(sfw_batch_t *tsb, int force);
119 void sfw_destroy_session(sfw_session_t *sn);
120
121 static inline sfw_test_case_t *
122 sfw_find_test_case(int id)
123 {
124 sfw_test_case_t *tsc;
125
126 LASSERT(id <= SRPC_SERVICE_MAX_ID);
127 LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
128
129 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
130 if (tsc->tsc_srv_service->sv_id == id)
131 return tsc;
132 }
133
134 return NULL;
135 }
136
137 static int
138 sfw_register_test(srpc_service_t *service, sfw_test_client_ops_t *cliops)
139 {
140 sfw_test_case_t *tsc;
141
142 if (sfw_find_test_case(service->sv_id)) {
143 CERROR("Failed to register test %s (%d)\n",
144 service->sv_name, service->sv_id);
145 return -EEXIST;
146 }
147
148 LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
149 if (!tsc)
150 return -ENOMEM;
151
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
159 static void
160 sfw_add_session_timer(void)
161 {
162 sfw_session_t *sn = sfw_data.fw_session;
163 struct stt_timer *timer = &sn->sn_timer;
164
165 LASSERT(!sfw_data.fw_shuttingdown);
166
167 if (!sn || !sn->sn_timeout)
168 return;
169
170 LASSERT(!sn->sn_timer_active);
171
172 sn->sn_timer_active = 1;
173 timer->stt_expires = ktime_get_real_seconds() + sn->sn_timeout;
174 stt_add_timer(timer);
175 }
176
177 static int
178 sfw_del_session_timer(void)
179 {
180 sfw_session_t *sn = sfw_data.fw_session;
181
182 if (!sn || !sn->sn_timer_active)
183 return 0;
184
185 LASSERT(sn->sn_timeout);
186
187 if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
188 sn->sn_timer_active = 0;
189 return 0;
190 }
191
192 return EBUSY; /* racing with sfw_session_expired() */
193 }
194
195 static void
196 sfw_deactivate_session(void)
197 __must_hold(&sfw_data.fw_lock)
198 {
199 sfw_session_t *sn = sfw_data.fw_session;
200 int nactive = 0;
201 sfw_batch_t *tsb;
202 sfw_test_case_t *tsc;
203
204 if (!sn)
205 return;
206
207 LASSERT(!sn->sn_timer_active);
208
209 sfw_data.fw_session = NULL;
210 atomic_inc(&sfw_data.fw_nzombies);
211 list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
212
213 spin_unlock(&sfw_data.fw_lock);
214
215 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
216 srpc_abort_service(tsc->tsc_srv_service);
217 }
218
219 spin_lock(&sfw_data.fw_lock);
220
221 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
222 if (sfw_batch_active(tsb)) {
223 nactive++;
224 sfw_stop_batch(tsb, 1);
225 }
226 }
227
228 if (nactive)
229 return; /* wait for active batches to stop */
230
231 list_del_init(&sn->sn_list);
232 spin_unlock(&sfw_data.fw_lock);
233
234 sfw_destroy_session(sn);
235
236 spin_lock(&sfw_data.fw_lock);
237 }
238
239 static void
240 sfw_session_expired(void *data)
241 {
242 sfw_session_t *sn = data;
243
244 spin_lock(&sfw_data.fw_lock);
245
246 LASSERT(sn->sn_timer_active);
247 LASSERT(sn == sfw_data.fw_session);
248
249 CWARN("Session expired! sid: %s-%llu, name: %s\n",
250 libcfs_nid2str(sn->sn_id.ses_nid),
251 sn->sn_id.ses_stamp, &sn->sn_name[0]);
252
253 sn->sn_timer_active = 0;
254 sfw_deactivate_session();
255
256 spin_unlock(&sfw_data.fw_lock);
257 }
258
259 static inline void
260 sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
261 unsigned features, const char *name)
262 {
263 struct stt_timer *timer = &sn->sn_timer;
264
265 memset(sn, 0, sizeof(sfw_session_t));
266 INIT_LIST_HEAD(&sn->sn_list);
267 INIT_LIST_HEAD(&sn->sn_batches);
268 atomic_set(&sn->sn_refcount, 1); /* +1 for caller */
269 atomic_set(&sn->sn_brw_errors, 0);
270 atomic_set(&sn->sn_ping_errors, 0);
271 strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
272
273 sn->sn_timer_active = 0;
274 sn->sn_id = sid;
275 sn->sn_features = features;
276 sn->sn_timeout = session_timeout;
277 sn->sn_started = cfs_time_current();
278
279 timer->stt_data = sn;
280 timer->stt_func = sfw_session_expired;
281 INIT_LIST_HEAD(&timer->stt_list);
282 }
283
284 /* completion handler for incoming framework RPCs */
285 static void
286 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
287 {
288 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
289 int status = rpc->srpc_status;
290
291 CDEBUG(D_NET, "Incoming framework RPC done: service %s, peer %s, status %s:%d\n",
292 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
293 swi_state2str(rpc->srpc_wi.swi_state),
294 status);
295
296 if (rpc->srpc_bulk)
297 sfw_free_pages(rpc);
298 }
299
300 static void
301 sfw_client_rpc_fini(srpc_client_rpc_t *rpc)
302 {
303 LASSERT(!rpc->crpc_bulk.bk_niov);
304 LASSERT(list_empty(&rpc->crpc_list));
305 LASSERT(!atomic_read(&rpc->crpc_refcount));
306
307 CDEBUG(D_NET, "Outgoing framework RPC done: service %d, peer %s, status %s:%d:%d\n",
308 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
309 swi_state2str(rpc->crpc_wi.swi_state),
310 rpc->crpc_aborted, rpc->crpc_status);
311
312 spin_lock(&sfw_data.fw_lock);
313
314 /* my callers must finish all RPCs before shutting me down */
315 LASSERT(!sfw_data.fw_shuttingdown);
316 list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
317
318 spin_unlock(&sfw_data.fw_lock);
319 }
320
321 static sfw_batch_t *
322 sfw_find_batch(lst_bid_t bid)
323 {
324 sfw_session_t *sn = sfw_data.fw_session;
325 sfw_batch_t *bat;
326
327 LASSERT(sn);
328
329 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
330 if (bat->bat_id.bat_id == bid.bat_id)
331 return bat;
332 }
333
334 return NULL;
335 }
336
337 static sfw_batch_t *
338 sfw_bid2batch(lst_bid_t bid)
339 {
340 sfw_session_t *sn = sfw_data.fw_session;
341 sfw_batch_t *bat;
342
343 LASSERT(sn);
344
345 bat = sfw_find_batch(bid);
346 if (bat)
347 return bat;
348
349 LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
350 if (!bat)
351 return NULL;
352
353 bat->bat_error = 0;
354 bat->bat_session = sn;
355 bat->bat_id = bid;
356 atomic_set(&bat->bat_nactive, 0);
357 INIT_LIST_HEAD(&bat->bat_tests);
358
359 list_add_tail(&bat->bat_list, &sn->sn_batches);
360 return bat;
361 }
362
363 static int
364 sfw_get_stats(srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
365 {
366 sfw_session_t *sn = sfw_data.fw_session;
367 sfw_counters_t *cnt = &reply->str_fw;
368 sfw_batch_t *bat;
369
370 reply->str_sid = !sn ? LST_INVALID_SID : sn->sn_id;
371
372 if (request->str_sid.ses_nid == LNET_NID_ANY) {
373 reply->str_status = EINVAL;
374 return 0;
375 }
376
377 if (!sn || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
378 reply->str_status = ESRCH;
379 return 0;
380 }
381
382 lnet_counters_get(&reply->str_lnet);
383 srpc_get_counters(&reply->str_rpc);
384
385 /*
386 * send over the msecs since the session was started
387 * with 32 bits to send, this is ~49 days
388 */
389 cnt->running_ms = jiffies_to_msecs(jiffies - sn->sn_started);
390 cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
391 cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
392 cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
393
394 cnt->active_batches = 0;
395 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
396 if (atomic_read(&bat->bat_nactive) > 0)
397 cnt->active_batches++;
398 }
399
400 reply->str_status = 0;
401 return 0;
402 }
403
404 int
405 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
406 {
407 sfw_session_t *sn = sfw_data.fw_session;
408 srpc_msg_t *msg = container_of(request, srpc_msg_t,
409 msg_body.mksn_reqst);
410 int cplen = 0;
411
412 if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
413 reply->mksn_sid = !sn ? LST_INVALID_SID : sn->sn_id;
414 reply->mksn_status = EINVAL;
415 return 0;
416 }
417
418 if (sn) {
419 reply->mksn_status = 0;
420 reply->mksn_sid = sn->sn_id;
421 reply->mksn_timeout = sn->sn_timeout;
422
423 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
424 atomic_inc(&sn->sn_refcount);
425 return 0;
426 }
427
428 if (!request->mksn_force) {
429 reply->mksn_status = EBUSY;
430 cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
431 sizeof(reply->mksn_name));
432 if (cplen >= sizeof(reply->mksn_name))
433 return -E2BIG;
434 return 0;
435 }
436 }
437
438 /*
439 * reject the request if it requires unknown features
440 * NB: old version will always accept all features because it's not
441 * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
442 * harmless because it will return zero feature to console, and it's
443 * console's responsibility to make sure all nodes in a session have
444 * same feature mask.
445 */
446 if (msg->msg_ses_feats & ~LST_FEATS_MASK) {
447 reply->mksn_status = EPROTO;
448 return 0;
449 }
450
451 /* brand new or create by force */
452 LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
453 if (!sn) {
454 CERROR("dropping RPC mksn under memory pressure\n");
455 return -ENOMEM;
456 }
457
458 sfw_init_session(sn, request->mksn_sid,
459 msg->msg_ses_feats, &request->mksn_name[0]);
460
461 spin_lock(&sfw_data.fw_lock);
462
463 sfw_deactivate_session();
464 LASSERT(!sfw_data.fw_session);
465 sfw_data.fw_session = sn;
466
467 spin_unlock(&sfw_data.fw_lock);
468
469 reply->mksn_status = 0;
470 reply->mksn_sid = sn->sn_id;
471 reply->mksn_timeout = sn->sn_timeout;
472 return 0;
473 }
474
475 static int
476 sfw_remove_session(srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
477 {
478 sfw_session_t *sn = sfw_data.fw_session;
479
480 reply->rmsn_sid = !sn ? LST_INVALID_SID : sn->sn_id;
481
482 if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
483 reply->rmsn_status = EINVAL;
484 return 0;
485 }
486
487 if (!sn || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
488 reply->rmsn_status = !sn ? ESRCH : EBUSY;
489 return 0;
490 }
491
492 if (!atomic_dec_and_test(&sn->sn_refcount)) {
493 reply->rmsn_status = 0;
494 return 0;
495 }
496
497 spin_lock(&sfw_data.fw_lock);
498 sfw_deactivate_session();
499 spin_unlock(&sfw_data.fw_lock);
500
501 reply->rmsn_status = 0;
502 reply->rmsn_sid = LST_INVALID_SID;
503 LASSERT(!sfw_data.fw_session);
504 return 0;
505 }
506
507 static int
508 sfw_debug_session(srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
509 {
510 sfw_session_t *sn = sfw_data.fw_session;
511
512 if (!sn) {
513 reply->dbg_status = ESRCH;
514 reply->dbg_sid = LST_INVALID_SID;
515 return 0;
516 }
517
518 reply->dbg_status = 0;
519 reply->dbg_sid = sn->sn_id;
520 reply->dbg_timeout = sn->sn_timeout;
521 if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
522 >= sizeof(reply->dbg_name))
523 return -E2BIG;
524
525 return 0;
526 }
527
528 static void
529 sfw_test_rpc_fini(srpc_client_rpc_t *rpc)
530 {
531 sfw_test_unit_t *tsu = rpc->crpc_priv;
532 sfw_test_instance_t *tsi = tsu->tsu_instance;
533
534 /* Called with hold of tsi->tsi_lock */
535 LASSERT(list_empty(&rpc->crpc_list));
536 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
537 }
538
539 static inline int
540 sfw_test_buffers(sfw_test_instance_t *tsi)
541 {
542 struct sfw_test_case *tsc;
543 struct srpc_service *svc;
544 int nbuf;
545
546 LASSERT(tsi);
547 tsc = sfw_find_test_case(tsi->tsi_service);
548 LASSERT(tsc);
549 svc = tsc->tsc_srv_service;
550 LASSERT(svc);
551
552 nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
553 return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
554 }
555
556 static int
557 sfw_load_test(struct sfw_test_instance *tsi)
558 {
559 struct sfw_test_case *tsc;
560 struct srpc_service *svc;
561 int nbuf;
562 int rc;
563
564 LASSERT(tsi);
565 tsc = sfw_find_test_case(tsi->tsi_service);
566 nbuf = sfw_test_buffers(tsi);
567 LASSERT(tsc);
568 svc = tsc->tsc_srv_service;
569
570 if (tsi->tsi_is_client) {
571 tsi->tsi_ops = tsc->tsc_cli_ops;
572 return 0;
573 }
574
575 rc = srpc_service_add_buffers(svc, nbuf);
576 if (rc) {
577 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
578 svc->sv_name, nbuf, rc);
579 /*
580 * NB: this error handler is not strictly correct, because
581 * it may release more buffers than already allocated,
582 * but it doesn't matter because request portal should
583 * be lazy portal and will grow buffers if necessary.
584 */
585 srpc_service_remove_buffers(svc, nbuf);
586 return -ENOMEM;
587 }
588
589 CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
590 nbuf * (srpc_serv_is_framework(svc) ?
591 2 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
592 return 0;
593 }
594
595 static void
596 sfw_unload_test(struct sfw_test_instance *tsi)
597 {
598 struct sfw_test_case *tsc;
599
600 LASSERT(tsi);
601 tsc = sfw_find_test_case(tsi->tsi_service);
602 LASSERT(tsc);
603
604 if (tsi->tsi_is_client)
605 return;
606
607 /*
608 * shrink buffers, because request portal is lazy portal
609 * which can grow buffers at runtime so we may leave
610 * some buffers behind, but never mind...
611 */
612 srpc_service_remove_buffers(tsc->tsc_srv_service,
613 sfw_test_buffers(tsi));
614 }
615
616 static void
617 sfw_destroy_test_instance(sfw_test_instance_t *tsi)
618 {
619 srpc_client_rpc_t *rpc;
620 sfw_test_unit_t *tsu;
621
622 if (!tsi->tsi_is_client)
623 goto clean;
624
625 tsi->tsi_ops->tso_fini(tsi);
626
627 LASSERT(!tsi->tsi_stopping);
628 LASSERT(list_empty(&tsi->tsi_active_rpcs));
629 LASSERT(!sfw_test_active(tsi));
630
631 while (!list_empty(&tsi->tsi_units)) {
632 tsu = list_entry(tsi->tsi_units.next,
633 sfw_test_unit_t, tsu_list);
634 list_del(&tsu->tsu_list);
635 LIBCFS_FREE(tsu, sizeof(*tsu));
636 }
637
638 while (!list_empty(&tsi->tsi_free_rpcs)) {
639 rpc = list_entry(tsi->tsi_free_rpcs.next,
640 srpc_client_rpc_t, crpc_list);
641 list_del(&rpc->crpc_list);
642 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
643 }
644
645 clean:
646 sfw_unload_test(tsi);
647 LIBCFS_FREE(tsi, sizeof(*tsi));
648 }
649
650 static void
651 sfw_destroy_batch(sfw_batch_t *tsb)
652 {
653 sfw_test_instance_t *tsi;
654
655 LASSERT(!sfw_batch_active(tsb));
656 LASSERT(list_empty(&tsb->bat_list));
657
658 while (!list_empty(&tsb->bat_tests)) {
659 tsi = list_entry(tsb->bat_tests.next,
660 sfw_test_instance_t, tsi_list);
661 list_del_init(&tsi->tsi_list);
662 sfw_destroy_test_instance(tsi);
663 }
664
665 LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
666 }
667
668 void
669 sfw_destroy_session(sfw_session_t *sn)
670 {
671 sfw_batch_t *batch;
672
673 LASSERT(list_empty(&sn->sn_list));
674 LASSERT(sn != sfw_data.fw_session);
675
676 while (!list_empty(&sn->sn_batches)) {
677 batch = list_entry(sn->sn_batches.next,
678 sfw_batch_t, bat_list);
679 list_del_init(&batch->bat_list);
680 sfw_destroy_batch(batch);
681 }
682
683 LIBCFS_FREE(sn, sizeof(*sn));
684 atomic_dec(&sfw_data.fw_nzombies);
685 }
686
687 static void
688 sfw_unpack_addtest_req(srpc_msg_t *msg)
689 {
690 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
691
692 LASSERT(msg->msg_type == SRPC_MSG_TEST_REQST);
693 LASSERT(req->tsr_is_client);
694
695 if (msg->msg_magic == SRPC_MSG_MAGIC)
696 return; /* no flipping needed */
697
698 LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
699
700 if (req->tsr_service == SRPC_SERVICE_BRW) {
701 if (!(msg->msg_ses_feats & LST_FEAT_BULK_LEN)) {
702 test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
703
704 __swab32s(&bulk->blk_opc);
705 __swab32s(&bulk->blk_npg);
706 __swab32s(&bulk->blk_flags);
707
708 } else {
709 test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
710
711 __swab16s(&bulk->blk_opc);
712 __swab16s(&bulk->blk_flags);
713 __swab32s(&bulk->blk_offset);
714 __swab32s(&bulk->blk_len);
715 }
716
717 return;
718 }
719
720 if (req->tsr_service == SRPC_SERVICE_PING) {
721 test_ping_req_t *ping = &req->tsr_u.ping;
722
723 __swab32s(&ping->png_size);
724 __swab32s(&ping->png_flags);
725 return;
726 }
727
728 LBUG();
729 }
730
731 static int
732 sfw_add_test_instance(sfw_batch_t *tsb, struct srpc_server_rpc *rpc)
733 {
734 srpc_msg_t *msg = &rpc->srpc_reqstbuf->buf_msg;
735 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
736 srpc_bulk_t *bk = rpc->srpc_bulk;
737 int ndest = req->tsr_ndest;
738 sfw_test_unit_t *tsu;
739 sfw_test_instance_t *tsi;
740 int i;
741 int rc;
742
743 LIBCFS_ALLOC(tsi, sizeof(*tsi));
744 if (!tsi) {
745 CERROR("Can't allocate test instance for batch: %llu\n",
746 tsb->bat_id.bat_id);
747 return -ENOMEM;
748 }
749
750 spin_lock_init(&tsi->tsi_lock);
751 atomic_set(&tsi->tsi_nactive, 0);
752 INIT_LIST_HEAD(&tsi->tsi_units);
753 INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
754 INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
755
756 tsi->tsi_stopping = 0;
757 tsi->tsi_batch = tsb;
758 tsi->tsi_loop = req->tsr_loop;
759 tsi->tsi_concur = req->tsr_concur;
760 tsi->tsi_service = req->tsr_service;
761 tsi->tsi_is_client = !!(req->tsr_is_client);
762 tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
763
764 rc = sfw_load_test(tsi);
765 if (rc) {
766 LIBCFS_FREE(tsi, sizeof(*tsi));
767 return rc;
768 }
769
770 LASSERT(!sfw_batch_active(tsb));
771
772 if (!tsi->tsi_is_client) {
773 /* it's test server, just add it to tsb */
774 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
775 return 0;
776 }
777
778 LASSERT(bk);
779 LASSERT(bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
780 LASSERT((unsigned int)bk->bk_len >=
781 sizeof(lnet_process_id_packed_t) * ndest);
782
783 sfw_unpack_addtest_req(msg);
784 memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
785
786 for (i = 0; i < ndest; i++) {
787 lnet_process_id_packed_t *dests;
788 lnet_process_id_packed_t id;
789 int j;
790
791 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
792 LASSERT(dests); /* my pages are within KVM always */
793 id = dests[i % SFW_ID_PER_PAGE];
794 if (msg->msg_magic != SRPC_MSG_MAGIC)
795 sfw_unpack_id(id);
796
797 for (j = 0; j < tsi->tsi_concur; j++) {
798 LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
799 if (!tsu) {
800 rc = -ENOMEM;
801 CERROR("Can't allocate tsu for %d\n",
802 tsi->tsi_service);
803 goto error;
804 }
805
806 tsu->tsu_dest.nid = id.nid;
807 tsu->tsu_dest.pid = id.pid;
808 tsu->tsu_instance = tsi;
809 tsu->tsu_private = NULL;
810 list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
811 }
812 }
813
814 rc = tsi->tsi_ops->tso_init(tsi);
815 if (!rc) {
816 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
817 return 0;
818 }
819
820 error:
821 LASSERT(rc);
822 sfw_destroy_test_instance(tsi);
823 return rc;
824 }
825
826 static void
827 sfw_test_unit_done(sfw_test_unit_t *tsu)
828 {
829 sfw_test_instance_t *tsi = tsu->tsu_instance;
830 sfw_batch_t *tsb = tsi->tsi_batch;
831 sfw_session_t *sn = tsb->bat_session;
832
833 LASSERT(sfw_test_active(tsi));
834
835 if (!atomic_dec_and_test(&tsi->tsi_nactive))
836 return;
837
838 /* the test instance is done */
839 spin_lock(&tsi->tsi_lock);
840
841 tsi->tsi_stopping = 0;
842
843 spin_unlock(&tsi->tsi_lock);
844
845 spin_lock(&sfw_data.fw_lock);
846
847 if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
848 sn == sfw_data.fw_session) { /* sn also active */
849 spin_unlock(&sfw_data.fw_lock);
850 return;
851 }
852
853 LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
854
855 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
856 if (sfw_batch_active(tsb)) {
857 spin_unlock(&sfw_data.fw_lock);
858 return;
859 }
860 }
861
862 list_del_init(&sn->sn_list);
863 spin_unlock(&sfw_data.fw_lock);
864
865 sfw_destroy_session(sn);
866 }
867
868 static void
869 sfw_test_rpc_done(srpc_client_rpc_t *rpc)
870 {
871 sfw_test_unit_t *tsu = rpc->crpc_priv;
872 sfw_test_instance_t *tsi = tsu->tsu_instance;
873 int done = 0;
874
875 tsi->tsi_ops->tso_done_rpc(tsu, rpc);
876
877 spin_lock(&tsi->tsi_lock);
878
879 LASSERT(sfw_test_active(tsi));
880 LASSERT(!list_empty(&rpc->crpc_list));
881
882 list_del_init(&rpc->crpc_list);
883
884 /* batch is stopping or loop is done or get error */
885 if (tsi->tsi_stopping || !tsu->tsu_loop ||
886 (rpc->crpc_status && tsi->tsi_stoptsu_onerr))
887 done = 1;
888
889 /* dec ref for poster */
890 srpc_client_rpc_decref(rpc);
891
892 spin_unlock(&tsi->tsi_lock);
893
894 if (!done) {
895 swi_schedule_workitem(&tsu->tsu_worker);
896 return;
897 }
898
899 sfw_test_unit_done(tsu);
900 }
901
902 int
903 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
904 unsigned features, int nblk, int blklen,
905 srpc_client_rpc_t **rpcpp)
906 {
907 srpc_client_rpc_t *rpc = NULL;
908 sfw_test_instance_t *tsi = tsu->tsu_instance;
909
910 spin_lock(&tsi->tsi_lock);
911
912 LASSERT(sfw_test_active(tsi));
913 /* pick request from buffer */
914 rpc = list_first_entry_or_null(&tsi->tsi_free_rpcs,
915 srpc_client_rpc_t, crpc_list);
916 if (rpc) {
917 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
918 list_del_init(&rpc->crpc_list);
919 }
920
921 spin_unlock(&tsi->tsi_lock);
922
923 if (!rpc) {
924 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
925 blklen, sfw_test_rpc_done,
926 sfw_test_rpc_fini, tsu);
927 } else {
928 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
929 blklen, sfw_test_rpc_done,
930 sfw_test_rpc_fini, tsu);
931 }
932
933 if (!rpc) {
934 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
935 return -ENOMEM;
936 }
937
938 rpc->crpc_reqstmsg.msg_ses_feats = features;
939 *rpcpp = rpc;
940
941 return 0;
942 }
943
944 static int
945 sfw_run_test(swi_workitem_t *wi)
946 {
947 sfw_test_unit_t *tsu = wi->swi_workitem.wi_data;
948 sfw_test_instance_t *tsi = tsu->tsu_instance;
949 srpc_client_rpc_t *rpc = NULL;
950
951 LASSERT(wi == &tsu->tsu_worker);
952
953 if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc)) {
954 LASSERT(!rpc);
955 goto test_done;
956 }
957
958 LASSERT(rpc);
959
960 spin_lock(&tsi->tsi_lock);
961
962 if (tsi->tsi_stopping) {
963 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
964 spin_unlock(&tsi->tsi_lock);
965 goto test_done;
966 }
967
968 if (tsu->tsu_loop > 0)
969 tsu->tsu_loop--;
970
971 list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
972 spin_unlock(&tsi->tsi_lock);
973
974 spin_lock(&rpc->crpc_lock);
975 rpc->crpc_timeout = rpc_timeout;
976 srpc_post_rpc(rpc);
977 spin_unlock(&rpc->crpc_lock);
978 return 0;
979
980 test_done:
981 /*
982 * No one can schedule me now since:
983 * - previous RPC, if any, has done and
984 * - no new RPC is initiated.
985 * - my batch is still active; no one can run it again now.
986 * Cancel pending schedules and prevent future schedule attempts:
987 */
988 swi_exit_workitem(wi);
989 sfw_test_unit_done(tsu);
990 return 1;
991 }
992
993 static int
994 sfw_run_batch(sfw_batch_t *tsb)
995 {
996 swi_workitem_t *wi;
997 sfw_test_unit_t *tsu;
998 sfw_test_instance_t *tsi;
999
1000 if (sfw_batch_active(tsb)) {
1001 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1002 tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1003 return 0;
1004 }
1005
1006 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1007 if (!tsi->tsi_is_client) /* skip server instances */
1008 continue;
1009
1010 LASSERT(!tsi->tsi_stopping);
1011 LASSERT(!sfw_test_active(tsi));
1012
1013 atomic_inc(&tsb->bat_nactive);
1014
1015 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1016 atomic_inc(&tsi->tsi_nactive);
1017 tsu->tsu_loop = tsi->tsi_loop;
1018 wi = &tsu->tsu_worker;
1019 swi_init_workitem(wi, tsu, sfw_run_test,
1020 lst_sched_test[lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1021 swi_schedule_workitem(wi);
1022 }
1023 }
1024
1025 return 0;
1026 }
1027
1028 int
1029 sfw_stop_batch(sfw_batch_t *tsb, int force)
1030 {
1031 sfw_test_instance_t *tsi;
1032 srpc_client_rpc_t *rpc;
1033
1034 if (!sfw_batch_active(tsb)) {
1035 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1036 return 0;
1037 }
1038
1039 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
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
1070 static int
1071 sfw_query_batch(sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1072 {
1073 sfw_test_instance_t *tsi;
1074
1075 if (testidx < 0)
1076 return -EINVAL;
1077
1078 if (!testidx) {
1079 reply->bar_active = atomic_read(&tsb->bat_nactive);
1080 return 0;
1081 }
1082
1083 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
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
1094 void
1095 sfw_free_pages(struct srpc_server_rpc *rpc)
1096 {
1097 srpc_free_bulk(rpc->srpc_bulk);
1098 rpc->srpc_bulk = NULL;
1099 }
1100
1101 int
1102 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1103 int sink)
1104 {
1105 LASSERT(!rpc->srpc_bulk);
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)
1110 return -ENOMEM;
1111
1112 return 0;
1113 }
1114
1115 static int
1116 sfw_add_test(struct srpc_server_rpc *rpc)
1117 {
1118 sfw_session_t *sn = sfw_data.fw_session;
1119 srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1120 srpc_test_reqst_t *request;
1121 int rc;
1122 sfw_batch_t *bat;
1123
1124 request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1125 reply->tsr_sid = !sn ? LST_INVALID_SID : sn->sn_id;
1126
1127 if (!request->tsr_loop ||
1128 !request->tsr_concur ||
1129 request->tsr_sid.ses_nid == LNET_NID_ANY ||
1130 request->tsr_ndest > SFW_MAX_NDESTS ||
1131 (request->tsr_is_client && !request->tsr_ndest) ||
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 || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1140 !sfw_find_test_case(request->tsr_service)) {
1141 reply->tsr_status = ENOENT;
1142 return 0;
1143 }
1144
1145 bat = sfw_bid2batch(request->tsr_bid);
1146 if (!bat) {
1147 CERROR("dropping RPC %s from %s under memory pressure\n",
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) {
1159 /* rpc will be resumed later in sfw_bulk_ready */
1160 int npg = sfw_id_pages(request->tsr_ndest);
1161 int len;
1162
1163 if (!(sn->sn_features & LST_FEAT_BULK_LEN)) {
1164 len = npg * PAGE_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);
1175 CDEBUG(!rc ? D_NET : D_WARNING,
1176 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1177 !rc ? "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
1185 static int
1186 sfw_control_batch(srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1187 {
1188 sfw_session_t *sn = sfw_data.fw_session;
1189 int rc = 0;
1190 sfw_batch_t *bat;
1191
1192 reply->bar_sid = !sn ? LST_INVALID_SID : sn->sn_id;
1193
1194 if (!sn || !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) {
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
1226 static int
1227 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1228 {
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;
1234
1235 LASSERT(!sfw_data.fw_active_srpc);
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()) {
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 &&
1267 sn->sn_features != request->msg_ses_feats) {
1268 CNETERR("Features of framework RPC don't match features of current session: %x/%x\n",
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) {
1276 /**
1277 * NB: at this point, old version will ignore features and
1278 * create new session anyway, so console should be able
1279 * to handle this
1280 */
1281 reply->msg_body.reply.status = EPROTO;
1282 goto out;
1283 }
1284
1285 switch (sv->sv_id) {
1286 default:
1287 LBUG();
1288 case SRPC_SERVICE_TEST:
1289 rc = sfw_add_test(rpc);
1290 break;
1291
1292 case SRPC_SERVICE_BATCH:
1293 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1294 &reply->msg_body.bat_reply);
1295 break;
1296
1297 case SRPC_SERVICE_QUERY_STAT:
1298 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1299 &reply->msg_body.stat_reply);
1300 break;
1301
1302 case SRPC_SERVICE_DEBUG:
1303 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1304 &reply->msg_body.dbg_reply);
1305 break;
1306
1307 case SRPC_SERVICE_MAKE_SESSION:
1308 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1309 &reply->msg_body.mksn_reply);
1310 break;
1311
1312 case SRPC_SERVICE_REMOVE_SESSION:
1313 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1314 &reply->msg_body.rmsn_reply);
1315 break;
1316 }
1317
1318 if (sfw_data.fw_session)
1319 features = sfw_data.fw_session->sn_features;
1320 out:
1321 reply->msg_ses_feats = features;
1322 rpc->srpc_done = sfw_server_rpc_done;
1323 spin_lock(&sfw_data.fw_lock);
1324
1325 if (!sfw_data.fw_shuttingdown)
1326 sfw_add_session_timer();
1327
1328 sfw_data.fw_active_srpc = NULL;
1329 spin_unlock(&sfw_data.fw_lock);
1330 return rc;
1331 }
1332
1333 static int
1334 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1335 {
1336 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1337 int rc;
1338
1339 LASSERT(rpc->srpc_bulk);
1340 LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1341 LASSERT(!sfw_data.fw_active_srpc);
1342 LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1343
1344 spin_lock(&sfw_data.fw_lock);
1345
1346 if (status) {
1347 CERROR("Bulk transfer failed for RPC: service %s, peer %s, status %d\n",
1348 sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1349 spin_unlock(&sfw_data.fw_lock);
1350 return -EIO;
1351 }
1352
1353 if (sfw_data.fw_shuttingdown) {
1354 spin_unlock(&sfw_data.fw_lock);
1355 return -ESHUTDOWN;
1356 }
1357
1358 if (sfw_del_session_timer()) {
1359 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1360 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1361 spin_unlock(&sfw_data.fw_lock);
1362 return -EAGAIN;
1363 }
1364
1365 sfw_data.fw_active_srpc = rpc;
1366 spin_unlock(&sfw_data.fw_lock);
1367
1368 rc = sfw_add_test(rpc);
1369
1370 spin_lock(&sfw_data.fw_lock);
1371
1372 if (!sfw_data.fw_shuttingdown)
1373 sfw_add_session_timer();
1374
1375 sfw_data.fw_active_srpc = NULL;
1376 spin_unlock(&sfw_data.fw_lock);
1377 return rc;
1378 }
1379
1380 srpc_client_rpc_t *
1381 sfw_create_rpc(lnet_process_id_t peer, int service,
1382 unsigned features, int nbulkiov, int bulklen,
1383 void (*done)(srpc_client_rpc_t *), void *priv)
1384 {
1385 srpc_client_rpc_t *rpc = NULL;
1386
1387 spin_lock(&sfw_data.fw_lock);
1388
1389 LASSERT(!sfw_data.fw_shuttingdown);
1390 LASSERT(service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1391
1392 if (!nbulkiov && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1393 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1394 srpc_client_rpc_t, crpc_list);
1395 list_del(&rpc->crpc_list);
1396
1397 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1398 done, sfw_client_rpc_fini, priv);
1399 }
1400
1401 spin_unlock(&sfw_data.fw_lock);
1402
1403 if (!rpc) {
1404 rpc = srpc_create_client_rpc(peer, service,
1405 nbulkiov, bulklen, done,
1406 nbulkiov ? NULL :
1407 sfw_client_rpc_fini,
1408 priv);
1409 }
1410
1411 if (rpc) /* "session" is concept in framework */
1412 rpc->crpc_reqstmsg.msg_ses_feats = features;
1413
1414 return rpc;
1415 }
1416
1417 void
1418 sfw_unpack_message(srpc_msg_t *msg)
1419 {
1420 if (msg->msg_magic == SRPC_MSG_MAGIC)
1421 return; /* no flipping needed */
1422
1423 /* srpc module should guarantee I wouldn't get crap */
1424 LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1425
1426 if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1427 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1428
1429 __swab32s(&req->str_type);
1430 __swab64s(&req->str_rpyid);
1431 sfw_unpack_sid(req->str_sid);
1432 return;
1433 }
1434
1435 if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1436 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1437
1438 __swab32s(&rep->str_status);
1439 sfw_unpack_sid(rep->str_sid);
1440 sfw_unpack_fw_counters(rep->str_fw);
1441 sfw_unpack_rpc_counters(rep->str_rpc);
1442 sfw_unpack_lnet_counters(rep->str_lnet);
1443 return;
1444 }
1445
1446 if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1447 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1448
1449 __swab64s(&req->mksn_rpyid);
1450 __swab32s(&req->mksn_force);
1451 sfw_unpack_sid(req->mksn_sid);
1452 return;
1453 }
1454
1455 if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1456 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1457
1458 __swab32s(&rep->mksn_status);
1459 __swab32s(&rep->mksn_timeout);
1460 sfw_unpack_sid(rep->mksn_sid);
1461 return;
1462 }
1463
1464 if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1465 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1466
1467 __swab64s(&req->rmsn_rpyid);
1468 sfw_unpack_sid(req->rmsn_sid);
1469 return;
1470 }
1471
1472 if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1473 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1474
1475 __swab32s(&rep->rmsn_status);
1476 sfw_unpack_sid(rep->rmsn_sid);
1477 return;
1478 }
1479
1480 if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1481 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1482
1483 __swab64s(&req->dbg_rpyid);
1484 __swab32s(&req->dbg_flags);
1485 sfw_unpack_sid(req->dbg_sid);
1486 return;
1487 }
1488
1489 if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1490 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1491
1492 __swab32s(&rep->dbg_nbatch);
1493 __swab32s(&rep->dbg_timeout);
1494 sfw_unpack_sid(rep->dbg_sid);
1495 return;
1496 }
1497
1498 if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1499 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1500
1501 __swab32s(&req->bar_opc);
1502 __swab64s(&req->bar_rpyid);
1503 __swab32s(&req->bar_testidx);
1504 __swab32s(&req->bar_arg);
1505 sfw_unpack_sid(req->bar_sid);
1506 __swab64s(&req->bar_bid.bat_id);
1507 return;
1508 }
1509
1510 if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1511 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1512
1513 __swab32s(&rep->bar_status);
1514 sfw_unpack_sid(rep->bar_sid);
1515 return;
1516 }
1517
1518 if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1519 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1520
1521 __swab64s(&req->tsr_rpyid);
1522 __swab64s(&req->tsr_bulkid);
1523 __swab32s(&req->tsr_loop);
1524 __swab32s(&req->tsr_ndest);
1525 __swab32s(&req->tsr_concur);
1526 __swab32s(&req->tsr_service);
1527 sfw_unpack_sid(req->tsr_sid);
1528 __swab64s(&req->tsr_bid.bat_id);
1529 return;
1530 }
1531
1532 if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1533 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1534
1535 __swab32s(&rep->tsr_status);
1536 sfw_unpack_sid(rep->tsr_sid);
1537 return;
1538 }
1539
1540 if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1541 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1542
1543 __swab64s(&req->join_rpyid);
1544 sfw_unpack_sid(req->join_sid);
1545 return;
1546 }
1547
1548 if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1549 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1550
1551 __swab32s(&rep->join_status);
1552 __swab32s(&rep->join_timeout);
1553 sfw_unpack_sid(rep->join_sid);
1554 return;
1555 }
1556
1557 LBUG();
1558 }
1559
1560 void
1561 sfw_abort_rpc(srpc_client_rpc_t *rpc)
1562 {
1563 LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1564 LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1565
1566 spin_lock(&rpc->crpc_lock);
1567 srpc_abort_rpc(rpc, -EINTR);
1568 spin_unlock(&rpc->crpc_lock);
1569 }
1570
1571 void
1572 sfw_post_rpc(srpc_client_rpc_t *rpc)
1573 {
1574 spin_lock(&rpc->crpc_lock);
1575
1576 LASSERT(!rpc->crpc_closed);
1577 LASSERT(!rpc->crpc_aborted);
1578 LASSERT(list_empty(&rpc->crpc_list));
1579 LASSERT(!sfw_data.fw_shuttingdown);
1580
1581 rpc->crpc_timeout = rpc_timeout;
1582 srpc_post_rpc(rpc);
1583
1584 spin_unlock(&rpc->crpc_lock);
1585 }
1586
1587 static srpc_service_t sfw_services[] = {
1588 {
1589 /* sv_id */ SRPC_SERVICE_DEBUG,
1590 /* sv_name */ "debug",
1591 0
1592 },
1593 {
1594 /* sv_id */ SRPC_SERVICE_QUERY_STAT,
1595 /* sv_name */ "query stats",
1596 0
1597 },
1598 {
1599 /* sv_id */ SRPC_SERVICE_MAKE_SESSION,
1600 /* sv_name */ "make session",
1601 0
1602 },
1603 {
1604 /* sv_id */ SRPC_SERVICE_REMOVE_SESSION,
1605 /* sv_name */ "remove session",
1606 0
1607 },
1608 {
1609 /* sv_id */ SRPC_SERVICE_BATCH,
1610 /* sv_name */ "batch service",
1611 0
1612 },
1613 {
1614 /* sv_id */ SRPC_SERVICE_TEST,
1615 /* sv_name */ "test service",
1616 0
1617 },
1618 {
1619 /* sv_id */ 0,
1620 /* sv_name */ NULL,
1621 0
1622 }
1623 };
1624
1625 int
1626 sfw_startup(void)
1627 {
1628 int i;
1629 int rc;
1630 int error;
1631 srpc_service_t *sv;
1632 sfw_test_case_t *tsc;
1633
1634 if (session_timeout < 0) {
1635 CERROR("Session timeout must be non-negative: %d\n",
1636 session_timeout);
1637 return -EINVAL;
1638 }
1639
1640 if (rpc_timeout < 0) {
1641 CERROR("RPC timeout must be non-negative: %d\n",
1642 rpc_timeout);
1643 return -EINVAL;
1644 }
1645
1646 if (!session_timeout)
1647 CWARN("Zero session_timeout specified - test sessions never expire.\n");
1648
1649 if (!rpc_timeout)
1650 CWARN("Zero rpc_timeout specified - test RPC never expire.\n");
1651
1652 memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1653
1654 sfw_data.fw_session = NULL;
1655 sfw_data.fw_active_srpc = NULL;
1656 spin_lock_init(&sfw_data.fw_lock);
1657 atomic_set(&sfw_data.fw_nzombies, 0);
1658 INIT_LIST_HEAD(&sfw_data.fw_tests);
1659 INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1660 INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1661
1662 brw_init_test_client();
1663 brw_init_test_service();
1664 rc = sfw_register_test(&brw_test_service, &brw_test_client);
1665 LASSERT(!rc);
1666
1667 ping_init_test_client();
1668 ping_init_test_service();
1669 rc = sfw_register_test(&ping_test_service, &ping_test_client);
1670 LASSERT(!rc);
1671
1672 error = 0;
1673 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1674 sv = tsc->tsc_srv_service;
1675
1676 rc = srpc_add_service(sv);
1677 LASSERT(rc != -EBUSY);
1678 if (rc) {
1679 CWARN("Failed to add %s service: %d\n",
1680 sv->sv_name, rc);
1681 error = rc;
1682 }
1683 }
1684
1685 for (i = 0; ; i++) {
1686 sv = &sfw_services[i];
1687 if (!sv->sv_name)
1688 break;
1689
1690 sv->sv_bulk_ready = NULL;
1691 sv->sv_handler = sfw_handle_server_rpc;
1692 sv->sv_wi_total = SFW_FRWK_WI_MAX;
1693 if (sv->sv_id == SRPC_SERVICE_TEST)
1694 sv->sv_bulk_ready = sfw_bulk_ready;
1695
1696 rc = srpc_add_service(sv);
1697 LASSERT(rc != -EBUSY);
1698 if (rc) {
1699 CWARN("Failed to add %s service: %d\n",
1700 sv->sv_name, rc);
1701 error = rc;
1702 }
1703
1704 /* about to sfw_shutdown, no need to add buffer */
1705 if (error)
1706 continue;
1707
1708 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1709 if (rc) {
1710 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
1711 sv->sv_name, sv->sv_wi_total, rc);
1712 error = -ENOMEM;
1713 }
1714 }
1715
1716 if (error)
1717 sfw_shutdown();
1718 return error;
1719 }
1720
1721 void
1722 sfw_shutdown(void)
1723 {
1724 srpc_service_t *sv;
1725 sfw_test_case_t *tsc;
1726 int i;
1727
1728 spin_lock(&sfw_data.fw_lock);
1729
1730 sfw_data.fw_shuttingdown = 1;
1731 lst_wait_until(!sfw_data.fw_active_srpc, sfw_data.fw_lock,
1732 "waiting for active RPC to finish.\n");
1733
1734 if (sfw_del_session_timer())
1735 lst_wait_until(!sfw_data.fw_session, sfw_data.fw_lock,
1736 "waiting for session timer to explode.\n");
1737
1738 sfw_deactivate_session();
1739 lst_wait_until(!atomic_read(&sfw_data.fw_nzombies),
1740 sfw_data.fw_lock,
1741 "waiting for %d zombie sessions to die.\n",
1742 atomic_read(&sfw_data.fw_nzombies));
1743
1744 spin_unlock(&sfw_data.fw_lock);
1745
1746 for (i = 0; ; i++) {
1747 sv = &sfw_services[i];
1748 if (!sv->sv_name)
1749 break;
1750
1751 srpc_shutdown_service(sv);
1752 srpc_remove_service(sv);
1753 }
1754
1755 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1756 sv = tsc->tsc_srv_service;
1757 srpc_shutdown_service(sv);
1758 srpc_remove_service(sv);
1759 }
1760
1761 while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1762 srpc_client_rpc_t *rpc;
1763
1764 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1765 srpc_client_rpc_t, crpc_list);
1766 list_del(&rpc->crpc_list);
1767
1768 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1769 }
1770
1771 for (i = 0; ; i++) {
1772 sv = &sfw_services[i];
1773 if (!sv->sv_name)
1774 break;
1775
1776 srpc_wait_service_shutdown(sv);
1777 }
1778
1779 while (!list_empty(&sfw_data.fw_tests)) {
1780 tsc = list_entry(sfw_data.fw_tests.next,
1781 sfw_test_case_t, tsc_list);
1782
1783 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1784
1785 list_del(&tsc->tsc_list);
1786 LIBCFS_FREE(tsc, sizeof(*tsc));
1787 }
1788 }
Linux kernel - Deliverables
This page took 0.067451 seconds and 5 git commands to generate.