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) != 0)
104 #define sfw_batch_active(b) (atomic_read(&(b)->bat_nactive) != 0)
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 srpc_server_rpc_t *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 *
sfw_find_test_case(int id)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
sfw_register_test(srpc_service_t * service,sfw_test_client_ops_t * cliops)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) != NULL) {
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 == NULL)
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
sfw_add_session_timer(void)160 sfw_add_session_timer(void)
161 {
162 sfw_session_t *sn = sfw_data.fw_session;
163 stt_timer_t *timer = &sn->sn_timer;
164
165 LASSERT(!sfw_data.fw_shuttingdown);
166
167 if (sn == NULL || sn->sn_timeout == 0)
168 return;
169
170 LASSERT(!sn->sn_timer_active);
171
172 sn->sn_timer_active = 1;
173 timer->stt_expires = cfs_time_add(sn->sn_timeout,
174 get_seconds());
175 stt_add_timer(timer);
176 return;
177 }
178
179 static int
sfw_del_session_timer(void)180 sfw_del_session_timer(void)
181 {
182 sfw_session_t *sn = sfw_data.fw_session;
183
184 if (sn == NULL || !sn->sn_timer_active)
185 return 0;
186
187 LASSERT(sn->sn_timeout != 0);
188
189 if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
190 sn->sn_timer_active = 0;
191 return 0;
192 }
193
194 return EBUSY; /* racing with sfw_session_expired() */
195 }
196
197 static void
sfw_deactivate_session(void)198 sfw_deactivate_session(void)
199 __must_hold(&sfw_data.fw_lock)
200 {
201 sfw_session_t *sn = sfw_data.fw_session;
202 int nactive = 0;
203 sfw_batch_t *tsb;
204 sfw_test_case_t *tsc;
205
206 if (sn == NULL) return;
207
208 LASSERT(!sn->sn_timer_active);
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
222 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
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
241 static void
sfw_session_expired(void * data)242 sfw_session_expired(void *data)
243 {
244 sfw_session_t *sn = data;
245
246 spin_lock(&sfw_data.fw_lock);
247
248 LASSERT(sn->sn_timer_active);
249 LASSERT(sn == sfw_data.fw_session);
250
251 CWARN("Session expired! sid: %s-%llu, name: %s\n",
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
261 static inline void
sfw_init_session(sfw_session_t * sn,lst_sid_t sid,unsigned features,const char * name)262 sfw_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;
276 sn->sn_id = sid;
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 */
287 static void
sfw_server_rpc_done(struct srpc_server_rpc * rpc)288 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
289 {
290 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
291 int status = rpc->srpc_status;
292
293 CDEBUG(D_NET,
294 "Incoming framework RPC done: service %s, peer %s, status %s:%d\n",
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
304 static void
sfw_client_rpc_fini(srpc_client_rpc_t * rpc)305 sfw_client_rpc_fini(srpc_client_rpc_t *rpc)
306 {
307 LASSERT(rpc->crpc_bulk.bk_niov == 0);
308 LASSERT(list_empty(&rpc->crpc_list));
309 LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
310
311 CDEBUG(D_NET,
312 "Outgoing framework RPC done: service %d, peer %s, status %s:%d:%d\n",
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
326 static sfw_batch_t *
sfw_find_batch(lst_bid_t bid)327 sfw_find_batch(lst_bid_t bid)
328 {
329 sfw_session_t *sn = sfw_data.fw_session;
330 sfw_batch_t *bat;
331
332 LASSERT(sn != NULL);
333
334 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
335 if (bat->bat_id.bat_id == bid.bat_id)
336 return bat;
337 }
338
339 return NULL;
340 }
341
342 static sfw_batch_t *
sfw_bid2batch(lst_bid_t bid)343 sfw_bid2batch(lst_bid_t bid)
344 {
345 sfw_session_t *sn = sfw_data.fw_session;
346 sfw_batch_t *bat;
347
348 LASSERT(sn != NULL);
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
368 static int
sfw_get_stats(srpc_stat_reqst_t * request,srpc_stat_reply_t * reply)369 sfw_get_stats(srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
370 {
371 sfw_session_t *sn = sfw_data.fw_session;
372 sfw_counters_t *cnt = &reply->str_fw;
373 sfw_batch_t *bat;
374 struct timeval tv;
375
376 reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
377
378 if (request->str_sid.ses_nid == LNET_NID_ANY) {
379 reply->str_status = EINVAL;
380 return 0;
381 }
382
383 if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
384 reply->str_status = ESRCH;
385 return 0;
386 }
387
388 lnet_counters_get(&reply->str_lnet);
389 srpc_get_counters(&reply->str_rpc);
390
391 /* send over the msecs since the session was started
392 - with 32 bits to send, this is ~49 days */
393 cfs_duration_usec(cfs_time_sub(cfs_time_current(),
394 sn->sn_started), &tv);
395
396 cnt->running_ms = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
397 cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
398 cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
399 cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
400
401 cnt->active_batches = 0;
402 list_for_each_entry(bat, &sn->sn_batches, bat_list) {
403 if (atomic_read(&bat->bat_nactive) > 0)
404 cnt->active_batches++;
405 }
406
407 reply->str_status = 0;
408 return 0;
409 }
410
411 int
sfw_make_session(srpc_mksn_reqst_t * request,srpc_mksn_reply_t * reply)412 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
413 {
414 sfw_session_t *sn = sfw_data.fw_session;
415 srpc_msg_t *msg = container_of(request, srpc_msg_t,
416 msg_body.mksn_reqst);
417 int cplen = 0;
418
419 if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
420 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
421 reply->mksn_status = EINVAL;
422 return 0;
423 }
424
425 if (sn != NULL) {
426 reply->mksn_status = 0;
427 reply->mksn_sid = sn->sn_id;
428 reply->mksn_timeout = sn->sn_timeout;
429
430 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
431 atomic_inc(&sn->sn_refcount);
432 return 0;
433 }
434
435 if (!request->mksn_force) {
436 reply->mksn_status = EBUSY;
437 cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
438 sizeof(reply->mksn_name));
439 if (cplen >= sizeof(reply->mksn_name))
440 return -E2BIG;
441 return 0;
442 }
443 }
444
445 /* reject the request if it requires unknown features
446 * NB: old version will always accept all features because it's not
447 * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
448 * harmless because it will return zero feature to console, and it's
449 * console's responsibility to make sure all nodes in a session have
450 * same feature mask. */
451 if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
452 reply->mksn_status = EPROTO;
453 return 0;
454 }
455
456 /* brand new or create by force */
457 LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
458 if (sn == NULL) {
459 CERROR("Dropping RPC (mksn) under memory pressure.\n");
460 return -ENOMEM;
461 }
462
463 sfw_init_session(sn, request->mksn_sid,
464 msg->msg_ses_feats, &request->mksn_name[0]);
465
466 spin_lock(&sfw_data.fw_lock);
467
468 sfw_deactivate_session();
469 LASSERT(sfw_data.fw_session == NULL);
470 sfw_data.fw_session = sn;
471
472 spin_unlock(&sfw_data.fw_lock);
473
474 reply->mksn_status = 0;
475 reply->mksn_sid = sn->sn_id;
476 reply->mksn_timeout = sn->sn_timeout;
477 return 0;
478 }
479
480 static int
sfw_remove_session(srpc_rmsn_reqst_t * request,srpc_rmsn_reply_t * reply)481 sfw_remove_session(srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
482 {
483 sfw_session_t *sn = sfw_data.fw_session;
484
485 reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
486
487 if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
488 reply->rmsn_status = EINVAL;
489 return 0;
490 }
491
492 if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
493 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
494 return 0;
495 }
496
497 if (!atomic_dec_and_test(&sn->sn_refcount)) {
498 reply->rmsn_status = 0;
499 return 0;
500 }
501
502 spin_lock(&sfw_data.fw_lock);
503 sfw_deactivate_session();
504 spin_unlock(&sfw_data.fw_lock);
505
506 reply->rmsn_status = 0;
507 reply->rmsn_sid = LST_INVALID_SID;
508 LASSERT(sfw_data.fw_session == NULL);
509 return 0;
510 }
511
512 static int
sfw_debug_session(srpc_debug_reqst_t * request,srpc_debug_reply_t * reply)513 sfw_debug_session(srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
514 {
515 sfw_session_t *sn = sfw_data.fw_session;
516
517 if (sn == NULL) {
518 reply->dbg_status = ESRCH;
519 reply->dbg_sid = LST_INVALID_SID;
520 return 0;
521 }
522
523 reply->dbg_status = 0;
524 reply->dbg_sid = sn->sn_id;
525 reply->dbg_timeout = sn->sn_timeout;
526 if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
527 >= sizeof(reply->dbg_name))
528 return -E2BIG;
529
530 return 0;
531 }
532
533 static void
sfw_test_rpc_fini(srpc_client_rpc_t * rpc)534 sfw_test_rpc_fini(srpc_client_rpc_t *rpc)
535 {
536 sfw_test_unit_t *tsu = rpc->crpc_priv;
537 sfw_test_instance_t *tsi = tsu->tsu_instance;
538
539 /* Called with hold of tsi->tsi_lock */
540 LASSERT(list_empty(&rpc->crpc_list));
541 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
542 }
543
544 static inline int
sfw_test_buffers(sfw_test_instance_t * tsi)545 sfw_test_buffers(sfw_test_instance_t *tsi)
546 {
547 struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
548 struct srpc_service *svc = tsc->tsc_srv_service;
549 int nbuf;
550
551 nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
552 return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
553 }
554
555 static int
sfw_load_test(struct sfw_test_instance * tsi)556 sfw_load_test(struct sfw_test_instance *tsi)
557 {
558 struct sfw_test_case *tsc;
559 struct srpc_service *svc;
560 int nbuf;
561 int rc;
562
563 LASSERT(tsi != NULL);
564 tsc = sfw_find_test_case(tsi->tsi_service);
565 nbuf = sfw_test_buffers(tsi);
566 LASSERT(tsc != NULL);
567 svc = tsc->tsc_srv_service;
568
569 if (tsi->tsi_is_client) {
570 tsi->tsi_ops = tsc->tsc_cli_ops;
571 return 0;
572 }
573
574 rc = srpc_service_add_buffers(svc, nbuf);
575 if (rc != 0) {
576 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
577 svc->sv_name, nbuf, rc);
578 /* NB: this error handler is not strictly correct, because
579 * it may release more buffers than already allocated,
580 * but it doesn't matter because request portal should
581 * be lazy portal and will grow buffers if necessary. */
582 srpc_service_remove_buffers(svc, nbuf);
583 return -ENOMEM;
584 }
585
586 CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
587 nbuf * (srpc_serv_is_framework(svc) ?
588 1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
589 return 0;
590 }
591
592 static void
sfw_unload_test(struct sfw_test_instance * tsi)593 sfw_unload_test(struct sfw_test_instance *tsi)
594 {
595 struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
596
597 LASSERT(tsc != NULL);
598
599 if (tsi->tsi_is_client)
600 return;
601
602 /* shrink buffers, because request portal is lazy portal
603 * which can grow buffers at runtime so we may leave
604 * some buffers behind, but never mind... */
605 srpc_service_remove_buffers(tsc->tsc_srv_service,
606 sfw_test_buffers(tsi));
607 return;
608 }
609
610 static void
sfw_destroy_test_instance(sfw_test_instance_t * tsi)611 sfw_destroy_test_instance(sfw_test_instance_t *tsi)
612 {
613 srpc_client_rpc_t *rpc;
614 sfw_test_unit_t *tsu;
615
616 if (!tsi->tsi_is_client) goto clean;
617
618 tsi->tsi_ops->tso_fini(tsi);
619
620 LASSERT(!tsi->tsi_stopping);
621 LASSERT(list_empty(&tsi->tsi_active_rpcs));
622 LASSERT(!sfw_test_active(tsi));
623
624 while (!list_empty(&tsi->tsi_units)) {
625 tsu = list_entry(tsi->tsi_units.next,
626 sfw_test_unit_t, tsu_list);
627 list_del(&tsu->tsu_list);
628 LIBCFS_FREE(tsu, sizeof(*tsu));
629 }
630
631 while (!list_empty(&tsi->tsi_free_rpcs)) {
632 rpc = list_entry(tsi->tsi_free_rpcs.next,
633 srpc_client_rpc_t, crpc_list);
634 list_del(&rpc->crpc_list);
635 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
636 }
637
638 clean:
639 sfw_unload_test(tsi);
640 LIBCFS_FREE(tsi, sizeof(*tsi));
641 return;
642 }
643
644 static void
sfw_destroy_batch(sfw_batch_t * tsb)645 sfw_destroy_batch(sfw_batch_t *tsb)
646 {
647 sfw_test_instance_t *tsi;
648
649 LASSERT(!sfw_batch_active(tsb));
650 LASSERT(list_empty(&tsb->bat_list));
651
652 while (!list_empty(&tsb->bat_tests)) {
653 tsi = list_entry(tsb->bat_tests.next,
654 sfw_test_instance_t, tsi_list);
655 list_del_init(&tsi->tsi_list);
656 sfw_destroy_test_instance(tsi);
657 }
658
659 LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
660 return;
661 }
662
663 void
sfw_destroy_session(sfw_session_t * sn)664 sfw_destroy_session(sfw_session_t *sn)
665 {
666 sfw_batch_t *batch;
667
668 LASSERT(list_empty(&sn->sn_list));
669 LASSERT(sn != sfw_data.fw_session);
670
671 while (!list_empty(&sn->sn_batches)) {
672 batch = list_entry(sn->sn_batches.next,
673 sfw_batch_t, bat_list);
674 list_del_init(&batch->bat_list);
675 sfw_destroy_batch(batch);
676 }
677
678 LIBCFS_FREE(sn, sizeof(*sn));
679 atomic_dec(&sfw_data.fw_nzombies);
680 return;
681 }
682
683 static void
sfw_unpack_addtest_req(srpc_msg_t * msg)684 sfw_unpack_addtest_req(srpc_msg_t *msg)
685 {
686 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
687
688 LASSERT(msg->msg_type == SRPC_MSG_TEST_REQST);
689 LASSERT(req->tsr_is_client);
690
691 if (msg->msg_magic == SRPC_MSG_MAGIC)
692 return; /* no flipping needed */
693
694 LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
695
696 if (req->tsr_service == SRPC_SERVICE_BRW) {
697 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
698 test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
699
700 __swab32s(&bulk->blk_opc);
701 __swab32s(&bulk->blk_npg);
702 __swab32s(&bulk->blk_flags);
703
704 } else {
705 test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
706
707 __swab16s(&bulk->blk_opc);
708 __swab16s(&bulk->blk_flags);
709 __swab32s(&bulk->blk_offset);
710 __swab32s(&bulk->blk_len);
711 }
712
713 return;
714 }
715
716 if (req->tsr_service == SRPC_SERVICE_PING) {
717 test_ping_req_t *ping = &req->tsr_u.ping;
718
719 __swab32s(&ping->png_size);
720 __swab32s(&ping->png_flags);
721 return;
722 }
723
724 LBUG();
725 return;
726 }
727
728 static int
sfw_add_test_instance(sfw_batch_t * tsb,srpc_server_rpc_t * rpc)729 sfw_add_test_instance(sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
730 {
731 srpc_msg_t *msg = &rpc->srpc_reqstbuf->buf_msg;
732 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
733 srpc_bulk_t *bk = rpc->srpc_bulk;
734 int ndest = req->tsr_ndest;
735 sfw_test_unit_t *tsu;
736 sfw_test_instance_t *tsi;
737 int i;
738 int rc;
739
740 LIBCFS_ALLOC(tsi, sizeof(*tsi));
741 if (tsi == NULL) {
742 CERROR("Can't allocate test instance for batch: %llu\n",
743 tsb->bat_id.bat_id);
744 return -ENOMEM;
745 }
746
747 spin_lock_init(&tsi->tsi_lock);
748 atomic_set(&tsi->tsi_nactive, 0);
749 INIT_LIST_HEAD(&tsi->tsi_units);
750 INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
751 INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
752
753 tsi->tsi_stopping = 0;
754 tsi->tsi_batch = tsb;
755 tsi->tsi_loop = req->tsr_loop;
756 tsi->tsi_concur = req->tsr_concur;
757 tsi->tsi_service = req->tsr_service;
758 tsi->tsi_is_client = !!(req->tsr_is_client);
759 tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
760
761 rc = sfw_load_test(tsi);
762 if (rc != 0) {
763 LIBCFS_FREE(tsi, sizeof(*tsi));
764 return rc;
765 }
766
767 LASSERT(!sfw_batch_active(tsb));
768
769 if (!tsi->tsi_is_client) {
770 /* it's test server, just add it to tsb */
771 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
772 return 0;
773 }
774
775 LASSERT(bk != NULL);
776 LASSERT(bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
777 LASSERT((unsigned int)bk->bk_len >=
778 sizeof(lnet_process_id_packed_t) * ndest);
779
780 sfw_unpack_addtest_req(msg);
781 memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
782
783 for (i = 0; i < ndest; i++) {
784 lnet_process_id_packed_t *dests;
785 lnet_process_id_packed_t id;
786 int j;
787
788 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
789 LASSERT(dests != NULL); /* my pages are within KVM always */
790 id = dests[i % SFW_ID_PER_PAGE];
791 if (msg->msg_magic != SRPC_MSG_MAGIC)
792 sfw_unpack_id(id);
793
794 for (j = 0; j < tsi->tsi_concur; j++) {
795 LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
796 if (tsu == NULL) {
797 rc = -ENOMEM;
798 CERROR("Can't allocate tsu for %d\n",
799 tsi->tsi_service);
800 goto error;
801 }
802
803 tsu->tsu_dest.nid = id.nid;
804 tsu->tsu_dest.pid = id.pid;
805 tsu->tsu_instance = tsi;
806 tsu->tsu_private = NULL;
807 list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
808 }
809 }
810
811 rc = tsi->tsi_ops->tso_init(tsi);
812 if (rc == 0) {
813 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
814 return 0;
815 }
816
817 error:
818 LASSERT(rc != 0);
819 sfw_destroy_test_instance(tsi);
820 return rc;
821 }
822
823 static void
sfw_test_unit_done(sfw_test_unit_t * tsu)824 sfw_test_unit_done(sfw_test_unit_t *tsu)
825 {
826 sfw_test_instance_t *tsi = tsu->tsu_instance;
827 sfw_batch_t *tsb = tsi->tsi_batch;
828 sfw_session_t *sn = tsb->bat_session;
829
830 LASSERT(sfw_test_active(tsi));
831
832 if (!atomic_dec_and_test(&tsi->tsi_nactive))
833 return;
834
835 /* the test instance is done */
836 spin_lock(&tsi->tsi_lock);
837
838 tsi->tsi_stopping = 0;
839
840 spin_unlock(&tsi->tsi_lock);
841
842 spin_lock(&sfw_data.fw_lock);
843
844 if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
845 sn == sfw_data.fw_session) { /* sn also active */
846 spin_unlock(&sfw_data.fw_lock);
847 return;
848 }
849
850 LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
851
852 list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
853 if (sfw_batch_active(tsb)) {
854 spin_unlock(&sfw_data.fw_lock);
855 return;
856 }
857 }
858
859 list_del_init(&sn->sn_list);
860 spin_unlock(&sfw_data.fw_lock);
861
862 sfw_destroy_session(sn);
863 return;
864 }
865
866 static void
sfw_test_rpc_done(srpc_client_rpc_t * rpc)867 sfw_test_rpc_done(srpc_client_rpc_t *rpc)
868 {
869 sfw_test_unit_t *tsu = rpc->crpc_priv;
870 sfw_test_instance_t *tsi = tsu->tsu_instance;
871 int done = 0;
872
873 tsi->tsi_ops->tso_done_rpc(tsu, rpc);
874
875 spin_lock(&tsi->tsi_lock);
876
877 LASSERT(sfw_test_active(tsi));
878 LASSERT(!list_empty(&rpc->crpc_list));
879
880 list_del_init(&rpc->crpc_list);
881
882 /* batch is stopping or loop is done or get error */
883 if (tsi->tsi_stopping ||
884 tsu->tsu_loop == 0 ||
885 (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
886 done = 1;
887
888 /* dec ref for poster */
889 srpc_client_rpc_decref(rpc);
890
891 spin_unlock(&tsi->tsi_lock);
892
893 if (!done) {
894 swi_schedule_workitem(&tsu->tsu_worker);
895 return;
896 }
897
898 sfw_test_unit_done(tsu);
899 return;
900 }
901
902 int
sfw_create_test_rpc(sfw_test_unit_t * tsu,lnet_process_id_t peer,unsigned features,int nblk,int blklen,srpc_client_rpc_t ** rpcpp)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
914 if (!list_empty(&tsi->tsi_free_rpcs)) {
915 /* pick request from buffer */
916 rpc = list_entry(tsi->tsi_free_rpcs.next,
917 srpc_client_rpc_t, crpc_list);
918 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
919 list_del_init(&rpc->crpc_list);
920 }
921
922 spin_unlock(&tsi->tsi_lock);
923
924 if (rpc == NULL) {
925 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
926 blklen, sfw_test_rpc_done,
927 sfw_test_rpc_fini, tsu);
928 } else {
929 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
930 blklen, sfw_test_rpc_done,
931 sfw_test_rpc_fini, tsu);
932 }
933
934 if (rpc == NULL) {
935 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
936 return -ENOMEM;
937 }
938
939 rpc->crpc_reqstmsg.msg_ses_feats = features;
940 *rpcpp = rpc;
941
942 return 0;
943 }
944
945 static int
sfw_run_test(swi_workitem_t * wi)946 sfw_run_test(swi_workitem_t *wi)
947 {
948 sfw_test_unit_t *tsu = wi->swi_workitem.wi_data;
949 sfw_test_instance_t *tsi = tsu->tsu_instance;
950 srpc_client_rpc_t *rpc = NULL;
951
952 LASSERT(wi == &tsu->tsu_worker);
953
954 if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
955 LASSERT(rpc == NULL);
956 goto test_done;
957 }
958
959 LASSERT(rpc != NULL);
960
961 spin_lock(&tsi->tsi_lock);
962
963 if (tsi->tsi_stopping) {
964 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
965 spin_unlock(&tsi->tsi_lock);
966 goto test_done;
967 }
968
969 if (tsu->tsu_loop > 0)
970 tsu->tsu_loop--;
971
972 list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
973 spin_unlock(&tsi->tsi_lock);
974
975 rpc->crpc_timeout = rpc_timeout;
976
977 spin_lock(&rpc->crpc_lock);
978 srpc_post_rpc(rpc);
979 spin_unlock(&rpc->crpc_lock);
980 return 0;
981
982 test_done:
983 /*
984 * No one can schedule me now since:
985 * - previous RPC, if any, has done and
986 * - no new RPC is initiated.
987 * - my batch is still active; no one can run it again now.
988 * Cancel pending schedules and prevent future schedule attempts:
989 */
990 swi_exit_workitem(wi);
991 sfw_test_unit_done(tsu);
992 return 1;
993 }
994
995 static int
sfw_run_batch(sfw_batch_t * tsb)996 sfw_run_batch(sfw_batch_t *tsb)
997 {
998 swi_workitem_t *wi;
999 sfw_test_unit_t *tsu;
1000 sfw_test_instance_t *tsi;
1001
1002 if (sfw_batch_active(tsb)) {
1003 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1004 tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1005 return 0;
1006 }
1007
1008 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1009 if (!tsi->tsi_is_client) /* skip server instances */
1010 continue;
1011
1012 LASSERT(!tsi->tsi_stopping);
1013 LASSERT(!sfw_test_active(tsi));
1014
1015 atomic_inc(&tsb->bat_nactive);
1016
1017 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1018 atomic_inc(&tsi->tsi_nactive);
1019 tsu->tsu_loop = tsi->tsi_loop;
1020 wi = &tsu->tsu_worker;
1021 swi_init_workitem(wi, tsu, sfw_run_test,
1022 lst_sched_test[\
1023 lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1024 swi_schedule_workitem(wi);
1025 }
1026 }
1027
1028 return 0;
1029 }
1030
1031 int
sfw_stop_batch(sfw_batch_t * tsb,int force)1032 sfw_stop_batch(sfw_batch_t *tsb, int force)
1033 {
1034 sfw_test_instance_t *tsi;
1035 srpc_client_rpc_t *rpc;
1036
1037 if (!sfw_batch_active(tsb)) {
1038 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1039 return 0;
1040 }
1041
1042 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1043 spin_lock(&tsi->tsi_lock);
1044
1045 if (!tsi->tsi_is_client ||
1046 !sfw_test_active(tsi) || tsi->tsi_stopping) {
1047 spin_unlock(&tsi->tsi_lock);
1048 continue;
1049 }
1050
1051 tsi->tsi_stopping = 1;
1052
1053 if (!force) {
1054 spin_unlock(&tsi->tsi_lock);
1055 continue;
1056 }
1057
1058 /* abort launched rpcs in the test */
1059 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1060 spin_lock(&rpc->crpc_lock);
1061
1062 srpc_abort_rpc(rpc, -EINTR);
1063
1064 spin_unlock(&rpc->crpc_lock);
1065 }
1066
1067 spin_unlock(&tsi->tsi_lock);
1068 }
1069
1070 return 0;
1071 }
1072
1073 static int
sfw_query_batch(sfw_batch_t * tsb,int testidx,srpc_batch_reply_t * reply)1074 sfw_query_batch(sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1075 {
1076 sfw_test_instance_t *tsi;
1077
1078 if (testidx < 0)
1079 return -EINVAL;
1080
1081 if (testidx == 0) {
1082 reply->bar_active = atomic_read(&tsb->bat_nactive);
1083 return 0;
1084 }
1085
1086 list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1087 if (testidx-- > 1)
1088 continue;
1089
1090 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1091 return 0;
1092 }
1093
1094 return -ENOENT;
1095 }
1096
1097 void
sfw_free_pages(srpc_server_rpc_t * rpc)1098 sfw_free_pages(srpc_server_rpc_t *rpc)
1099 {
1100 srpc_free_bulk(rpc->srpc_bulk);
1101 rpc->srpc_bulk = NULL;
1102 }
1103
1104 int
sfw_alloc_pages(struct srpc_server_rpc * rpc,int cpt,int npages,int len,int sink)1105 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1106 int sink)
1107 {
1108 LASSERT(rpc->srpc_bulk == NULL);
1109 LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1110
1111 rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1112 if (rpc->srpc_bulk == NULL)
1113 return -ENOMEM;
1114
1115 return 0;
1116 }
1117
1118 static int
sfw_add_test(srpc_server_rpc_t * rpc)1119 sfw_add_test(srpc_server_rpc_t *rpc)
1120 {
1121 sfw_session_t *sn = sfw_data.fw_session;
1122 srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1123 srpc_test_reqst_t *request;
1124 int rc;
1125 sfw_batch_t *bat;
1126
1127 request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1128 reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1129
1130 if (request->tsr_loop == 0 ||
1131 request->tsr_concur == 0 ||
1132 request->tsr_sid.ses_nid == LNET_NID_ANY ||
1133 request->tsr_ndest > SFW_MAX_NDESTS ||
1134 (request->tsr_is_client && request->tsr_ndest == 0) ||
1135 request->tsr_concur > SFW_MAX_CONCUR ||
1136 request->tsr_service > SRPC_SERVICE_MAX_ID ||
1137 request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1138 reply->tsr_status = EINVAL;
1139 return 0;
1140 }
1141
1142 if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1143 sfw_find_test_case(request->tsr_service) == NULL) {
1144 reply->tsr_status = ENOENT;
1145 return 0;
1146 }
1147
1148 bat = sfw_bid2batch(request->tsr_bid);
1149 if (bat == NULL) {
1150 CERROR("Dropping RPC (%s) from %s under memory pressure.\n",
1151 rpc->srpc_scd->scd_svc->sv_name,
1152 libcfs_id2str(rpc->srpc_peer));
1153 return -ENOMEM;
1154 }
1155
1156 if (sfw_batch_active(bat)) {
1157 reply->tsr_status = EBUSY;
1158 return 0;
1159 }
1160
1161 if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1162 /* rpc will be resumed later in sfw_bulk_ready */
1163 int npg = sfw_id_pages(request->tsr_ndest);
1164 int len;
1165
1166 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1167 len = npg * PAGE_CACHE_SIZE;
1168
1169 } else {
1170 len = sizeof(lnet_process_id_packed_t) *
1171 request->tsr_ndest;
1172 }
1173
1174 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1175 }
1176
1177 rc = sfw_add_test_instance(bat, rpc);
1178 CDEBUG(rc == 0 ? D_NET : D_WARNING,
1179 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1180 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1181 request->tsr_is_client ? "client" : "server",
1182 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1183
1184 reply->tsr_status = (rc < 0) ? -rc : rc;
1185 return 0;
1186 }
1187
1188 static int
sfw_control_batch(srpc_batch_reqst_t * request,srpc_batch_reply_t * reply)1189 sfw_control_batch(srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1190 {
1191 sfw_session_t *sn = sfw_data.fw_session;
1192 int rc = 0;
1193 sfw_batch_t *bat;
1194
1195 reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1196
1197 if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1198 reply->bar_status = ESRCH;
1199 return 0;
1200 }
1201
1202 bat = sfw_find_batch(request->bar_bid);
1203 if (bat == NULL) {
1204 reply->bar_status = ENOENT;
1205 return 0;
1206 }
1207
1208 switch (request->bar_opc) {
1209 case SRPC_BATCH_OPC_RUN:
1210 rc = sfw_run_batch(bat);
1211 break;
1212
1213 case SRPC_BATCH_OPC_STOP:
1214 rc = sfw_stop_batch(bat, request->bar_arg);
1215 break;
1216
1217 case SRPC_BATCH_OPC_QUERY:
1218 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1219 break;
1220
1221 default:
1222 return -EINVAL; /* drop it */
1223 }
1224
1225 reply->bar_status = (rc < 0) ? -rc : rc;
1226 return 0;
1227 }
1228
1229 static int
sfw_handle_server_rpc(struct srpc_server_rpc * rpc)1230 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1231 {
1232 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1233 srpc_msg_t *reply = &rpc->srpc_replymsg;
1234 srpc_msg_t *request = &rpc->srpc_reqstbuf->buf_msg;
1235 unsigned features = LST_FEATS_MASK;
1236 int rc = 0;
1237
1238 LASSERT(sfw_data.fw_active_srpc == NULL);
1239 LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1240
1241 spin_lock(&sfw_data.fw_lock);
1242
1243 if (sfw_data.fw_shuttingdown) {
1244 spin_unlock(&sfw_data.fw_lock);
1245 return -ESHUTDOWN;
1246 }
1247
1248 /* Remove timer to avoid racing with it or expiring active session */
1249 if (sfw_del_session_timer() != 0) {
1250 CERROR("Dropping RPC (%s) from %s: racing with expiry timer.",
1251 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1252 spin_unlock(&sfw_data.fw_lock);
1253 return -EAGAIN;
1254 }
1255
1256 sfw_data.fw_active_srpc = rpc;
1257 spin_unlock(&sfw_data.fw_lock);
1258
1259 sfw_unpack_message(request);
1260 LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1261
1262 /* rpc module should have checked this */
1263 LASSERT(request->msg_version == SRPC_MSG_VERSION);
1264
1265 if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1266 sv->sv_id != SRPC_SERVICE_DEBUG) {
1267 sfw_session_t *sn = sfw_data.fw_session;
1268
1269 if (sn != NULL &&
1270 sn->sn_features != request->msg_ses_feats) {
1271 CNETERR("Features of framework RPC don't match features of current session: %x/%x\n",
1272 request->msg_ses_feats, sn->sn_features);
1273 reply->msg_body.reply.status = EPROTO;
1274 reply->msg_body.reply.sid = sn->sn_id;
1275 goto out;
1276 }
1277
1278 } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1279 /* NB: at this point, old version will ignore features and
1280 * create new session anyway, so console should be able
1281 * to handle this */
1282 reply->msg_body.reply.status = EPROTO;
1283 goto out;
1284 }
1285
1286 switch (sv->sv_id) {
1287 default:
1288 LBUG();
1289 case SRPC_SERVICE_TEST:
1290 rc = sfw_add_test(rpc);
1291 break;
1292
1293 case SRPC_SERVICE_BATCH:
1294 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1295 &reply->msg_body.bat_reply);
1296 break;
1297
1298 case SRPC_SERVICE_QUERY_STAT:
1299 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1300 &reply->msg_body.stat_reply);
1301 break;
1302
1303 case SRPC_SERVICE_DEBUG:
1304 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1305 &reply->msg_body.dbg_reply);
1306 break;
1307
1308 case SRPC_SERVICE_MAKE_SESSION:
1309 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1310 &reply->msg_body.mksn_reply);
1311 break;
1312
1313 case SRPC_SERVICE_REMOVE_SESSION:
1314 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1315 &reply->msg_body.rmsn_reply);
1316 break;
1317 }
1318
1319 if (sfw_data.fw_session != NULL)
1320 features = sfw_data.fw_session->sn_features;
1321 out:
1322 reply->msg_ses_feats = features;
1323 rpc->srpc_done = sfw_server_rpc_done;
1324 spin_lock(&sfw_data.fw_lock);
1325
1326 if (!sfw_data.fw_shuttingdown)
1327 sfw_add_session_timer();
1328
1329 sfw_data.fw_active_srpc = NULL;
1330 spin_unlock(&sfw_data.fw_lock);
1331 return rc;
1332 }
1333
1334 static int
sfw_bulk_ready(struct srpc_server_rpc * rpc,int status)1335 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1336 {
1337 struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1338 int rc;
1339
1340 LASSERT(rpc->srpc_bulk != NULL);
1341 LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1342 LASSERT(sfw_data.fw_active_srpc == NULL);
1343 LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1344
1345 spin_lock(&sfw_data.fw_lock);
1346
1347 if (status != 0) {
1348 CERROR("Bulk transfer failed for RPC: service %s, peer %s, status %d\n",
1349 sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1350 spin_unlock(&sfw_data.fw_lock);
1351 return -EIO;
1352 }
1353
1354 if (sfw_data.fw_shuttingdown) {
1355 spin_unlock(&sfw_data.fw_lock);
1356 return -ESHUTDOWN;
1357 }
1358
1359 if (sfw_del_session_timer() != 0) {
1360 CERROR("Dropping RPC (%s) from %s: racing with expiry timer",
1361 sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1362 spin_unlock(&sfw_data.fw_lock);
1363 return -EAGAIN;
1364 }
1365
1366 sfw_data.fw_active_srpc = rpc;
1367 spin_unlock(&sfw_data.fw_lock);
1368
1369 rc = sfw_add_test(rpc);
1370
1371 spin_lock(&sfw_data.fw_lock);
1372
1373 if (!sfw_data.fw_shuttingdown)
1374 sfw_add_session_timer();
1375
1376 sfw_data.fw_active_srpc = NULL;
1377 spin_unlock(&sfw_data.fw_lock);
1378 return rc;
1379 }
1380
1381 srpc_client_rpc_t *
sfw_create_rpc(lnet_process_id_t peer,int service,unsigned features,int nbulkiov,int bulklen,void (* done)(srpc_client_rpc_t *),void * priv)1382 sfw_create_rpc(lnet_process_id_t peer, int service,
1383 unsigned features, int nbulkiov, int bulklen,
1384 void (*done)(srpc_client_rpc_t *), void *priv)
1385 {
1386 srpc_client_rpc_t *rpc = NULL;
1387
1388 spin_lock(&sfw_data.fw_lock);
1389
1390 LASSERT(!sfw_data.fw_shuttingdown);
1391 LASSERT(service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1392
1393 if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1394 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1395 srpc_client_rpc_t, crpc_list);
1396 list_del(&rpc->crpc_list);
1397
1398 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1399 done, sfw_client_rpc_fini, priv);
1400 }
1401
1402 spin_unlock(&sfw_data.fw_lock);
1403
1404 if (rpc == NULL) {
1405 rpc = srpc_create_client_rpc(peer, service,
1406 nbulkiov, bulklen, done,
1407 nbulkiov != 0 ? NULL :
1408 sfw_client_rpc_fini,
1409 priv);
1410 }
1411
1412 if (rpc != NULL) /* "session" is concept in framework */
1413 rpc->crpc_reqstmsg.msg_ses_feats = features;
1414
1415 return rpc;
1416 }
1417
1418 void
sfw_unpack_message(srpc_msg_t * msg)1419 sfw_unpack_message(srpc_msg_t *msg)
1420 {
1421 if (msg->msg_magic == SRPC_MSG_MAGIC)
1422 return; /* no flipping needed */
1423
1424 /* srpc module should guarantee I wouldn't get crap */
1425 LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1426
1427 if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1428 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1429
1430 __swab32s(&req->str_type);
1431 __swab64s(&req->str_rpyid);
1432 sfw_unpack_sid(req->str_sid);
1433 return;
1434 }
1435
1436 if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1437 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1438
1439 __swab32s(&rep->str_status);
1440 sfw_unpack_sid(rep->str_sid);
1441 sfw_unpack_fw_counters(rep->str_fw);
1442 sfw_unpack_rpc_counters(rep->str_rpc);
1443 sfw_unpack_lnet_counters(rep->str_lnet);
1444 return;
1445 }
1446
1447 if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1448 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1449
1450 __swab64s(&req->mksn_rpyid);
1451 __swab32s(&req->mksn_force);
1452 sfw_unpack_sid(req->mksn_sid);
1453 return;
1454 }
1455
1456 if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1457 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1458
1459 __swab32s(&rep->mksn_status);
1460 __swab32s(&rep->mksn_timeout);
1461 sfw_unpack_sid(rep->mksn_sid);
1462 return;
1463 }
1464
1465 if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1466 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1467
1468 __swab64s(&req->rmsn_rpyid);
1469 sfw_unpack_sid(req->rmsn_sid);
1470 return;
1471 }
1472
1473 if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1474 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1475
1476 __swab32s(&rep->rmsn_status);
1477 sfw_unpack_sid(rep->rmsn_sid);
1478 return;
1479 }
1480
1481 if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1482 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1483
1484 __swab64s(&req->dbg_rpyid);
1485 __swab32s(&req->dbg_flags);
1486 sfw_unpack_sid(req->dbg_sid);
1487 return;
1488 }
1489
1490 if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1491 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1492
1493 __swab32s(&rep->dbg_nbatch);
1494 __swab32s(&rep->dbg_timeout);
1495 sfw_unpack_sid(rep->dbg_sid);
1496 return;
1497 }
1498
1499 if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1500 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1501
1502 __swab32s(&req->bar_opc);
1503 __swab64s(&req->bar_rpyid);
1504 __swab32s(&req->bar_testidx);
1505 __swab32s(&req->bar_arg);
1506 sfw_unpack_sid(req->bar_sid);
1507 __swab64s(&req->bar_bid.bat_id);
1508 return;
1509 }
1510
1511 if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1512 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1513
1514 __swab32s(&rep->bar_status);
1515 sfw_unpack_sid(rep->bar_sid);
1516 return;
1517 }
1518
1519 if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1520 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1521
1522 __swab64s(&req->tsr_rpyid);
1523 __swab64s(&req->tsr_bulkid);
1524 __swab32s(&req->tsr_loop);
1525 __swab32s(&req->tsr_ndest);
1526 __swab32s(&req->tsr_concur);
1527 __swab32s(&req->tsr_service);
1528 sfw_unpack_sid(req->tsr_sid);
1529 __swab64s(&req->tsr_bid.bat_id);
1530 return;
1531 }
1532
1533 if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1534 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1535
1536 __swab32s(&rep->tsr_status);
1537 sfw_unpack_sid(rep->tsr_sid);
1538 return;
1539 }
1540
1541 if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1542 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1543
1544 __swab64s(&req->join_rpyid);
1545 sfw_unpack_sid(req->join_sid);
1546 return;
1547 }
1548
1549 if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1550 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1551
1552 __swab32s(&rep->join_status);
1553 __swab32s(&rep->join_timeout);
1554 sfw_unpack_sid(rep->join_sid);
1555 return;
1556 }
1557
1558 LBUG();
1559 return;
1560 }
1561
1562 void
sfw_abort_rpc(srpc_client_rpc_t * rpc)1563 sfw_abort_rpc(srpc_client_rpc_t *rpc)
1564 {
1565 LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1566 LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1567
1568 spin_lock(&rpc->crpc_lock);
1569 srpc_abort_rpc(rpc, -EINTR);
1570 spin_unlock(&rpc->crpc_lock);
1571 return;
1572 }
1573
1574 void
sfw_post_rpc(srpc_client_rpc_t * rpc)1575 sfw_post_rpc(srpc_client_rpc_t *rpc)
1576 {
1577 spin_lock(&rpc->crpc_lock);
1578
1579 LASSERT(!rpc->crpc_closed);
1580 LASSERT(!rpc->crpc_aborted);
1581 LASSERT(list_empty(&rpc->crpc_list));
1582 LASSERT(!sfw_data.fw_shuttingdown);
1583
1584 rpc->crpc_timeout = rpc_timeout;
1585 srpc_post_rpc(rpc);
1586
1587 spin_unlock(&rpc->crpc_lock);
1588 return;
1589 }
1590
1591 static srpc_service_t sfw_services[] = {
1592 {
1593 /* sv_id */ SRPC_SERVICE_DEBUG,
1594 /* sv_name */ "debug",
1595 0
1596 },
1597 {
1598 /* sv_id */ SRPC_SERVICE_QUERY_STAT,
1599 /* sv_name */ "query stats",
1600 0
1601 },
1602 {
1603 /* sv_id */ SRPC_SERVICE_MAKE_SESSION,
1604 /* sv_name */ "make session",
1605 0
1606 },
1607 {
1608 /* sv_id */ SRPC_SERVICE_REMOVE_SESSION,
1609 /* sv_name */ "remove session",
1610 0
1611 },
1612 {
1613 /* sv_id */ SRPC_SERVICE_BATCH,
1614 /* sv_name */ "batch service",
1615 0
1616 },
1617 {
1618 /* sv_id */ SRPC_SERVICE_TEST,
1619 /* sv_name */ "test service",
1620 0
1621 },
1622 {
1623 /* sv_id */ 0,
1624 /* sv_name */ NULL,
1625 0
1626 }
1627 };
1628
1629 extern sfw_test_client_ops_t ping_test_client;
1630 extern srpc_service_t ping_test_service;
1631 extern void ping_init_test_client(void);
1632 extern void ping_init_test_service(void);
1633
1634 extern sfw_test_client_ops_t brw_test_client;
1635 extern srpc_service_t brw_test_service;
1636 extern void brw_init_test_client(void);
1637 extern void brw_init_test_service(void);
1638
1639
1640 int
sfw_startup(void)1641 sfw_startup(void)
1642 {
1643 int i;
1644 int rc;
1645 int error;
1646 srpc_service_t *sv;
1647 sfw_test_case_t *tsc;
1648
1649
1650 if (session_timeout < 0) {
1651 CERROR("Session timeout must be non-negative: %d\n",
1652 session_timeout);
1653 return -EINVAL;
1654 }
1655
1656 if (rpc_timeout < 0) {
1657 CERROR("RPC timeout must be non-negative: %d\n",
1658 rpc_timeout);
1659 return -EINVAL;
1660 }
1661
1662 if (session_timeout == 0)
1663 CWARN("Zero session_timeout specified - test sessions never expire.\n");
1664
1665 if (rpc_timeout == 0)
1666 CWARN("Zero rpc_timeout specified - test RPC never expire.\n");
1667
1668 memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1669
1670 sfw_data.fw_session = NULL;
1671 sfw_data.fw_active_srpc = NULL;
1672 spin_lock_init(&sfw_data.fw_lock);
1673 atomic_set(&sfw_data.fw_nzombies, 0);
1674 INIT_LIST_HEAD(&sfw_data.fw_tests);
1675 INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1676 INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1677
1678 brw_init_test_client();
1679 brw_init_test_service();
1680 rc = sfw_register_test(&brw_test_service, &brw_test_client);
1681 LASSERT(rc == 0);
1682
1683 ping_init_test_client();
1684 ping_init_test_service();
1685 rc = sfw_register_test(&ping_test_service, &ping_test_client);
1686 LASSERT(rc == 0);
1687
1688 error = 0;
1689 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1690 sv = tsc->tsc_srv_service;
1691
1692 rc = srpc_add_service(sv);
1693 LASSERT(rc != -EBUSY);
1694 if (rc != 0) {
1695 CWARN("Failed to add %s service: %d\n",
1696 sv->sv_name, rc);
1697 error = rc;
1698 }
1699 }
1700
1701 for (i = 0; ; i++) {
1702 sv = &sfw_services[i];
1703 if (sv->sv_name == NULL) break;
1704
1705 sv->sv_bulk_ready = NULL;
1706 sv->sv_handler = sfw_handle_server_rpc;
1707 sv->sv_wi_total = SFW_FRWK_WI_MAX;
1708 if (sv->sv_id == SRPC_SERVICE_TEST)
1709 sv->sv_bulk_ready = sfw_bulk_ready;
1710
1711 rc = srpc_add_service(sv);
1712 LASSERT(rc != -EBUSY);
1713 if (rc != 0) {
1714 CWARN("Failed to add %s service: %d\n",
1715 sv->sv_name, rc);
1716 error = rc;
1717 }
1718
1719 /* about to sfw_shutdown, no need to add buffer */
1720 if (error) continue;
1721
1722 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1723 if (rc != 0) {
1724 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
1725 sv->sv_name, sv->sv_wi_total, rc);
1726 error = -ENOMEM;
1727 }
1728 }
1729
1730 if (error != 0)
1731 sfw_shutdown();
1732 return error;
1733 }
1734
1735 void
sfw_shutdown(void)1736 sfw_shutdown(void)
1737 {
1738 srpc_service_t *sv;
1739 sfw_test_case_t *tsc;
1740 int i;
1741
1742 spin_lock(&sfw_data.fw_lock);
1743
1744 sfw_data.fw_shuttingdown = 1;
1745 lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1746 "waiting for active RPC to finish.\n");
1747
1748 if (sfw_del_session_timer() != 0)
1749 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1750 "waiting for session timer to explode.\n");
1751
1752 sfw_deactivate_session();
1753 lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1754 sfw_data.fw_lock,
1755 "waiting for %d zombie sessions to die.\n",
1756 atomic_read(&sfw_data.fw_nzombies));
1757
1758 spin_unlock(&sfw_data.fw_lock);
1759
1760 for (i = 0; ; i++) {
1761 sv = &sfw_services[i];
1762 if (sv->sv_name == NULL)
1763 break;
1764
1765 srpc_shutdown_service(sv);
1766 srpc_remove_service(sv);
1767 }
1768
1769 list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1770 sv = tsc->tsc_srv_service;
1771 srpc_shutdown_service(sv);
1772 srpc_remove_service(sv);
1773 }
1774
1775 while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1776 srpc_client_rpc_t *rpc;
1777
1778 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1779 srpc_client_rpc_t, crpc_list);
1780 list_del(&rpc->crpc_list);
1781
1782 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1783 }
1784
1785 for (i = 0; ; i++) {
1786 sv = &sfw_services[i];
1787 if (sv->sv_name == NULL)
1788 break;
1789
1790 srpc_wait_service_shutdown(sv);
1791 }
1792
1793 while (!list_empty(&sfw_data.fw_tests)) {
1794 tsc = list_entry(sfw_data.fw_tests.next,
1795 sfw_test_case_t, tsc_list);
1796
1797 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1798
1799 list_del(&tsc->tsc_list);
1800 LIBCFS_FREE(tsc, sizeof(*tsc));
1801 }
1802
1803 return;
1804 }
1805