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.gnu.org/licenses/gpl-2.0.html
19 *
20 * GPL HEADER END
21 */
22 /*
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
25 *
26 * Copyright (c) 2012, 2015, Intel Corporation.
27 */
28 /*
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
31 *
32 * lnet/selftest/rpc.c
33 *
34 * Author: Isaac Huang <isaac@clusterfs.com>
35 *
36 * 2012-05-13: Liang Zhen <liang@whamcloud.com>
37 * - percpt data for service to improve smp performance
38 * - code cleanup
39 */
40
41 #define DEBUG_SUBSYSTEM S_LNET
42
43 #include "selftest.h"
44
45 enum srpc_state {
46 SRPC_STATE_NONE,
47 SRPC_STATE_NI_INIT,
48 SRPC_STATE_EQ_INIT,
49 SRPC_STATE_RUNNING,
50 SRPC_STATE_STOPPING,
51 };
52
53 static struct smoketest_rpc {
54 spinlock_t rpc_glock; /* global lock */
55 struct srpc_service *rpc_services[SRPC_SERVICE_MAX_ID + 1];
56 lnet_handle_eq_t rpc_lnet_eq; /* _the_ LNet event queue */
57 enum srpc_state rpc_state;
58 srpc_counters_t rpc_counters;
59 __u64 rpc_matchbits; /* matchbits counter */
60 } srpc_data;
61
62 static inline int
srpc_serv_portal(int svc_id)63 srpc_serv_portal(int svc_id)
64 {
65 return svc_id < SRPC_FRAMEWORK_SERVICE_MAX_ID ?
66 SRPC_FRAMEWORK_REQUEST_PORTAL : SRPC_REQUEST_PORTAL;
67 }
68
69 /* forward ref's */
70 int srpc_handle_rpc(struct swi_workitem *wi);
71
srpc_get_counters(srpc_counters_t * cnt)72 void srpc_get_counters(srpc_counters_t *cnt)
73 {
74 spin_lock(&srpc_data.rpc_glock);
75 *cnt = srpc_data.rpc_counters;
76 spin_unlock(&srpc_data.rpc_glock);
77 }
78
srpc_set_counters(const srpc_counters_t * cnt)79 void srpc_set_counters(const srpc_counters_t *cnt)
80 {
81 spin_lock(&srpc_data.rpc_glock);
82 srpc_data.rpc_counters = *cnt;
83 spin_unlock(&srpc_data.rpc_glock);
84 }
85
86 static int
srpc_add_bulk_page(struct srpc_bulk * bk,struct page * pg,int i,int nob)87 srpc_add_bulk_page(struct srpc_bulk *bk, struct page *pg, int i, int nob)
88 {
89 nob = min_t(int, nob, PAGE_SIZE);
90
91 LASSERT(nob > 0);
92 LASSERT(i >= 0 && i < bk->bk_niov);
93
94 bk->bk_iovs[i].bv_offset = 0;
95 bk->bk_iovs[i].bv_page = pg;
96 bk->bk_iovs[i].bv_len = nob;
97 return nob;
98 }
99
100 void
srpc_free_bulk(struct srpc_bulk * bk)101 srpc_free_bulk(struct srpc_bulk *bk)
102 {
103 int i;
104 struct page *pg;
105
106 LASSERT(bk);
107
108 for (i = 0; i < bk->bk_niov; i++) {
109 pg = bk->bk_iovs[i].bv_page;
110 if (!pg)
111 break;
112
113 __free_page(pg);
114 }
115
116 LIBCFS_FREE(bk, offsetof(struct srpc_bulk, bk_iovs[bk->bk_niov]));
117 }
118
119 struct srpc_bulk *
srpc_alloc_bulk(int cpt,unsigned bulk_npg,unsigned bulk_len,int sink)120 srpc_alloc_bulk(int cpt, unsigned bulk_npg, unsigned bulk_len, int sink)
121 {
122 struct srpc_bulk *bk;
123 int i;
124
125 LASSERT(bulk_npg > 0 && bulk_npg <= LNET_MAX_IOV);
126
127 LIBCFS_CPT_ALLOC(bk, lnet_cpt_table(), cpt,
128 offsetof(struct srpc_bulk, bk_iovs[bulk_npg]));
129 if (!bk) {
130 CERROR("Can't allocate descriptor for %d pages\n", bulk_npg);
131 return NULL;
132 }
133
134 memset(bk, 0, offsetof(struct srpc_bulk, bk_iovs[bulk_npg]));
135 bk->bk_sink = sink;
136 bk->bk_len = bulk_len;
137 bk->bk_niov = bulk_npg;
138
139 for (i = 0; i < bulk_npg; i++) {
140 struct page *pg;
141 int nob;
142
143 pg = alloc_pages_node(cfs_cpt_spread_node(lnet_cpt_table(), cpt),
144 GFP_KERNEL, 0);
145 if (!pg) {
146 CERROR("Can't allocate page %d of %d\n", i, bulk_npg);
147 srpc_free_bulk(bk);
148 return NULL;
149 }
150
151 nob = srpc_add_bulk_page(bk, pg, i, bulk_len);
152 bulk_len -= nob;
153 }
154
155 return bk;
156 }
157
158 static inline __u64
srpc_next_id(void)159 srpc_next_id(void)
160 {
161 __u64 id;
162
163 spin_lock(&srpc_data.rpc_glock);
164 id = srpc_data.rpc_matchbits++;
165 spin_unlock(&srpc_data.rpc_glock);
166 return id;
167 }
168
169 static void
srpc_init_server_rpc(struct srpc_server_rpc * rpc,struct srpc_service_cd * scd,struct srpc_buffer * buffer)170 srpc_init_server_rpc(struct srpc_server_rpc *rpc,
171 struct srpc_service_cd *scd,
172 struct srpc_buffer *buffer)
173 {
174 memset(rpc, 0, sizeof(*rpc));
175 swi_init_workitem(&rpc->srpc_wi, rpc, srpc_handle_rpc,
176 srpc_serv_is_framework(scd->scd_svc) ?
177 lst_sched_serial : lst_sched_test[scd->scd_cpt]);
178
179 rpc->srpc_ev.ev_fired = 1; /* no event expected now */
180
181 rpc->srpc_scd = scd;
182 rpc->srpc_reqstbuf = buffer;
183 rpc->srpc_peer = buffer->buf_peer;
184 rpc->srpc_self = buffer->buf_self;
185 LNetInvalidateHandle(&rpc->srpc_replymdh);
186 }
187
188 static void
srpc_service_fini(struct srpc_service * svc)189 srpc_service_fini(struct srpc_service *svc)
190 {
191 struct srpc_service_cd *scd;
192 struct srpc_server_rpc *rpc;
193 struct srpc_buffer *buf;
194 struct list_head *q;
195 int i;
196
197 if (!svc->sv_cpt_data)
198 return;
199
200 cfs_percpt_for_each(scd, i, svc->sv_cpt_data) {
201 while (1) {
202 if (!list_empty(&scd->scd_buf_posted))
203 q = &scd->scd_buf_posted;
204 else if (!list_empty(&scd->scd_buf_blocked))
205 q = &scd->scd_buf_blocked;
206 else
207 break;
208
209 while (!list_empty(q)) {
210 buf = list_entry(q->next, struct srpc_buffer,
211 buf_list);
212 list_del(&buf->buf_list);
213 LIBCFS_FREE(buf, sizeof(*buf));
214 }
215 }
216
217 LASSERT(list_empty(&scd->scd_rpc_active));
218
219 while (!list_empty(&scd->scd_rpc_free)) {
220 rpc = list_entry(scd->scd_rpc_free.next,
221 struct srpc_server_rpc,
222 srpc_list);
223 list_del(&rpc->srpc_list);
224 LIBCFS_FREE(rpc, sizeof(*rpc));
225 }
226 }
227
228 cfs_percpt_free(svc->sv_cpt_data);
229 svc->sv_cpt_data = NULL;
230 }
231
232 static int
srpc_service_nrpcs(struct srpc_service * svc)233 srpc_service_nrpcs(struct srpc_service *svc)
234 {
235 int nrpcs = svc->sv_wi_total / svc->sv_ncpts;
236
237 return srpc_serv_is_framework(svc) ?
238 max(nrpcs, SFW_FRWK_WI_MIN) : max(nrpcs, SFW_TEST_WI_MIN);
239 }
240
241 int srpc_add_buffer(struct swi_workitem *wi);
242
243 static int
srpc_service_init(struct srpc_service * svc)244 srpc_service_init(struct srpc_service *svc)
245 {
246 struct srpc_service_cd *scd;
247 struct srpc_server_rpc *rpc;
248 int nrpcs;
249 int i;
250 int j;
251
252 svc->sv_shuttingdown = 0;
253
254 svc->sv_cpt_data = cfs_percpt_alloc(lnet_cpt_table(),
255 sizeof(**svc->sv_cpt_data));
256 if (!svc->sv_cpt_data)
257 return -ENOMEM;
258
259 svc->sv_ncpts = srpc_serv_is_framework(svc) ?
260 1 : cfs_cpt_number(lnet_cpt_table());
261 nrpcs = srpc_service_nrpcs(svc);
262
263 cfs_percpt_for_each(scd, i, svc->sv_cpt_data) {
264 scd->scd_cpt = i;
265 scd->scd_svc = svc;
266 spin_lock_init(&scd->scd_lock);
267 INIT_LIST_HEAD(&scd->scd_rpc_free);
268 INIT_LIST_HEAD(&scd->scd_rpc_active);
269 INIT_LIST_HEAD(&scd->scd_buf_posted);
270 INIT_LIST_HEAD(&scd->scd_buf_blocked);
271
272 scd->scd_ev.ev_data = scd;
273 scd->scd_ev.ev_type = SRPC_REQUEST_RCVD;
274
275 /*
276 * NB: don't use lst_sched_serial for adding buffer,
277 * see details in srpc_service_add_buffers()
278 */
279 swi_init_workitem(&scd->scd_buf_wi, scd,
280 srpc_add_buffer, lst_sched_test[i]);
281
282 if (i && srpc_serv_is_framework(svc)) {
283 /*
284 * NB: framework service only needs srpc_service_cd for
285 * one partition, but we allocate for all to make
286 * it easier to implement, it will waste a little
287 * memory but nobody should care about this
288 */
289 continue;
290 }
291
292 for (j = 0; j < nrpcs; j++) {
293 LIBCFS_CPT_ALLOC(rpc, lnet_cpt_table(),
294 i, sizeof(*rpc));
295 if (!rpc) {
296 srpc_service_fini(svc);
297 return -ENOMEM;
298 }
299 list_add(&rpc->srpc_list, &scd->scd_rpc_free);
300 }
301 }
302
303 return 0;
304 }
305
306 int
srpc_add_service(struct srpc_service * sv)307 srpc_add_service(struct srpc_service *sv)
308 {
309 int id = sv->sv_id;
310
311 LASSERT(0 <= id && id <= SRPC_SERVICE_MAX_ID);
312
313 if (srpc_service_init(sv))
314 return -ENOMEM;
315
316 spin_lock(&srpc_data.rpc_glock);
317
318 LASSERT(srpc_data.rpc_state == SRPC_STATE_RUNNING);
319
320 if (srpc_data.rpc_services[id]) {
321 spin_unlock(&srpc_data.rpc_glock);
322 goto failed;
323 }
324
325 srpc_data.rpc_services[id] = sv;
326 spin_unlock(&srpc_data.rpc_glock);
327
328 CDEBUG(D_NET, "Adding service: id %d, name %s\n", id, sv->sv_name);
329 return 0;
330
331 failed:
332 srpc_service_fini(sv);
333 return -EBUSY;
334 }
335
336 int
srpc_remove_service(struct srpc_service * sv)337 srpc_remove_service(struct srpc_service *sv)
338 {
339 int id = sv->sv_id;
340
341 spin_lock(&srpc_data.rpc_glock);
342
343 if (srpc_data.rpc_services[id] != sv) {
344 spin_unlock(&srpc_data.rpc_glock);
345 return -ENOENT;
346 }
347
348 srpc_data.rpc_services[id] = NULL;
349 spin_unlock(&srpc_data.rpc_glock);
350 return 0;
351 }
352
353 static int
srpc_post_passive_rdma(int portal,int local,__u64 matchbits,void * buf,int len,int options,lnet_process_id_t peer,lnet_handle_md_t * mdh,struct srpc_event * ev)354 srpc_post_passive_rdma(int portal, int local, __u64 matchbits, void *buf,
355 int len, int options, lnet_process_id_t peer,
356 lnet_handle_md_t *mdh, struct srpc_event *ev)
357 {
358 int rc;
359 lnet_md_t md;
360 lnet_handle_me_t meh;
361
362 rc = LNetMEAttach(portal, peer, matchbits, 0, LNET_UNLINK,
363 local ? LNET_INS_LOCAL : LNET_INS_AFTER, &meh);
364 if (rc) {
365 CERROR("LNetMEAttach failed: %d\n", rc);
366 LASSERT(rc == -ENOMEM);
367 return -ENOMEM;
368 }
369
370 md.threshold = 1;
371 md.user_ptr = ev;
372 md.start = buf;
373 md.length = len;
374 md.options = options;
375 md.eq_handle = srpc_data.rpc_lnet_eq;
376
377 rc = LNetMDAttach(meh, md, LNET_UNLINK, mdh);
378 if (rc) {
379 CERROR("LNetMDAttach failed: %d\n", rc);
380 LASSERT(rc == -ENOMEM);
381
382 rc = LNetMEUnlink(meh);
383 LASSERT(!rc);
384 return -ENOMEM;
385 }
386
387 CDEBUG(D_NET, "Posted passive RDMA: peer %s, portal %d, matchbits %#llx\n",
388 libcfs_id2str(peer), portal, matchbits);
389 return 0;
390 }
391
392 static int
srpc_post_active_rdma(int portal,__u64 matchbits,void * buf,int len,int options,lnet_process_id_t peer,lnet_nid_t self,lnet_handle_md_t * mdh,struct srpc_event * ev)393 srpc_post_active_rdma(int portal, __u64 matchbits, void *buf, int len,
394 int options, lnet_process_id_t peer, lnet_nid_t self,
395 lnet_handle_md_t *mdh, struct srpc_event *ev)
396 {
397 int rc;
398 lnet_md_t md;
399
400 md.user_ptr = ev;
401 md.start = buf;
402 md.length = len;
403 md.eq_handle = srpc_data.rpc_lnet_eq;
404 md.threshold = options & LNET_MD_OP_GET ? 2 : 1;
405 md.options = options & ~(LNET_MD_OP_PUT | LNET_MD_OP_GET);
406
407 rc = LNetMDBind(md, LNET_UNLINK, mdh);
408 if (rc) {
409 CERROR("LNetMDBind failed: %d\n", rc);
410 LASSERT(rc == -ENOMEM);
411 return -ENOMEM;
412 }
413
414 /*
415 * this is kind of an abuse of the LNET_MD_OP_{PUT,GET} options.
416 * they're only meaningful for MDs attached to an ME (i.e. passive
417 * buffers...
418 */
419 if (options & LNET_MD_OP_PUT) {
420 rc = LNetPut(self, *mdh, LNET_NOACK_REQ, peer,
421 portal, matchbits, 0, 0);
422 } else {
423 LASSERT(options & LNET_MD_OP_GET);
424
425 rc = LNetGet(self, *mdh, peer, portal, matchbits, 0);
426 }
427
428 if (rc) {
429 CERROR("LNet%s(%s, %d, %lld) failed: %d\n",
430 options & LNET_MD_OP_PUT ? "Put" : "Get",
431 libcfs_id2str(peer), portal, matchbits, rc);
432
433 /*
434 * The forthcoming unlink event will complete this operation
435 * with failure, so fall through and return success here.
436 */
437 rc = LNetMDUnlink(*mdh);
438 LASSERT(!rc);
439 } else {
440 CDEBUG(D_NET, "Posted active RDMA: peer %s, portal %u, matchbits %#llx\n",
441 libcfs_id2str(peer), portal, matchbits);
442 }
443 return 0;
444 }
445
446 static int
srpc_post_passive_rqtbuf(int service,int local,void * buf,int len,lnet_handle_md_t * mdh,struct srpc_event * ev)447 srpc_post_passive_rqtbuf(int service, int local, void *buf, int len,
448 lnet_handle_md_t *mdh, struct srpc_event *ev)
449 {
450 lnet_process_id_t any = { 0 };
451
452 any.nid = LNET_NID_ANY;
453 any.pid = LNET_PID_ANY;
454
455 return srpc_post_passive_rdma(srpc_serv_portal(service),
456 local, service, buf, len,
457 LNET_MD_OP_PUT, any, mdh, ev);
458 }
459
460 static int
srpc_service_post_buffer(struct srpc_service_cd * scd,struct srpc_buffer * buf)461 srpc_service_post_buffer(struct srpc_service_cd *scd, struct srpc_buffer *buf)
462 __must_hold(&scd->scd_lock)
463 {
464 struct srpc_service *sv = scd->scd_svc;
465 struct srpc_msg *msg = &buf->buf_msg;
466 int rc;
467
468 LNetInvalidateHandle(&buf->buf_mdh);
469 list_add(&buf->buf_list, &scd->scd_buf_posted);
470 scd->scd_buf_nposted++;
471 spin_unlock(&scd->scd_lock);
472
473 rc = srpc_post_passive_rqtbuf(sv->sv_id,
474 !srpc_serv_is_framework(sv),
475 msg, sizeof(*msg), &buf->buf_mdh,
476 &scd->scd_ev);
477
478 /*
479 * At this point, a RPC (new or delayed) may have arrived in
480 * msg and its event handler has been called. So we must add
481 * buf to scd_buf_posted _before_ dropping scd_lock
482 */
483 spin_lock(&scd->scd_lock);
484
485 if (!rc) {
486 if (!sv->sv_shuttingdown)
487 return 0;
488
489 spin_unlock(&scd->scd_lock);
490 /*
491 * srpc_shutdown_service might have tried to unlink me
492 * when my buf_mdh was still invalid
493 */
494 LNetMDUnlink(buf->buf_mdh);
495 spin_lock(&scd->scd_lock);
496 return 0;
497 }
498
499 scd->scd_buf_nposted--;
500 if (sv->sv_shuttingdown)
501 return rc; /* don't allow to change scd_buf_posted */
502
503 list_del(&buf->buf_list);
504 spin_unlock(&scd->scd_lock);
505
506 LIBCFS_FREE(buf, sizeof(*buf));
507
508 spin_lock(&scd->scd_lock);
509 return rc;
510 }
511
512 int
srpc_add_buffer(struct swi_workitem * wi)513 srpc_add_buffer(struct swi_workitem *wi)
514 {
515 struct srpc_service_cd *scd = wi->swi_workitem.wi_data;
516 struct srpc_buffer *buf;
517 int rc = 0;
518
519 /*
520 * it's called by workitem scheduler threads, these threads
521 * should have been set CPT affinity, so buffers will be posted
522 * on CPT local list of Portal
523 */
524 spin_lock(&scd->scd_lock);
525
526 while (scd->scd_buf_adjust > 0 &&
527 !scd->scd_svc->sv_shuttingdown) {
528 scd->scd_buf_adjust--; /* consume it */
529 scd->scd_buf_posting++;
530
531 spin_unlock(&scd->scd_lock);
532
533 LIBCFS_ALLOC(buf, sizeof(*buf));
534 if (!buf) {
535 CERROR("Failed to add new buf to service: %s\n",
536 scd->scd_svc->sv_name);
537 spin_lock(&scd->scd_lock);
538 rc = -ENOMEM;
539 break;
540 }
541
542 spin_lock(&scd->scd_lock);
543 if (scd->scd_svc->sv_shuttingdown) {
544 spin_unlock(&scd->scd_lock);
545 LIBCFS_FREE(buf, sizeof(*buf));
546
547 spin_lock(&scd->scd_lock);
548 rc = -ESHUTDOWN;
549 break;
550 }
551
552 rc = srpc_service_post_buffer(scd, buf);
553 if (rc)
554 break; /* buf has been freed inside */
555
556 LASSERT(scd->scd_buf_posting > 0);
557 scd->scd_buf_posting--;
558 scd->scd_buf_total++;
559 scd->scd_buf_low = max(2, scd->scd_buf_total / 4);
560 }
561
562 if (rc) {
563 scd->scd_buf_err_stamp = ktime_get_real_seconds();
564 scd->scd_buf_err = rc;
565
566 LASSERT(scd->scd_buf_posting > 0);
567 scd->scd_buf_posting--;
568 }
569
570 spin_unlock(&scd->scd_lock);
571 return 0;
572 }
573
574 int
srpc_service_add_buffers(struct srpc_service * sv,int nbuffer)575 srpc_service_add_buffers(struct srpc_service *sv, int nbuffer)
576 {
577 struct srpc_service_cd *scd;
578 int rc = 0;
579 int i;
580
581 LASSERTF(nbuffer > 0, "nbuffer must be positive: %d\n", nbuffer);
582
583 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
584 spin_lock(&scd->scd_lock);
585
586 scd->scd_buf_err = 0;
587 scd->scd_buf_err_stamp = 0;
588 scd->scd_buf_posting = 0;
589 scd->scd_buf_adjust = nbuffer;
590 /* start to post buffers */
591 swi_schedule_workitem(&scd->scd_buf_wi);
592 spin_unlock(&scd->scd_lock);
593
594 /* framework service only post buffer for one partition */
595 if (srpc_serv_is_framework(sv))
596 break;
597 }
598
599 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
600 spin_lock(&scd->scd_lock);
601 /*
602 * NB: srpc_service_add_buffers() can be called inside
603 * thread context of lst_sched_serial, and we don't normally
604 * allow to sleep inside thread context of WI scheduler
605 * because it will block current scheduler thread from doing
606 * anything else, even worse, it could deadlock if it's
607 * waiting on result from another WI of the same scheduler.
608 * However, it's safe at here because scd_buf_wi is scheduled
609 * by thread in a different WI scheduler (lst_sched_test),
610 * so we don't have any risk of deadlock, though this could
611 * block all WIs pending on lst_sched_serial for a moment
612 * which is not good but not fatal.
613 */
614 lst_wait_until(scd->scd_buf_err ||
615 (!scd->scd_buf_adjust &&
616 !scd->scd_buf_posting),
617 scd->scd_lock, "waiting for adding buffer\n");
618
619 if (scd->scd_buf_err && !rc)
620 rc = scd->scd_buf_err;
621
622 spin_unlock(&scd->scd_lock);
623 }
624
625 return rc;
626 }
627
628 void
srpc_service_remove_buffers(struct srpc_service * sv,int nbuffer)629 srpc_service_remove_buffers(struct srpc_service *sv, int nbuffer)
630 {
631 struct srpc_service_cd *scd;
632 int num;
633 int i;
634
635 LASSERT(!sv->sv_shuttingdown);
636
637 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
638 spin_lock(&scd->scd_lock);
639
640 num = scd->scd_buf_total + scd->scd_buf_posting;
641 scd->scd_buf_adjust -= min(nbuffer, num);
642
643 spin_unlock(&scd->scd_lock);
644 }
645 }
646
647 /* returns 1 if sv has finished, otherwise 0 */
648 int
srpc_finish_service(struct srpc_service * sv)649 srpc_finish_service(struct srpc_service *sv)
650 {
651 struct srpc_service_cd *scd;
652 struct srpc_server_rpc *rpc;
653 int i;
654
655 LASSERT(sv->sv_shuttingdown); /* srpc_shutdown_service called */
656
657 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
658 spin_lock(&scd->scd_lock);
659 if (!swi_deschedule_workitem(&scd->scd_buf_wi)) {
660 spin_unlock(&scd->scd_lock);
661 return 0;
662 }
663
664 if (scd->scd_buf_nposted > 0) {
665 CDEBUG(D_NET, "waiting for %d posted buffers to unlink\n",
666 scd->scd_buf_nposted);
667 spin_unlock(&scd->scd_lock);
668 return 0;
669 }
670
671 if (list_empty(&scd->scd_rpc_active)) {
672 spin_unlock(&scd->scd_lock);
673 continue;
674 }
675
676 rpc = list_entry(scd->scd_rpc_active.next,
677 struct srpc_server_rpc, srpc_list);
678 CNETERR("Active RPC %p on shutdown: sv %s, peer %s, wi %s scheduled %d running %d, ev fired %d type %d status %d lnet %d\n",
679 rpc, sv->sv_name, libcfs_id2str(rpc->srpc_peer),
680 swi_state2str(rpc->srpc_wi.swi_state),
681 rpc->srpc_wi.swi_workitem.wi_scheduled,
682 rpc->srpc_wi.swi_workitem.wi_running,
683 rpc->srpc_ev.ev_fired, rpc->srpc_ev.ev_type,
684 rpc->srpc_ev.ev_status, rpc->srpc_ev.ev_lnet);
685 spin_unlock(&scd->scd_lock);
686 return 0;
687 }
688
689 /* no lock needed from now on */
690 srpc_service_fini(sv);
691 return 1;
692 }
693
694 /* called with sv->sv_lock held */
695 static void
srpc_service_recycle_buffer(struct srpc_service_cd * scd,struct srpc_buffer * buf)696 srpc_service_recycle_buffer(struct srpc_service_cd *scd, struct srpc_buffer *buf)
697 __must_hold(&scd->scd_lock)
698 {
699 if (!scd->scd_svc->sv_shuttingdown && scd->scd_buf_adjust >= 0) {
700 if (srpc_service_post_buffer(scd, buf)) {
701 CWARN("Failed to post %s buffer\n",
702 scd->scd_svc->sv_name);
703 }
704 return;
705 }
706
707 /* service is shutting down, or we want to recycle some buffers */
708 scd->scd_buf_total--;
709
710 if (scd->scd_buf_adjust < 0) {
711 scd->scd_buf_adjust++;
712 if (scd->scd_buf_adjust < 0 &&
713 !scd->scd_buf_total && !scd->scd_buf_posting) {
714 CDEBUG(D_INFO,
715 "Try to recycle %d buffers but nothing left\n",
716 scd->scd_buf_adjust);
717 scd->scd_buf_adjust = 0;
718 }
719 }
720
721 spin_unlock(&scd->scd_lock);
722 LIBCFS_FREE(buf, sizeof(*buf));
723 spin_lock(&scd->scd_lock);
724 }
725
726 void
srpc_abort_service(struct srpc_service * sv)727 srpc_abort_service(struct srpc_service *sv)
728 {
729 struct srpc_service_cd *scd;
730 struct srpc_server_rpc *rpc;
731 int i;
732
733 CDEBUG(D_NET, "Aborting service: id %d, name %s\n",
734 sv->sv_id, sv->sv_name);
735
736 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
737 spin_lock(&scd->scd_lock);
738
739 /*
740 * schedule in-flight RPCs to notice the abort, NB:
741 * racing with incoming RPCs; complete fix should make test
742 * RPCs carry session ID in its headers
743 */
744 list_for_each_entry(rpc, &scd->scd_rpc_active, srpc_list) {
745 rpc->srpc_aborted = 1;
746 swi_schedule_workitem(&rpc->srpc_wi);
747 }
748
749 spin_unlock(&scd->scd_lock);
750 }
751 }
752
753 void
srpc_shutdown_service(struct srpc_service * sv)754 srpc_shutdown_service(struct srpc_service *sv)
755 {
756 struct srpc_service_cd *scd;
757 struct srpc_server_rpc *rpc;
758 struct srpc_buffer *buf;
759 int i;
760
761 CDEBUG(D_NET, "Shutting down service: id %d, name %s\n",
762 sv->sv_id, sv->sv_name);
763
764 cfs_percpt_for_each(scd, i, sv->sv_cpt_data)
765 spin_lock(&scd->scd_lock);
766
767 sv->sv_shuttingdown = 1; /* i.e. no new active RPC */
768
769 cfs_percpt_for_each(scd, i, sv->sv_cpt_data)
770 spin_unlock(&scd->scd_lock);
771
772 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
773 spin_lock(&scd->scd_lock);
774
775 /* schedule in-flight RPCs to notice the shutdown */
776 list_for_each_entry(rpc, &scd->scd_rpc_active, srpc_list)
777 swi_schedule_workitem(&rpc->srpc_wi);
778
779 spin_unlock(&scd->scd_lock);
780
781 /*
782 * OK to traverse scd_buf_posted without lock, since no one
783 * touches scd_buf_posted now
784 */
785 list_for_each_entry(buf, &scd->scd_buf_posted, buf_list)
786 LNetMDUnlink(buf->buf_mdh);
787 }
788 }
789
790 static int
srpc_send_request(struct srpc_client_rpc * rpc)791 srpc_send_request(struct srpc_client_rpc *rpc)
792 {
793 struct srpc_event *ev = &rpc->crpc_reqstev;
794 int rc;
795
796 ev->ev_fired = 0;
797 ev->ev_data = rpc;
798 ev->ev_type = SRPC_REQUEST_SENT;
799
800 rc = srpc_post_active_rdma(srpc_serv_portal(rpc->crpc_service),
801 rpc->crpc_service, &rpc->crpc_reqstmsg,
802 sizeof(struct srpc_msg), LNET_MD_OP_PUT,
803 rpc->crpc_dest, LNET_NID_ANY,
804 &rpc->crpc_reqstmdh, ev);
805 if (rc) {
806 LASSERT(rc == -ENOMEM);
807 ev->ev_fired = 1; /* no more event expected */
808 }
809 return rc;
810 }
811
812 static int
srpc_prepare_reply(struct srpc_client_rpc * rpc)813 srpc_prepare_reply(struct srpc_client_rpc *rpc)
814 {
815 struct srpc_event *ev = &rpc->crpc_replyev;
816 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.rpyid;
817 int rc;
818
819 ev->ev_fired = 0;
820 ev->ev_data = rpc;
821 ev->ev_type = SRPC_REPLY_RCVD;
822
823 *id = srpc_next_id();
824
825 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, 0, *id,
826 &rpc->crpc_replymsg,
827 sizeof(struct srpc_msg),
828 LNET_MD_OP_PUT, rpc->crpc_dest,
829 &rpc->crpc_replymdh, ev);
830 if (rc) {
831 LASSERT(rc == -ENOMEM);
832 ev->ev_fired = 1; /* no more event expected */
833 }
834 return rc;
835 }
836
837 static int
srpc_prepare_bulk(struct srpc_client_rpc * rpc)838 srpc_prepare_bulk(struct srpc_client_rpc *rpc)
839 {
840 struct srpc_bulk *bk = &rpc->crpc_bulk;
841 struct srpc_event *ev = &rpc->crpc_bulkev;
842 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.bulkid;
843 int rc;
844 int opt;
845
846 LASSERT(bk->bk_niov <= LNET_MAX_IOV);
847
848 if (!bk->bk_niov)
849 return 0; /* nothing to do */
850
851 opt = bk->bk_sink ? LNET_MD_OP_PUT : LNET_MD_OP_GET;
852 opt |= LNET_MD_KIOV;
853
854 ev->ev_fired = 0;
855 ev->ev_data = rpc;
856 ev->ev_type = SRPC_BULK_REQ_RCVD;
857
858 *id = srpc_next_id();
859
860 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, 0, *id,
861 &bk->bk_iovs[0], bk->bk_niov, opt,
862 rpc->crpc_dest, &bk->bk_mdh, ev);
863 if (rc) {
864 LASSERT(rc == -ENOMEM);
865 ev->ev_fired = 1; /* no more event expected */
866 }
867 return rc;
868 }
869
870 static int
srpc_do_bulk(struct srpc_server_rpc * rpc)871 srpc_do_bulk(struct srpc_server_rpc *rpc)
872 {
873 struct srpc_event *ev = &rpc->srpc_ev;
874 struct srpc_bulk *bk = rpc->srpc_bulk;
875 __u64 id = rpc->srpc_reqstbuf->buf_msg.msg_body.reqst.bulkid;
876 int rc;
877 int opt;
878
879 LASSERT(bk);
880
881 opt = bk->bk_sink ? LNET_MD_OP_GET : LNET_MD_OP_PUT;
882 opt |= LNET_MD_KIOV;
883
884 ev->ev_fired = 0;
885 ev->ev_data = rpc;
886 ev->ev_type = bk->bk_sink ? SRPC_BULK_GET_RPLD : SRPC_BULK_PUT_SENT;
887
888 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, id,
889 &bk->bk_iovs[0], bk->bk_niov, opt,
890 rpc->srpc_peer, rpc->srpc_self,
891 &bk->bk_mdh, ev);
892 if (rc)
893 ev->ev_fired = 1; /* no more event expected */
894 return rc;
895 }
896
897 /* only called from srpc_handle_rpc */
898 static void
srpc_server_rpc_done(struct srpc_server_rpc * rpc,int status)899 srpc_server_rpc_done(struct srpc_server_rpc *rpc, int status)
900 {
901 struct srpc_service_cd *scd = rpc->srpc_scd;
902 struct srpc_service *sv = scd->scd_svc;
903 struct srpc_buffer *buffer;
904
905 LASSERT(status || rpc->srpc_wi.swi_state == SWI_STATE_DONE);
906
907 rpc->srpc_status = status;
908
909 CDEBUG_LIMIT(!status ? D_NET : D_NETERROR,
910 "Server RPC %p done: service %s, peer %s, status %s:%d\n",
911 rpc, sv->sv_name, libcfs_id2str(rpc->srpc_peer),
912 swi_state2str(rpc->srpc_wi.swi_state), status);
913
914 if (status) {
915 spin_lock(&srpc_data.rpc_glock);
916 srpc_data.rpc_counters.rpcs_dropped++;
917 spin_unlock(&srpc_data.rpc_glock);
918 }
919
920 if (rpc->srpc_done)
921 (*rpc->srpc_done) (rpc);
922 LASSERT(!rpc->srpc_bulk);
923
924 spin_lock(&scd->scd_lock);
925
926 if (rpc->srpc_reqstbuf) {
927 /*
928 * NB might drop sv_lock in srpc_service_recycle_buffer, but
929 * sv won't go away for scd_rpc_active must not be empty
930 */
931 srpc_service_recycle_buffer(scd, rpc->srpc_reqstbuf);
932 rpc->srpc_reqstbuf = NULL;
933 }
934
935 list_del(&rpc->srpc_list); /* from scd->scd_rpc_active */
936
937 /*
938 * No one can schedule me now since:
939 * - I'm not on scd_rpc_active.
940 * - all LNet events have been fired.
941 * Cancel pending schedules and prevent future schedule attempts:
942 */
943 LASSERT(rpc->srpc_ev.ev_fired);
944 swi_exit_workitem(&rpc->srpc_wi);
945
946 if (!sv->sv_shuttingdown && !list_empty(&scd->scd_buf_blocked)) {
947 buffer = list_entry(scd->scd_buf_blocked.next,
948 struct srpc_buffer, buf_list);
949 list_del(&buffer->buf_list);
950
951 srpc_init_server_rpc(rpc, scd, buffer);
952 list_add_tail(&rpc->srpc_list, &scd->scd_rpc_active);
953 swi_schedule_workitem(&rpc->srpc_wi);
954 } else {
955 list_add(&rpc->srpc_list, &scd->scd_rpc_free);
956 }
957
958 spin_unlock(&scd->scd_lock);
959 }
960
961 /* handles an incoming RPC */
962 int
srpc_handle_rpc(struct swi_workitem * wi)963 srpc_handle_rpc(struct swi_workitem *wi)
964 {
965 struct srpc_server_rpc *rpc = wi->swi_workitem.wi_data;
966 struct srpc_service_cd *scd = rpc->srpc_scd;
967 struct srpc_service *sv = scd->scd_svc;
968 struct srpc_event *ev = &rpc->srpc_ev;
969 int rc = 0;
970
971 LASSERT(wi == &rpc->srpc_wi);
972
973 spin_lock(&scd->scd_lock);
974
975 if (sv->sv_shuttingdown || rpc->srpc_aborted) {
976 spin_unlock(&scd->scd_lock);
977
978 if (rpc->srpc_bulk)
979 LNetMDUnlink(rpc->srpc_bulk->bk_mdh);
980 LNetMDUnlink(rpc->srpc_replymdh);
981
982 if (ev->ev_fired) { /* no more event, OK to finish */
983 srpc_server_rpc_done(rpc, -ESHUTDOWN);
984 return 1;
985 }
986 return 0;
987 }
988
989 spin_unlock(&scd->scd_lock);
990
991 switch (wi->swi_state) {
992 default:
993 LBUG();
994 case SWI_STATE_NEWBORN: {
995 struct srpc_msg *msg;
996 struct srpc_generic_reply *reply;
997
998 msg = &rpc->srpc_reqstbuf->buf_msg;
999 reply = &rpc->srpc_replymsg.msg_body.reply;
1000
1001 if (!msg->msg_magic) {
1002 /* moaned already in srpc_lnet_ev_handler */
1003 srpc_server_rpc_done(rpc, EBADMSG);
1004 return 1;
1005 }
1006
1007 srpc_unpack_msg_hdr(msg);
1008 if (msg->msg_version != SRPC_MSG_VERSION) {
1009 CWARN("Version mismatch: %u, %u expected, from %s\n",
1010 msg->msg_version, SRPC_MSG_VERSION,
1011 libcfs_id2str(rpc->srpc_peer));
1012 reply->status = EPROTO;
1013 /* drop through and send reply */
1014 } else {
1015 reply->status = 0;
1016 rc = (*sv->sv_handler)(rpc);
1017 LASSERT(!reply->status || !rpc->srpc_bulk);
1018 if (rc) {
1019 srpc_server_rpc_done(rpc, rc);
1020 return 1;
1021 }
1022 }
1023
1024 wi->swi_state = SWI_STATE_BULK_STARTED;
1025
1026 if (rpc->srpc_bulk) {
1027 rc = srpc_do_bulk(rpc);
1028 if (!rc)
1029 return 0; /* wait for bulk */
1030
1031 LASSERT(ev->ev_fired);
1032 ev->ev_status = rc;
1033 }
1034 }
1035 case SWI_STATE_BULK_STARTED:
1036 LASSERT(!rpc->srpc_bulk || ev->ev_fired);
1037
1038 if (rpc->srpc_bulk) {
1039 rc = ev->ev_status;
1040
1041 if (sv->sv_bulk_ready)
1042 rc = (*sv->sv_bulk_ready) (rpc, rc);
1043
1044 if (rc) {
1045 srpc_server_rpc_done(rpc, rc);
1046 return 1;
1047 }
1048 }
1049
1050 wi->swi_state = SWI_STATE_REPLY_SUBMITTED;
1051 rc = srpc_send_reply(rpc);
1052 if (!rc)
1053 return 0; /* wait for reply */
1054 srpc_server_rpc_done(rpc, rc);
1055 return 1;
1056
1057 case SWI_STATE_REPLY_SUBMITTED:
1058 if (!ev->ev_fired) {
1059 CERROR("RPC %p: bulk %p, service %d\n",
1060 rpc, rpc->srpc_bulk, sv->sv_id);
1061 CERROR("Event: status %d, type %d, lnet %d\n",
1062 ev->ev_status, ev->ev_type, ev->ev_lnet);
1063 LASSERT(ev->ev_fired);
1064 }
1065
1066 wi->swi_state = SWI_STATE_DONE;
1067 srpc_server_rpc_done(rpc, ev->ev_status);
1068 return 1;
1069 }
1070
1071 return 0;
1072 }
1073
1074 static void
srpc_client_rpc_expired(void * data)1075 srpc_client_rpc_expired(void *data)
1076 {
1077 struct srpc_client_rpc *rpc = data;
1078
1079 CWARN("Client RPC expired: service %d, peer %s, timeout %d.\n",
1080 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1081 rpc->crpc_timeout);
1082
1083 spin_lock(&rpc->crpc_lock);
1084
1085 rpc->crpc_timeout = 0;
1086 srpc_abort_rpc(rpc, -ETIMEDOUT);
1087
1088 spin_unlock(&rpc->crpc_lock);
1089
1090 spin_lock(&srpc_data.rpc_glock);
1091 srpc_data.rpc_counters.rpcs_expired++;
1092 spin_unlock(&srpc_data.rpc_glock);
1093 }
1094
1095 static void
srpc_add_client_rpc_timer(struct srpc_client_rpc * rpc)1096 srpc_add_client_rpc_timer(struct srpc_client_rpc *rpc)
1097 {
1098 struct stt_timer *timer = &rpc->crpc_timer;
1099
1100 if (!rpc->crpc_timeout)
1101 return;
1102
1103 INIT_LIST_HEAD(&timer->stt_list);
1104 timer->stt_data = rpc;
1105 timer->stt_func = srpc_client_rpc_expired;
1106 timer->stt_expires = ktime_get_real_seconds() + rpc->crpc_timeout;
1107 stt_add_timer(timer);
1108 }
1109
1110 /*
1111 * Called with rpc->crpc_lock held.
1112 *
1113 * Upon exit the RPC expiry timer is not queued and the handler is not
1114 * running on any CPU.
1115 */
1116 static void
srpc_del_client_rpc_timer(struct srpc_client_rpc * rpc)1117 srpc_del_client_rpc_timer(struct srpc_client_rpc *rpc)
1118 {
1119 /* timer not planted or already exploded */
1120 if (!rpc->crpc_timeout)
1121 return;
1122
1123 /* timer successfully defused */
1124 if (stt_del_timer(&rpc->crpc_timer))
1125 return;
1126
1127 /* timer detonated, wait for it to explode */
1128 while (rpc->crpc_timeout) {
1129 spin_unlock(&rpc->crpc_lock);
1130
1131 schedule();
1132
1133 spin_lock(&rpc->crpc_lock);
1134 }
1135 }
1136
1137 static void
srpc_client_rpc_done(struct srpc_client_rpc * rpc,int status)1138 srpc_client_rpc_done(struct srpc_client_rpc *rpc, int status)
1139 {
1140 struct swi_workitem *wi = &rpc->crpc_wi;
1141
1142 LASSERT(status || wi->swi_state == SWI_STATE_DONE);
1143
1144 spin_lock(&rpc->crpc_lock);
1145
1146 rpc->crpc_closed = 1;
1147 if (!rpc->crpc_status)
1148 rpc->crpc_status = status;
1149
1150 srpc_del_client_rpc_timer(rpc);
1151
1152 CDEBUG_LIMIT(!status ? D_NET : D_NETERROR,
1153 "Client RPC done: service %d, peer %s, status %s:%d:%d\n",
1154 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1155 swi_state2str(wi->swi_state), rpc->crpc_aborted, status);
1156
1157 /*
1158 * No one can schedule me now since:
1159 * - RPC timer has been defused.
1160 * - all LNet events have been fired.
1161 * - crpc_closed has been set, preventing srpc_abort_rpc from
1162 * scheduling me.
1163 * Cancel pending schedules and prevent future schedule attempts:
1164 */
1165 LASSERT(!srpc_event_pending(rpc));
1166 swi_exit_workitem(wi);
1167
1168 spin_unlock(&rpc->crpc_lock);
1169
1170 (*rpc->crpc_done)(rpc);
1171 }
1172
1173 /* sends an outgoing RPC */
1174 int
srpc_send_rpc(struct swi_workitem * wi)1175 srpc_send_rpc(struct swi_workitem *wi)
1176 {
1177 int rc = 0;
1178 struct srpc_client_rpc *rpc;
1179 struct srpc_msg *reply;
1180 int do_bulk;
1181
1182 LASSERT(wi);
1183
1184 rpc = wi->swi_workitem.wi_data;
1185
1186 LASSERT(rpc);
1187 LASSERT(wi == &rpc->crpc_wi);
1188
1189 reply = &rpc->crpc_replymsg;
1190 do_bulk = rpc->crpc_bulk.bk_niov > 0;
1191
1192 spin_lock(&rpc->crpc_lock);
1193
1194 if (rpc->crpc_aborted) {
1195 spin_unlock(&rpc->crpc_lock);
1196 goto abort;
1197 }
1198
1199 spin_unlock(&rpc->crpc_lock);
1200
1201 switch (wi->swi_state) {
1202 default:
1203 LBUG();
1204 case SWI_STATE_NEWBORN:
1205 LASSERT(!srpc_event_pending(rpc));
1206
1207 rc = srpc_prepare_reply(rpc);
1208 if (rc) {
1209 srpc_client_rpc_done(rpc, rc);
1210 return 1;
1211 }
1212
1213 rc = srpc_prepare_bulk(rpc);
1214 if (rc)
1215 break;
1216
1217 wi->swi_state = SWI_STATE_REQUEST_SUBMITTED;
1218 rc = srpc_send_request(rpc);
1219 break;
1220
1221 case SWI_STATE_REQUEST_SUBMITTED:
1222 /*
1223 * CAVEAT EMPTOR: rqtev, rpyev, and bulkev may come in any
1224 * order; however, they're processed in a strict order:
1225 * rqt, rpy, and bulk.
1226 */
1227 if (!rpc->crpc_reqstev.ev_fired)
1228 break;
1229
1230 rc = rpc->crpc_reqstev.ev_status;
1231 if (rc)
1232 break;
1233
1234 wi->swi_state = SWI_STATE_REQUEST_SENT;
1235 /* perhaps more events, fall thru */
1236 case SWI_STATE_REQUEST_SENT: {
1237 enum srpc_msg_type type = srpc_service2reply(rpc->crpc_service);
1238
1239 if (!rpc->crpc_replyev.ev_fired)
1240 break;
1241
1242 rc = rpc->crpc_replyev.ev_status;
1243 if (rc)
1244 break;
1245
1246 srpc_unpack_msg_hdr(reply);
1247 if (reply->msg_type != type ||
1248 (reply->msg_magic != SRPC_MSG_MAGIC &&
1249 reply->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1250 CWARN("Bad message from %s: type %u (%d expected), magic %u (%d expected).\n",
1251 libcfs_id2str(rpc->crpc_dest),
1252 reply->msg_type, type,
1253 reply->msg_magic, SRPC_MSG_MAGIC);
1254 rc = -EBADMSG;
1255 break;
1256 }
1257
1258 if (do_bulk && reply->msg_body.reply.status) {
1259 CWARN("Remote error %d at %s, unlink bulk buffer in case peer didn't initiate bulk transfer\n",
1260 reply->msg_body.reply.status,
1261 libcfs_id2str(rpc->crpc_dest));
1262 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1263 }
1264
1265 wi->swi_state = SWI_STATE_REPLY_RECEIVED;
1266 }
1267 case SWI_STATE_REPLY_RECEIVED:
1268 if (do_bulk && !rpc->crpc_bulkev.ev_fired)
1269 break;
1270
1271 rc = do_bulk ? rpc->crpc_bulkev.ev_status : 0;
1272
1273 /*
1274 * Bulk buffer was unlinked due to remote error. Clear error
1275 * since reply buffer still contains valid data.
1276 * NB rpc->crpc_done shouldn't look into bulk data in case of
1277 * remote error.
1278 */
1279 if (do_bulk && rpc->crpc_bulkev.ev_lnet == LNET_EVENT_UNLINK &&
1280 !rpc->crpc_status && reply->msg_body.reply.status)
1281 rc = 0;
1282
1283 wi->swi_state = SWI_STATE_DONE;
1284 srpc_client_rpc_done(rpc, rc);
1285 return 1;
1286 }
1287
1288 if (rc) {
1289 spin_lock(&rpc->crpc_lock);
1290 srpc_abort_rpc(rpc, rc);
1291 spin_unlock(&rpc->crpc_lock);
1292 }
1293
1294 abort:
1295 if (rpc->crpc_aborted) {
1296 LNetMDUnlink(rpc->crpc_reqstmdh);
1297 LNetMDUnlink(rpc->crpc_replymdh);
1298 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1299
1300 if (!srpc_event_pending(rpc)) {
1301 srpc_client_rpc_done(rpc, -EINTR);
1302 return 1;
1303 }
1304 }
1305 return 0;
1306 }
1307
1308 struct srpc_client_rpc *
srpc_create_client_rpc(lnet_process_id_t peer,int service,int nbulkiov,int bulklen,void (* rpc_done)(struct srpc_client_rpc *),void (* rpc_fini)(struct srpc_client_rpc *),void * priv)1309 srpc_create_client_rpc(lnet_process_id_t peer, int service,
1310 int nbulkiov, int bulklen,
1311 void (*rpc_done)(struct srpc_client_rpc *),
1312 void (*rpc_fini)(struct srpc_client_rpc *), void *priv)
1313 {
1314 struct srpc_client_rpc *rpc;
1315
1316 LIBCFS_ALLOC(rpc, offsetof(struct srpc_client_rpc,
1317 crpc_bulk.bk_iovs[nbulkiov]));
1318 if (!rpc)
1319 return NULL;
1320
1321 srpc_init_client_rpc(rpc, peer, service, nbulkiov,
1322 bulklen, rpc_done, rpc_fini, priv);
1323 return rpc;
1324 }
1325
1326 /* called with rpc->crpc_lock held */
1327 void
srpc_abort_rpc(struct srpc_client_rpc * rpc,int why)1328 srpc_abort_rpc(struct srpc_client_rpc *rpc, int why)
1329 {
1330 LASSERT(why);
1331
1332 if (rpc->crpc_aborted || /* already aborted */
1333 rpc->crpc_closed) /* callback imminent */
1334 return;
1335
1336 CDEBUG(D_NET, "Aborting RPC: service %d, peer %s, state %s, why %d\n",
1337 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1338 swi_state2str(rpc->crpc_wi.swi_state), why);
1339
1340 rpc->crpc_aborted = 1;
1341 rpc->crpc_status = why;
1342 swi_schedule_workitem(&rpc->crpc_wi);
1343 }
1344
1345 /* called with rpc->crpc_lock held */
1346 void
srpc_post_rpc(struct srpc_client_rpc * rpc)1347 srpc_post_rpc(struct srpc_client_rpc *rpc)
1348 {
1349 LASSERT(!rpc->crpc_aborted);
1350 LASSERT(srpc_data.rpc_state == SRPC_STATE_RUNNING);
1351
1352 CDEBUG(D_NET, "Posting RPC: peer %s, service %d, timeout %d\n",
1353 libcfs_id2str(rpc->crpc_dest), rpc->crpc_service,
1354 rpc->crpc_timeout);
1355
1356 srpc_add_client_rpc_timer(rpc);
1357 swi_schedule_workitem(&rpc->crpc_wi);
1358 }
1359
1360 int
srpc_send_reply(struct srpc_server_rpc * rpc)1361 srpc_send_reply(struct srpc_server_rpc *rpc)
1362 {
1363 struct srpc_event *ev = &rpc->srpc_ev;
1364 struct srpc_msg *msg = &rpc->srpc_replymsg;
1365 struct srpc_buffer *buffer = rpc->srpc_reqstbuf;
1366 struct srpc_service_cd *scd = rpc->srpc_scd;
1367 struct srpc_service *sv = scd->scd_svc;
1368 __u64 rpyid;
1369 int rc;
1370
1371 LASSERT(buffer);
1372 rpyid = buffer->buf_msg.msg_body.reqst.rpyid;
1373
1374 spin_lock(&scd->scd_lock);
1375
1376 if (!sv->sv_shuttingdown && !srpc_serv_is_framework(sv)) {
1377 /*
1378 * Repost buffer before replying since test client
1379 * might send me another RPC once it gets the reply
1380 */
1381 if (srpc_service_post_buffer(scd, buffer))
1382 CWARN("Failed to repost %s buffer\n", sv->sv_name);
1383 rpc->srpc_reqstbuf = NULL;
1384 }
1385
1386 spin_unlock(&scd->scd_lock);
1387
1388 ev->ev_fired = 0;
1389 ev->ev_data = rpc;
1390 ev->ev_type = SRPC_REPLY_SENT;
1391
1392 msg->msg_magic = SRPC_MSG_MAGIC;
1393 msg->msg_version = SRPC_MSG_VERSION;
1394 msg->msg_type = srpc_service2reply(sv->sv_id);
1395
1396 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, rpyid, msg,
1397 sizeof(*msg), LNET_MD_OP_PUT,
1398 rpc->srpc_peer, rpc->srpc_self,
1399 &rpc->srpc_replymdh, ev);
1400 if (rc)
1401 ev->ev_fired = 1; /* no more event expected */
1402 return rc;
1403 }
1404
1405 /* when in kernel always called with LNET_LOCK() held, and in thread context */
1406 static void
srpc_lnet_ev_handler(lnet_event_t * ev)1407 srpc_lnet_ev_handler(lnet_event_t *ev)
1408 {
1409 struct srpc_service_cd *scd;
1410 struct srpc_event *rpcev = ev->md.user_ptr;
1411 struct srpc_client_rpc *crpc;
1412 struct srpc_server_rpc *srpc;
1413 struct srpc_buffer *buffer;
1414 struct srpc_service *sv;
1415 struct srpc_msg *msg;
1416 enum srpc_msg_type type;
1417
1418 LASSERT(!in_interrupt());
1419
1420 if (ev->status) {
1421 __u32 errors;
1422
1423 spin_lock(&srpc_data.rpc_glock);
1424 if (ev->status != -ECANCELED) /* cancellation is not error */
1425 srpc_data.rpc_counters.errors++;
1426 errors = srpc_data.rpc_counters.errors;
1427 spin_unlock(&srpc_data.rpc_glock);
1428
1429 CNETERR("LNet event status %d type %d, RPC errors %u\n",
1430 ev->status, ev->type, errors);
1431 }
1432
1433 rpcev->ev_lnet = ev->type;
1434
1435 switch (rpcev->ev_type) {
1436 default:
1437 CERROR("Unknown event: status %d, type %d, lnet %d\n",
1438 rpcev->ev_status, rpcev->ev_type, rpcev->ev_lnet);
1439 LBUG();
1440 case SRPC_REQUEST_SENT:
1441 if (!ev->status && ev->type != LNET_EVENT_UNLINK) {
1442 spin_lock(&srpc_data.rpc_glock);
1443 srpc_data.rpc_counters.rpcs_sent++;
1444 spin_unlock(&srpc_data.rpc_glock);
1445 }
1446 case SRPC_REPLY_RCVD:
1447 case SRPC_BULK_REQ_RCVD:
1448 crpc = rpcev->ev_data;
1449
1450 if (rpcev != &crpc->crpc_reqstev &&
1451 rpcev != &crpc->crpc_replyev &&
1452 rpcev != &crpc->crpc_bulkev) {
1453 CERROR("rpcev %p, crpc %p, reqstev %p, replyev %p, bulkev %p\n",
1454 rpcev, crpc, &crpc->crpc_reqstev,
1455 &crpc->crpc_replyev, &crpc->crpc_bulkev);
1456 CERROR("Bad event: status %d, type %d, lnet %d\n",
1457 rpcev->ev_status, rpcev->ev_type, rpcev->ev_lnet);
1458 LBUG();
1459 }
1460
1461 spin_lock(&crpc->crpc_lock);
1462
1463 LASSERT(!rpcev->ev_fired);
1464 rpcev->ev_fired = 1;
1465 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1466 -EINTR : ev->status;
1467 swi_schedule_workitem(&crpc->crpc_wi);
1468
1469 spin_unlock(&crpc->crpc_lock);
1470 break;
1471
1472 case SRPC_REQUEST_RCVD:
1473 scd = rpcev->ev_data;
1474 sv = scd->scd_svc;
1475
1476 LASSERT(rpcev == &scd->scd_ev);
1477
1478 spin_lock(&scd->scd_lock);
1479
1480 LASSERT(ev->unlinked);
1481 LASSERT(ev->type == LNET_EVENT_PUT ||
1482 ev->type == LNET_EVENT_UNLINK);
1483 LASSERT(ev->type != LNET_EVENT_UNLINK ||
1484 sv->sv_shuttingdown);
1485
1486 buffer = container_of(ev->md.start, struct srpc_buffer, buf_msg);
1487 buffer->buf_peer = ev->initiator;
1488 buffer->buf_self = ev->target.nid;
1489
1490 LASSERT(scd->scd_buf_nposted > 0);
1491 scd->scd_buf_nposted--;
1492
1493 if (sv->sv_shuttingdown) {
1494 /*
1495 * Leave buffer on scd->scd_buf_nposted since
1496 * srpc_finish_service needs to traverse it.
1497 */
1498 spin_unlock(&scd->scd_lock);
1499 break;
1500 }
1501
1502 if (scd->scd_buf_err_stamp &&
1503 scd->scd_buf_err_stamp < ktime_get_real_seconds()) {
1504 /* re-enable adding buffer */
1505 scd->scd_buf_err_stamp = 0;
1506 scd->scd_buf_err = 0;
1507 }
1508
1509 if (!scd->scd_buf_err && /* adding buffer is enabled */
1510 !scd->scd_buf_adjust &&
1511 scd->scd_buf_nposted < scd->scd_buf_low) {
1512 scd->scd_buf_adjust = max(scd->scd_buf_total / 2,
1513 SFW_TEST_WI_MIN);
1514 swi_schedule_workitem(&scd->scd_buf_wi);
1515 }
1516
1517 list_del(&buffer->buf_list); /* from scd->scd_buf_posted */
1518 msg = &buffer->buf_msg;
1519 type = srpc_service2request(sv->sv_id);
1520
1521 if (ev->status || ev->mlength != sizeof(*msg) ||
1522 (msg->msg_type != type &&
1523 msg->msg_type != __swab32(type)) ||
1524 (msg->msg_magic != SRPC_MSG_MAGIC &&
1525 msg->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1526 CERROR("Dropping RPC (%s) from %s: status %d mlength %d type %u magic %u.\n",
1527 sv->sv_name, libcfs_id2str(ev->initiator),
1528 ev->status, ev->mlength,
1529 msg->msg_type, msg->msg_magic);
1530
1531 /*
1532 * NB can't call srpc_service_recycle_buffer here since
1533 * it may call LNetM[DE]Attach. The invalid magic tells
1534 * srpc_handle_rpc to drop this RPC
1535 */
1536 msg->msg_magic = 0;
1537 }
1538
1539 if (!list_empty(&scd->scd_rpc_free)) {
1540 srpc = list_entry(scd->scd_rpc_free.next,
1541 struct srpc_server_rpc,
1542 srpc_list);
1543 list_del(&srpc->srpc_list);
1544
1545 srpc_init_server_rpc(srpc, scd, buffer);
1546 list_add_tail(&srpc->srpc_list,
1547 &scd->scd_rpc_active);
1548 swi_schedule_workitem(&srpc->srpc_wi);
1549 } else {
1550 list_add_tail(&buffer->buf_list,
1551 &scd->scd_buf_blocked);
1552 }
1553
1554 spin_unlock(&scd->scd_lock);
1555
1556 spin_lock(&srpc_data.rpc_glock);
1557 srpc_data.rpc_counters.rpcs_rcvd++;
1558 spin_unlock(&srpc_data.rpc_glock);
1559 break;
1560
1561 case SRPC_BULK_GET_RPLD:
1562 LASSERT(ev->type == LNET_EVENT_SEND ||
1563 ev->type == LNET_EVENT_REPLY ||
1564 ev->type == LNET_EVENT_UNLINK);
1565
1566 if (!ev->unlinked)
1567 break; /* wait for final event */
1568
1569 case SRPC_BULK_PUT_SENT:
1570 if (!ev->status && ev->type != LNET_EVENT_UNLINK) {
1571 spin_lock(&srpc_data.rpc_glock);
1572
1573 if (rpcev->ev_type == SRPC_BULK_GET_RPLD)
1574 srpc_data.rpc_counters.bulk_get += ev->mlength;
1575 else
1576 srpc_data.rpc_counters.bulk_put += ev->mlength;
1577
1578 spin_unlock(&srpc_data.rpc_glock);
1579 }
1580 case SRPC_REPLY_SENT:
1581 srpc = rpcev->ev_data;
1582 scd = srpc->srpc_scd;
1583
1584 LASSERT(rpcev == &srpc->srpc_ev);
1585
1586 spin_lock(&scd->scd_lock);
1587
1588 rpcev->ev_fired = 1;
1589 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1590 -EINTR : ev->status;
1591 swi_schedule_workitem(&srpc->srpc_wi);
1592
1593 spin_unlock(&scd->scd_lock);
1594 break;
1595 }
1596 }
1597
1598 int
srpc_startup(void)1599 srpc_startup(void)
1600 {
1601 int rc;
1602
1603 memset(&srpc_data, 0, sizeof(struct smoketest_rpc));
1604 spin_lock_init(&srpc_data.rpc_glock);
1605
1606 /* 1 second pause to avoid timestamp reuse */
1607 set_current_state(TASK_UNINTERRUPTIBLE);
1608 schedule_timeout(cfs_time_seconds(1));
1609 srpc_data.rpc_matchbits = ((__u64)ktime_get_real_seconds()) << 48;
1610
1611 srpc_data.rpc_state = SRPC_STATE_NONE;
1612
1613 rc = LNetNIInit(LNET_PID_LUSTRE);
1614 if (rc < 0) {
1615 CERROR("LNetNIInit() has failed: %d\n", rc);
1616 return rc;
1617 }
1618
1619 srpc_data.rpc_state = SRPC_STATE_NI_INIT;
1620
1621 LNetInvalidateHandle(&srpc_data.rpc_lnet_eq);
1622 rc = LNetEQAlloc(0, srpc_lnet_ev_handler, &srpc_data.rpc_lnet_eq);
1623 if (rc) {
1624 CERROR("LNetEQAlloc() has failed: %d\n", rc);
1625 goto bail;
1626 }
1627
1628 rc = LNetSetLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1629 LASSERT(!rc);
1630 rc = LNetSetLazyPortal(SRPC_REQUEST_PORTAL);
1631 LASSERT(!rc);
1632
1633 srpc_data.rpc_state = SRPC_STATE_EQ_INIT;
1634
1635 rc = stt_startup();
1636
1637 bail:
1638 if (rc)
1639 srpc_shutdown();
1640 else
1641 srpc_data.rpc_state = SRPC_STATE_RUNNING;
1642
1643 return rc;
1644 }
1645
1646 void
srpc_shutdown(void)1647 srpc_shutdown(void)
1648 {
1649 int i;
1650 int rc;
1651 int state;
1652
1653 state = srpc_data.rpc_state;
1654 srpc_data.rpc_state = SRPC_STATE_STOPPING;
1655
1656 switch (state) {
1657 default:
1658 LBUG();
1659 case SRPC_STATE_RUNNING:
1660 spin_lock(&srpc_data.rpc_glock);
1661
1662 for (i = 0; i <= SRPC_SERVICE_MAX_ID; i++) {
1663 struct srpc_service *sv = srpc_data.rpc_services[i];
1664
1665 LASSERTF(!sv, "service not empty: id %d, name %s\n",
1666 i, sv->sv_name);
1667 }
1668
1669 spin_unlock(&srpc_data.rpc_glock);
1670
1671 stt_shutdown();
1672
1673 case SRPC_STATE_EQ_INIT:
1674 rc = LNetClearLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1675 rc = LNetClearLazyPortal(SRPC_REQUEST_PORTAL);
1676 LASSERT(!rc);
1677 rc = LNetEQFree(srpc_data.rpc_lnet_eq);
1678 LASSERT(!rc); /* the EQ should have no user by now */
1679
1680 case SRPC_STATE_NI_INIT:
1681 LNetNIFini();
1682 }
1683 }
1684