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 struct lnet_handle_eq rpc_lnet_eq; /* _the_ LNet event queue */
57 enum srpc_state rpc_state;
58 struct srpc_counters 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(struct srpc_counters * cnt)72 void srpc_get_counters(struct srpc_counters *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 struct srpc_counters * cnt)79 void srpc_set_counters(const struct srpc_counters *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 off,int nob)87 srpc_add_bulk_page(struct srpc_bulk *bk, struct page *pg, int i, int off,
88 int nob)
89 {
90 LASSERT(off < PAGE_SIZE);
91 LASSERT(nob > 0 && nob <= PAGE_SIZE);
92
93 bk->bk_iovs[i].bv_offset = off;
94 bk->bk_iovs[i].bv_page = pg;
95 bk->bk_iovs[i].bv_len = nob;
96 return nob;
97 }
98
99 void
srpc_free_bulk(struct srpc_bulk * bk)100 srpc_free_bulk(struct srpc_bulk *bk)
101 {
102 int i;
103 struct page *pg;
104
105 LASSERT(bk);
106
107 for (i = 0; i < bk->bk_niov; i++) {
108 pg = bk->bk_iovs[i].bv_page;
109 if (!pg)
110 break;
111
112 __free_page(pg);
113 }
114
115 LIBCFS_FREE(bk, offsetof(struct srpc_bulk, bk_iovs[bk->bk_niov]));
116 }
117
118 struct srpc_bulk *
srpc_alloc_bulk(int cpt,unsigned int bulk_off,unsigned int bulk_npg,unsigned int bulk_len,int sink)119 srpc_alloc_bulk(int cpt, unsigned int bulk_off, unsigned int bulk_npg,
120 unsigned int 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 = min_t(unsigned int, bulk_off + bulk_len, PAGE_SIZE) -
152 bulk_off;
153 srpc_add_bulk_page(bk, pg, i, bulk_off, nob);
154 bulk_len -= nob;
155 bulk_off = 0;
156 }
157
158 return bk;
159 }
160
161 static inline __u64
srpc_next_id(void)162 srpc_next_id(void)
163 {
164 __u64 id;
165
166 spin_lock(&srpc_data.rpc_glock);
167 id = srpc_data.rpc_matchbits++;
168 spin_unlock(&srpc_data.rpc_glock);
169 return id;
170 }
171
172 static void
srpc_init_server_rpc(struct srpc_server_rpc * rpc,struct srpc_service_cd * scd,struct srpc_buffer * buffer)173 srpc_init_server_rpc(struct srpc_server_rpc *rpc,
174 struct srpc_service_cd *scd,
175 struct srpc_buffer *buffer)
176 {
177 memset(rpc, 0, sizeof(*rpc));
178 swi_init_workitem(&rpc->srpc_wi, rpc, srpc_handle_rpc,
179 srpc_serv_is_framework(scd->scd_svc) ?
180 lst_sched_serial : lst_sched_test[scd->scd_cpt]);
181
182 rpc->srpc_ev.ev_fired = 1; /* no event expected now */
183
184 rpc->srpc_scd = scd;
185 rpc->srpc_reqstbuf = buffer;
186 rpc->srpc_peer = buffer->buf_peer;
187 rpc->srpc_self = buffer->buf_self;
188 LNetInvalidateMDHandle(&rpc->srpc_replymdh);
189 }
190
191 static void
srpc_service_fini(struct srpc_service * svc)192 srpc_service_fini(struct srpc_service *svc)
193 {
194 struct srpc_service_cd *scd;
195 struct srpc_server_rpc *rpc;
196 struct srpc_buffer *buf;
197 struct list_head *q;
198 int i;
199
200 if (!svc->sv_cpt_data)
201 return;
202
203 cfs_percpt_for_each(scd, i, svc->sv_cpt_data) {
204 while (1) {
205 if (!list_empty(&scd->scd_buf_posted))
206 q = &scd->scd_buf_posted;
207 else if (!list_empty(&scd->scd_buf_blocked))
208 q = &scd->scd_buf_blocked;
209 else
210 break;
211
212 while (!list_empty(q)) {
213 buf = list_entry(q->next, struct srpc_buffer,
214 buf_list);
215 list_del(&buf->buf_list);
216 LIBCFS_FREE(buf, sizeof(*buf));
217 }
218 }
219
220 LASSERT(list_empty(&scd->scd_rpc_active));
221
222 while (!list_empty(&scd->scd_rpc_free)) {
223 rpc = list_entry(scd->scd_rpc_free.next,
224 struct srpc_server_rpc,
225 srpc_list);
226 list_del(&rpc->srpc_list);
227 LIBCFS_FREE(rpc, sizeof(*rpc));
228 }
229 }
230
231 cfs_percpt_free(svc->sv_cpt_data);
232 svc->sv_cpt_data = NULL;
233 }
234
235 static int
srpc_service_nrpcs(struct srpc_service * svc)236 srpc_service_nrpcs(struct srpc_service *svc)
237 {
238 int nrpcs = svc->sv_wi_total / svc->sv_ncpts;
239
240 return srpc_serv_is_framework(svc) ?
241 max(nrpcs, SFW_FRWK_WI_MIN) : max(nrpcs, SFW_TEST_WI_MIN);
242 }
243
244 int srpc_add_buffer(struct swi_workitem *wi);
245
246 static int
srpc_service_init(struct srpc_service * svc)247 srpc_service_init(struct srpc_service *svc)
248 {
249 struct srpc_service_cd *scd;
250 struct srpc_server_rpc *rpc;
251 int nrpcs;
252 int i;
253 int j;
254
255 svc->sv_shuttingdown = 0;
256
257 svc->sv_cpt_data = cfs_percpt_alloc(lnet_cpt_table(),
258 sizeof(**svc->sv_cpt_data));
259 if (!svc->sv_cpt_data)
260 return -ENOMEM;
261
262 svc->sv_ncpts = srpc_serv_is_framework(svc) ?
263 1 : cfs_cpt_number(lnet_cpt_table());
264 nrpcs = srpc_service_nrpcs(svc);
265
266 cfs_percpt_for_each(scd, i, svc->sv_cpt_data) {
267 scd->scd_cpt = i;
268 scd->scd_svc = svc;
269 spin_lock_init(&scd->scd_lock);
270 INIT_LIST_HEAD(&scd->scd_rpc_free);
271 INIT_LIST_HEAD(&scd->scd_rpc_active);
272 INIT_LIST_HEAD(&scd->scd_buf_posted);
273 INIT_LIST_HEAD(&scd->scd_buf_blocked);
274
275 scd->scd_ev.ev_data = scd;
276 scd->scd_ev.ev_type = SRPC_REQUEST_RCVD;
277
278 /*
279 * NB: don't use lst_sched_serial for adding buffer,
280 * see details in srpc_service_add_buffers()
281 */
282 swi_init_workitem(&scd->scd_buf_wi, scd,
283 srpc_add_buffer, lst_sched_test[i]);
284
285 if (i && srpc_serv_is_framework(svc)) {
286 /*
287 * NB: framework service only needs srpc_service_cd for
288 * one partition, but we allocate for all to make
289 * it easier to implement, it will waste a little
290 * memory but nobody should care about this
291 */
292 continue;
293 }
294
295 for (j = 0; j < nrpcs; j++) {
296 LIBCFS_CPT_ALLOC(rpc, lnet_cpt_table(),
297 i, sizeof(*rpc));
298 if (!rpc) {
299 srpc_service_fini(svc);
300 return -ENOMEM;
301 }
302 list_add(&rpc->srpc_list, &scd->scd_rpc_free);
303 }
304 }
305
306 return 0;
307 }
308
309 int
srpc_add_service(struct srpc_service * sv)310 srpc_add_service(struct srpc_service *sv)
311 {
312 int id = sv->sv_id;
313
314 LASSERT(0 <= id && id <= SRPC_SERVICE_MAX_ID);
315
316 if (srpc_service_init(sv))
317 return -ENOMEM;
318
319 spin_lock(&srpc_data.rpc_glock);
320
321 LASSERT(srpc_data.rpc_state == SRPC_STATE_RUNNING);
322
323 if (srpc_data.rpc_services[id]) {
324 spin_unlock(&srpc_data.rpc_glock);
325 goto failed;
326 }
327
328 srpc_data.rpc_services[id] = sv;
329 spin_unlock(&srpc_data.rpc_glock);
330
331 CDEBUG(D_NET, "Adding service: id %d, name %s\n", id, sv->sv_name);
332 return 0;
333
334 failed:
335 srpc_service_fini(sv);
336 return -EBUSY;
337 }
338
339 int
srpc_remove_service(struct srpc_service * sv)340 srpc_remove_service(struct srpc_service *sv)
341 {
342 int id = sv->sv_id;
343
344 spin_lock(&srpc_data.rpc_glock);
345
346 if (srpc_data.rpc_services[id] != sv) {
347 spin_unlock(&srpc_data.rpc_glock);
348 return -ENOENT;
349 }
350
351 srpc_data.rpc_services[id] = NULL;
352 spin_unlock(&srpc_data.rpc_glock);
353 return 0;
354 }
355
356 static int
srpc_post_passive_rdma(int portal,int local,__u64 matchbits,void * buf,int len,int options,struct lnet_process_id peer,struct lnet_handle_md * mdh,struct srpc_event * ev)357 srpc_post_passive_rdma(int portal, int local, __u64 matchbits, void *buf,
358 int len, int options, struct lnet_process_id peer,
359 struct lnet_handle_md *mdh, struct srpc_event *ev)
360 {
361 int rc;
362 struct lnet_md md;
363 struct lnet_handle_me meh;
364
365 rc = LNetMEAttach(portal, peer, matchbits, 0, LNET_UNLINK,
366 local ? LNET_INS_LOCAL : LNET_INS_AFTER, &meh);
367 if (rc) {
368 CERROR("LNetMEAttach failed: %d\n", rc);
369 LASSERT(rc == -ENOMEM);
370 return -ENOMEM;
371 }
372
373 md.threshold = 1;
374 md.user_ptr = ev;
375 md.start = buf;
376 md.length = len;
377 md.options = options;
378 md.eq_handle = srpc_data.rpc_lnet_eq;
379
380 rc = LNetMDAttach(meh, md, LNET_UNLINK, mdh);
381 if (rc) {
382 CERROR("LNetMDAttach failed: %d\n", rc);
383 LASSERT(rc == -ENOMEM);
384
385 rc = LNetMEUnlink(meh);
386 LASSERT(!rc);
387 return -ENOMEM;
388 }
389
390 CDEBUG(D_NET, "Posted passive RDMA: peer %s, portal %d, matchbits %#llx\n",
391 libcfs_id2str(peer), portal, matchbits);
392 return 0;
393 }
394
395 static int
srpc_post_active_rdma(int portal,__u64 matchbits,void * buf,int len,int options,struct lnet_process_id peer,lnet_nid_t self,struct lnet_handle_md * mdh,struct srpc_event * ev)396 srpc_post_active_rdma(int portal, __u64 matchbits, void *buf, int len,
397 int options, struct lnet_process_id peer,
398 lnet_nid_t self, struct lnet_handle_md *mdh,
399 struct srpc_event *ev)
400 {
401 int rc;
402 struct lnet_md md;
403
404 md.user_ptr = ev;
405 md.start = buf;
406 md.length = len;
407 md.eq_handle = srpc_data.rpc_lnet_eq;
408 md.threshold = options & LNET_MD_OP_GET ? 2 : 1;
409 md.options = options & ~(LNET_MD_OP_PUT | LNET_MD_OP_GET);
410
411 rc = LNetMDBind(md, LNET_UNLINK, mdh);
412 if (rc) {
413 CERROR("LNetMDBind failed: %d\n", rc);
414 LASSERT(rc == -ENOMEM);
415 return -ENOMEM;
416 }
417
418 /*
419 * this is kind of an abuse of the LNET_MD_OP_{PUT,GET} options.
420 * they're only meaningful for MDs attached to an ME (i.e. passive
421 * buffers...
422 */
423 if (options & LNET_MD_OP_PUT) {
424 rc = LNetPut(self, *mdh, LNET_NOACK_REQ, peer,
425 portal, matchbits, 0, 0);
426 } else {
427 LASSERT(options & LNET_MD_OP_GET);
428
429 rc = LNetGet(self, *mdh, peer, portal, matchbits, 0);
430 }
431
432 if (rc) {
433 CERROR("LNet%s(%s, %d, %lld) failed: %d\n",
434 options & LNET_MD_OP_PUT ? "Put" : "Get",
435 libcfs_id2str(peer), portal, matchbits, rc);
436
437 /*
438 * The forthcoming unlink event will complete this operation
439 * with failure, so fall through and return success here.
440 */
441 rc = LNetMDUnlink(*mdh);
442 LASSERT(!rc);
443 } else {
444 CDEBUG(D_NET, "Posted active RDMA: peer %s, portal %u, matchbits %#llx\n",
445 libcfs_id2str(peer), portal, matchbits);
446 }
447 return 0;
448 }
449
450 static int
srpc_post_passive_rqtbuf(int service,int local,void * buf,int len,struct lnet_handle_md * mdh,struct srpc_event * ev)451 srpc_post_passive_rqtbuf(int service, int local, void *buf, int len,
452 struct lnet_handle_md *mdh, struct srpc_event *ev)
453 {
454 struct lnet_process_id any = { 0 };
455
456 any.nid = LNET_NID_ANY;
457 any.pid = LNET_PID_ANY;
458
459 return srpc_post_passive_rdma(srpc_serv_portal(service),
460 local, service, buf, len,
461 LNET_MD_OP_PUT, any, mdh, ev);
462 }
463
464 static int
srpc_service_post_buffer(struct srpc_service_cd * scd,struct srpc_buffer * buf)465 srpc_service_post_buffer(struct srpc_service_cd *scd, struct srpc_buffer *buf)
466 __must_hold(&scd->scd_lock)
467 {
468 struct srpc_service *sv = scd->scd_svc;
469 struct srpc_msg *msg = &buf->buf_msg;
470 int rc;
471
472 LNetInvalidateMDHandle(&buf->buf_mdh);
473 list_add(&buf->buf_list, &scd->scd_buf_posted);
474 scd->scd_buf_nposted++;
475 spin_unlock(&scd->scd_lock);
476
477 rc = srpc_post_passive_rqtbuf(sv->sv_id,
478 !srpc_serv_is_framework(sv),
479 msg, sizeof(*msg), &buf->buf_mdh,
480 &scd->scd_ev);
481
482 /*
483 * At this point, a RPC (new or delayed) may have arrived in
484 * msg and its event handler has been called. So we must add
485 * buf to scd_buf_posted _before_ dropping scd_lock
486 */
487 spin_lock(&scd->scd_lock);
488
489 if (!rc) {
490 if (!sv->sv_shuttingdown)
491 return 0;
492
493 spin_unlock(&scd->scd_lock);
494 /*
495 * srpc_shutdown_service might have tried to unlink me
496 * when my buf_mdh was still invalid
497 */
498 LNetMDUnlink(buf->buf_mdh);
499 spin_lock(&scd->scd_lock);
500 return 0;
501 }
502
503 scd->scd_buf_nposted--;
504 if (sv->sv_shuttingdown)
505 return rc; /* don't allow to change scd_buf_posted */
506
507 list_del(&buf->buf_list);
508 spin_unlock(&scd->scd_lock);
509
510 LIBCFS_FREE(buf, sizeof(*buf));
511
512 spin_lock(&scd->scd_lock);
513 return rc;
514 }
515
516 int
srpc_add_buffer(struct swi_workitem * wi)517 srpc_add_buffer(struct swi_workitem *wi)
518 {
519 struct srpc_service_cd *scd = wi->swi_workitem.wi_data;
520 struct srpc_buffer *buf;
521 int rc = 0;
522
523 /*
524 * it's called by workitem scheduler threads, these threads
525 * should have been set CPT affinity, so buffers will be posted
526 * on CPT local list of Portal
527 */
528 spin_lock(&scd->scd_lock);
529
530 while (scd->scd_buf_adjust > 0 &&
531 !scd->scd_svc->sv_shuttingdown) {
532 scd->scd_buf_adjust--; /* consume it */
533 scd->scd_buf_posting++;
534
535 spin_unlock(&scd->scd_lock);
536
537 LIBCFS_ALLOC(buf, sizeof(*buf));
538 if (!buf) {
539 CERROR("Failed to add new buf to service: %s\n",
540 scd->scd_svc->sv_name);
541 spin_lock(&scd->scd_lock);
542 rc = -ENOMEM;
543 break;
544 }
545
546 spin_lock(&scd->scd_lock);
547 if (scd->scd_svc->sv_shuttingdown) {
548 spin_unlock(&scd->scd_lock);
549 LIBCFS_FREE(buf, sizeof(*buf));
550
551 spin_lock(&scd->scd_lock);
552 rc = -ESHUTDOWN;
553 break;
554 }
555
556 rc = srpc_service_post_buffer(scd, buf);
557 if (rc)
558 break; /* buf has been freed inside */
559
560 LASSERT(scd->scd_buf_posting > 0);
561 scd->scd_buf_posting--;
562 scd->scd_buf_total++;
563 scd->scd_buf_low = max(2, scd->scd_buf_total / 4);
564 }
565
566 if (rc) {
567 scd->scd_buf_err_stamp = ktime_get_real_seconds();
568 scd->scd_buf_err = rc;
569
570 LASSERT(scd->scd_buf_posting > 0);
571 scd->scd_buf_posting--;
572 }
573
574 spin_unlock(&scd->scd_lock);
575 return 0;
576 }
577
578 int
srpc_service_add_buffers(struct srpc_service * sv,int nbuffer)579 srpc_service_add_buffers(struct srpc_service *sv, int nbuffer)
580 {
581 struct srpc_service_cd *scd;
582 int rc = 0;
583 int i;
584
585 LASSERTF(nbuffer > 0, "nbuffer must be positive: %d\n", nbuffer);
586
587 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
588 spin_lock(&scd->scd_lock);
589
590 scd->scd_buf_err = 0;
591 scd->scd_buf_err_stamp = 0;
592 scd->scd_buf_posting = 0;
593 scd->scd_buf_adjust = nbuffer;
594 /* start to post buffers */
595 swi_schedule_workitem(&scd->scd_buf_wi);
596 spin_unlock(&scd->scd_lock);
597
598 /* framework service only post buffer for one partition */
599 if (srpc_serv_is_framework(sv))
600 break;
601 }
602
603 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
604 spin_lock(&scd->scd_lock);
605 /*
606 * NB: srpc_service_add_buffers() can be called inside
607 * thread context of lst_sched_serial, and we don't normally
608 * allow to sleep inside thread context of WI scheduler
609 * because it will block current scheduler thread from doing
610 * anything else, even worse, it could deadlock if it's
611 * waiting on result from another WI of the same scheduler.
612 * However, it's safe at here because scd_buf_wi is scheduled
613 * by thread in a different WI scheduler (lst_sched_test),
614 * so we don't have any risk of deadlock, though this could
615 * block all WIs pending on lst_sched_serial for a moment
616 * which is not good but not fatal.
617 */
618 lst_wait_until(scd->scd_buf_err ||
619 (!scd->scd_buf_adjust &&
620 !scd->scd_buf_posting),
621 scd->scd_lock, "waiting for adding buffer\n");
622
623 if (scd->scd_buf_err && !rc)
624 rc = scd->scd_buf_err;
625
626 spin_unlock(&scd->scd_lock);
627 }
628
629 return rc;
630 }
631
632 void
srpc_service_remove_buffers(struct srpc_service * sv,int nbuffer)633 srpc_service_remove_buffers(struct srpc_service *sv, int nbuffer)
634 {
635 struct srpc_service_cd *scd;
636 int num;
637 int i;
638
639 LASSERT(!sv->sv_shuttingdown);
640
641 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
642 spin_lock(&scd->scd_lock);
643
644 num = scd->scd_buf_total + scd->scd_buf_posting;
645 scd->scd_buf_adjust -= min(nbuffer, num);
646
647 spin_unlock(&scd->scd_lock);
648 }
649 }
650
651 /* returns 1 if sv has finished, otherwise 0 */
652 int
srpc_finish_service(struct srpc_service * sv)653 srpc_finish_service(struct srpc_service *sv)
654 {
655 struct srpc_service_cd *scd;
656 struct srpc_server_rpc *rpc;
657 int i;
658
659 LASSERT(sv->sv_shuttingdown); /* srpc_shutdown_service called */
660
661 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
662 spin_lock(&scd->scd_lock);
663 if (!swi_deschedule_workitem(&scd->scd_buf_wi)) {
664 spin_unlock(&scd->scd_lock);
665 return 0;
666 }
667
668 if (scd->scd_buf_nposted > 0) {
669 CDEBUG(D_NET, "waiting for %d posted buffers to unlink\n",
670 scd->scd_buf_nposted);
671 spin_unlock(&scd->scd_lock);
672 return 0;
673 }
674
675 if (list_empty(&scd->scd_rpc_active)) {
676 spin_unlock(&scd->scd_lock);
677 continue;
678 }
679
680 rpc = list_entry(scd->scd_rpc_active.next,
681 struct srpc_server_rpc, srpc_list);
682 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",
683 rpc, sv->sv_name, libcfs_id2str(rpc->srpc_peer),
684 swi_state2str(rpc->srpc_wi.swi_state),
685 rpc->srpc_wi.swi_workitem.wi_scheduled,
686 rpc->srpc_wi.swi_workitem.wi_running,
687 rpc->srpc_ev.ev_fired, rpc->srpc_ev.ev_type,
688 rpc->srpc_ev.ev_status, rpc->srpc_ev.ev_lnet);
689 spin_unlock(&scd->scd_lock);
690 return 0;
691 }
692
693 /* no lock needed from now on */
694 srpc_service_fini(sv);
695 return 1;
696 }
697
698 /* called with sv->sv_lock held */
699 static void
srpc_service_recycle_buffer(struct srpc_service_cd * scd,struct srpc_buffer * buf)700 srpc_service_recycle_buffer(struct srpc_service_cd *scd,
701 struct srpc_buffer *buf)
702 __must_hold(&scd->scd_lock)
703 {
704 if (!scd->scd_svc->sv_shuttingdown && scd->scd_buf_adjust >= 0) {
705 if (srpc_service_post_buffer(scd, buf)) {
706 CWARN("Failed to post %s buffer\n",
707 scd->scd_svc->sv_name);
708 }
709 return;
710 }
711
712 /* service is shutting down, or we want to recycle some buffers */
713 scd->scd_buf_total--;
714
715 if (scd->scd_buf_adjust < 0) {
716 scd->scd_buf_adjust++;
717 if (scd->scd_buf_adjust < 0 &&
718 !scd->scd_buf_total && !scd->scd_buf_posting) {
719 CDEBUG(D_INFO,
720 "Try to recycle %d buffers but nothing left\n",
721 scd->scd_buf_adjust);
722 scd->scd_buf_adjust = 0;
723 }
724 }
725
726 spin_unlock(&scd->scd_lock);
727 LIBCFS_FREE(buf, sizeof(*buf));
728 spin_lock(&scd->scd_lock);
729 }
730
731 void
srpc_abort_service(struct srpc_service * sv)732 srpc_abort_service(struct srpc_service *sv)
733 {
734 struct srpc_service_cd *scd;
735 struct srpc_server_rpc *rpc;
736 int i;
737
738 CDEBUG(D_NET, "Aborting service: id %d, name %s\n",
739 sv->sv_id, sv->sv_name);
740
741 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
742 spin_lock(&scd->scd_lock);
743
744 /*
745 * schedule in-flight RPCs to notice the abort, NB:
746 * racing with incoming RPCs; complete fix should make test
747 * RPCs carry session ID in its headers
748 */
749 list_for_each_entry(rpc, &scd->scd_rpc_active, srpc_list) {
750 rpc->srpc_aborted = 1;
751 swi_schedule_workitem(&rpc->srpc_wi);
752 }
753
754 spin_unlock(&scd->scd_lock);
755 }
756 }
757
758 void
srpc_shutdown_service(struct srpc_service * sv)759 srpc_shutdown_service(struct srpc_service *sv)
760 {
761 struct srpc_service_cd *scd;
762 struct srpc_server_rpc *rpc;
763 struct srpc_buffer *buf;
764 int i;
765
766 CDEBUG(D_NET, "Shutting down service: id %d, name %s\n",
767 sv->sv_id, sv->sv_name);
768
769 cfs_percpt_for_each(scd, i, sv->sv_cpt_data)
770 spin_lock(&scd->scd_lock);
771
772 sv->sv_shuttingdown = 1; /* i.e. no new active RPC */
773
774 cfs_percpt_for_each(scd, i, sv->sv_cpt_data)
775 spin_unlock(&scd->scd_lock);
776
777 cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
778 spin_lock(&scd->scd_lock);
779
780 /* schedule in-flight RPCs to notice the shutdown */
781 list_for_each_entry(rpc, &scd->scd_rpc_active, srpc_list)
782 swi_schedule_workitem(&rpc->srpc_wi);
783
784 spin_unlock(&scd->scd_lock);
785
786 /*
787 * OK to traverse scd_buf_posted without lock, since no one
788 * touches scd_buf_posted now
789 */
790 list_for_each_entry(buf, &scd->scd_buf_posted, buf_list)
791 LNetMDUnlink(buf->buf_mdh);
792 }
793 }
794
795 static int
srpc_send_request(struct srpc_client_rpc * rpc)796 srpc_send_request(struct srpc_client_rpc *rpc)
797 {
798 struct srpc_event *ev = &rpc->crpc_reqstev;
799 int rc;
800
801 ev->ev_fired = 0;
802 ev->ev_data = rpc;
803 ev->ev_type = SRPC_REQUEST_SENT;
804
805 rc = srpc_post_active_rdma(srpc_serv_portal(rpc->crpc_service),
806 rpc->crpc_service, &rpc->crpc_reqstmsg,
807 sizeof(struct srpc_msg), LNET_MD_OP_PUT,
808 rpc->crpc_dest, LNET_NID_ANY,
809 &rpc->crpc_reqstmdh, ev);
810 if (rc) {
811 LASSERT(rc == -ENOMEM);
812 ev->ev_fired = 1; /* no more event expected */
813 }
814 return rc;
815 }
816
817 static int
srpc_prepare_reply(struct srpc_client_rpc * rpc)818 srpc_prepare_reply(struct srpc_client_rpc *rpc)
819 {
820 struct srpc_event *ev = &rpc->crpc_replyev;
821 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.rpyid;
822 int rc;
823
824 ev->ev_fired = 0;
825 ev->ev_data = rpc;
826 ev->ev_type = SRPC_REPLY_RCVD;
827
828 *id = srpc_next_id();
829
830 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, 0, *id,
831 &rpc->crpc_replymsg,
832 sizeof(struct srpc_msg),
833 LNET_MD_OP_PUT, rpc->crpc_dest,
834 &rpc->crpc_replymdh, ev);
835 if (rc) {
836 LASSERT(rc == -ENOMEM);
837 ev->ev_fired = 1; /* no more event expected */
838 }
839 return rc;
840 }
841
842 static int
srpc_prepare_bulk(struct srpc_client_rpc * rpc)843 srpc_prepare_bulk(struct srpc_client_rpc *rpc)
844 {
845 struct srpc_bulk *bk = &rpc->crpc_bulk;
846 struct srpc_event *ev = &rpc->crpc_bulkev;
847 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.bulkid;
848 int rc;
849 int opt;
850
851 LASSERT(bk->bk_niov <= LNET_MAX_IOV);
852
853 if (!bk->bk_niov)
854 return 0; /* nothing to do */
855
856 opt = bk->bk_sink ? LNET_MD_OP_PUT : LNET_MD_OP_GET;
857 opt |= LNET_MD_KIOV;
858
859 ev->ev_fired = 0;
860 ev->ev_data = rpc;
861 ev->ev_type = SRPC_BULK_REQ_RCVD;
862
863 *id = srpc_next_id();
864
865 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, 0, *id,
866 &bk->bk_iovs[0], bk->bk_niov, opt,
867 rpc->crpc_dest, &bk->bk_mdh, ev);
868 if (rc) {
869 LASSERT(rc == -ENOMEM);
870 ev->ev_fired = 1; /* no more event expected */
871 }
872 return rc;
873 }
874
875 static int
srpc_do_bulk(struct srpc_server_rpc * rpc)876 srpc_do_bulk(struct srpc_server_rpc *rpc)
877 {
878 struct srpc_event *ev = &rpc->srpc_ev;
879 struct srpc_bulk *bk = rpc->srpc_bulk;
880 __u64 id = rpc->srpc_reqstbuf->buf_msg.msg_body.reqst.bulkid;
881 int rc;
882 int opt;
883
884 LASSERT(bk);
885
886 opt = bk->bk_sink ? LNET_MD_OP_GET : LNET_MD_OP_PUT;
887 opt |= LNET_MD_KIOV;
888
889 ev->ev_fired = 0;
890 ev->ev_data = rpc;
891 ev->ev_type = bk->bk_sink ? SRPC_BULK_GET_RPLD : SRPC_BULK_PUT_SENT;
892
893 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, id,
894 &bk->bk_iovs[0], bk->bk_niov, opt,
895 rpc->srpc_peer, rpc->srpc_self,
896 &bk->bk_mdh, ev);
897 if (rc)
898 ev->ev_fired = 1; /* no more event expected */
899 return rc;
900 }
901
902 /* only called from srpc_handle_rpc */
903 static void
srpc_server_rpc_done(struct srpc_server_rpc * rpc,int status)904 srpc_server_rpc_done(struct srpc_server_rpc *rpc, int status)
905 {
906 struct srpc_service_cd *scd = rpc->srpc_scd;
907 struct srpc_service *sv = scd->scd_svc;
908 struct srpc_buffer *buffer;
909
910 LASSERT(status || rpc->srpc_wi.swi_state == SWI_STATE_DONE);
911
912 rpc->srpc_status = status;
913
914 CDEBUG_LIMIT(!status ? D_NET : D_NETERROR,
915 "Server RPC %p done: service %s, peer %s, status %s:%d\n",
916 rpc, sv->sv_name, libcfs_id2str(rpc->srpc_peer),
917 swi_state2str(rpc->srpc_wi.swi_state), status);
918
919 if (status) {
920 spin_lock(&srpc_data.rpc_glock);
921 srpc_data.rpc_counters.rpcs_dropped++;
922 spin_unlock(&srpc_data.rpc_glock);
923 }
924
925 if (rpc->srpc_done)
926 (*rpc->srpc_done) (rpc);
927 LASSERT(!rpc->srpc_bulk);
928
929 spin_lock(&scd->scd_lock);
930
931 if (rpc->srpc_reqstbuf) {
932 /*
933 * NB might drop sv_lock in srpc_service_recycle_buffer, but
934 * sv won't go away for scd_rpc_active must not be empty
935 */
936 srpc_service_recycle_buffer(scd, rpc->srpc_reqstbuf);
937 rpc->srpc_reqstbuf = NULL;
938 }
939
940 list_del(&rpc->srpc_list); /* from scd->scd_rpc_active */
941
942 /*
943 * No one can schedule me now since:
944 * - I'm not on scd_rpc_active.
945 * - all LNet events have been fired.
946 * Cancel pending schedules and prevent future schedule attempts:
947 */
948 LASSERT(rpc->srpc_ev.ev_fired);
949 swi_exit_workitem(&rpc->srpc_wi);
950
951 if (!sv->sv_shuttingdown && !list_empty(&scd->scd_buf_blocked)) {
952 buffer = list_entry(scd->scd_buf_blocked.next,
953 struct srpc_buffer, buf_list);
954 list_del(&buffer->buf_list);
955
956 srpc_init_server_rpc(rpc, scd, buffer);
957 list_add_tail(&rpc->srpc_list, &scd->scd_rpc_active);
958 swi_schedule_workitem(&rpc->srpc_wi);
959 } else {
960 list_add(&rpc->srpc_list, &scd->scd_rpc_free);
961 }
962
963 spin_unlock(&scd->scd_lock);
964 }
965
966 /* handles an incoming RPC */
967 int
srpc_handle_rpc(struct swi_workitem * wi)968 srpc_handle_rpc(struct swi_workitem *wi)
969 {
970 struct srpc_server_rpc *rpc = wi->swi_workitem.wi_data;
971 struct srpc_service_cd *scd = rpc->srpc_scd;
972 struct srpc_service *sv = scd->scd_svc;
973 struct srpc_event *ev = &rpc->srpc_ev;
974 int rc = 0;
975
976 LASSERT(wi == &rpc->srpc_wi);
977
978 spin_lock(&scd->scd_lock);
979
980 if (sv->sv_shuttingdown || rpc->srpc_aborted) {
981 spin_unlock(&scd->scd_lock);
982
983 if (rpc->srpc_bulk)
984 LNetMDUnlink(rpc->srpc_bulk->bk_mdh);
985 LNetMDUnlink(rpc->srpc_replymdh);
986
987 if (ev->ev_fired) { /* no more event, OK to finish */
988 srpc_server_rpc_done(rpc, -ESHUTDOWN);
989 return 1;
990 }
991 return 0;
992 }
993
994 spin_unlock(&scd->scd_lock);
995
996 switch (wi->swi_state) {
997 default:
998 LBUG();
999 case SWI_STATE_NEWBORN: {
1000 struct srpc_msg *msg;
1001 struct srpc_generic_reply *reply;
1002
1003 msg = &rpc->srpc_reqstbuf->buf_msg;
1004 reply = &rpc->srpc_replymsg.msg_body.reply;
1005
1006 if (!msg->msg_magic) {
1007 /* moaned already in srpc_lnet_ev_handler */
1008 srpc_server_rpc_done(rpc, EBADMSG);
1009 return 1;
1010 }
1011
1012 srpc_unpack_msg_hdr(msg);
1013 if (msg->msg_version != SRPC_MSG_VERSION) {
1014 CWARN("Version mismatch: %u, %u expected, from %s\n",
1015 msg->msg_version, SRPC_MSG_VERSION,
1016 libcfs_id2str(rpc->srpc_peer));
1017 reply->status = EPROTO;
1018 /* drop through and send reply */
1019 } else {
1020 reply->status = 0;
1021 rc = (*sv->sv_handler)(rpc);
1022 LASSERT(!reply->status || !rpc->srpc_bulk);
1023 if (rc) {
1024 srpc_server_rpc_done(rpc, rc);
1025 return 1;
1026 }
1027 }
1028
1029 wi->swi_state = SWI_STATE_BULK_STARTED;
1030
1031 if (rpc->srpc_bulk) {
1032 rc = srpc_do_bulk(rpc);
1033 if (!rc)
1034 return 0; /* wait for bulk */
1035
1036 LASSERT(ev->ev_fired);
1037 ev->ev_status = rc;
1038 }
1039 }
1040 case SWI_STATE_BULK_STARTED:
1041 LASSERT(!rpc->srpc_bulk || ev->ev_fired);
1042
1043 if (rpc->srpc_bulk) {
1044 rc = ev->ev_status;
1045
1046 if (sv->sv_bulk_ready)
1047 rc = (*sv->sv_bulk_ready) (rpc, rc);
1048
1049 if (rc) {
1050 srpc_server_rpc_done(rpc, rc);
1051 return 1;
1052 }
1053 }
1054
1055 wi->swi_state = SWI_STATE_REPLY_SUBMITTED;
1056 rc = srpc_send_reply(rpc);
1057 if (!rc)
1058 return 0; /* wait for reply */
1059 srpc_server_rpc_done(rpc, rc);
1060 return 1;
1061
1062 case SWI_STATE_REPLY_SUBMITTED:
1063 if (!ev->ev_fired) {
1064 CERROR("RPC %p: bulk %p, service %d\n",
1065 rpc, rpc->srpc_bulk, sv->sv_id);
1066 CERROR("Event: status %d, type %d, lnet %d\n",
1067 ev->ev_status, ev->ev_type, ev->ev_lnet);
1068 LASSERT(ev->ev_fired);
1069 }
1070
1071 wi->swi_state = SWI_STATE_DONE;
1072 srpc_server_rpc_done(rpc, ev->ev_status);
1073 return 1;
1074 }
1075
1076 return 0;
1077 }
1078
1079 static void
srpc_client_rpc_expired(void * data)1080 srpc_client_rpc_expired(void *data)
1081 {
1082 struct srpc_client_rpc *rpc = data;
1083
1084 CWARN("Client RPC expired: service %d, peer %s, timeout %d.\n",
1085 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1086 rpc->crpc_timeout);
1087
1088 spin_lock(&rpc->crpc_lock);
1089
1090 rpc->crpc_timeout = 0;
1091 srpc_abort_rpc(rpc, -ETIMEDOUT);
1092
1093 spin_unlock(&rpc->crpc_lock);
1094
1095 spin_lock(&srpc_data.rpc_glock);
1096 srpc_data.rpc_counters.rpcs_expired++;
1097 spin_unlock(&srpc_data.rpc_glock);
1098 }
1099
1100 static void
srpc_add_client_rpc_timer(struct srpc_client_rpc * rpc)1101 srpc_add_client_rpc_timer(struct srpc_client_rpc *rpc)
1102 {
1103 struct stt_timer *timer = &rpc->crpc_timer;
1104
1105 if (!rpc->crpc_timeout)
1106 return;
1107
1108 INIT_LIST_HEAD(&timer->stt_list);
1109 timer->stt_data = rpc;
1110 timer->stt_func = srpc_client_rpc_expired;
1111 timer->stt_expires = ktime_get_real_seconds() + rpc->crpc_timeout;
1112 stt_add_timer(timer);
1113 }
1114
1115 /*
1116 * Called with rpc->crpc_lock held.
1117 *
1118 * Upon exit the RPC expiry timer is not queued and the handler is not
1119 * running on any CPU.
1120 */
1121 static void
srpc_del_client_rpc_timer(struct srpc_client_rpc * rpc)1122 srpc_del_client_rpc_timer(struct srpc_client_rpc *rpc)
1123 {
1124 /* timer not planted or already exploded */
1125 if (!rpc->crpc_timeout)
1126 return;
1127
1128 /* timer successfully defused */
1129 if (stt_del_timer(&rpc->crpc_timer))
1130 return;
1131
1132 /* timer detonated, wait for it to explode */
1133 while (rpc->crpc_timeout) {
1134 spin_unlock(&rpc->crpc_lock);
1135
1136 schedule();
1137
1138 spin_lock(&rpc->crpc_lock);
1139 }
1140 }
1141
1142 static void
srpc_client_rpc_done(struct srpc_client_rpc * rpc,int status)1143 srpc_client_rpc_done(struct srpc_client_rpc *rpc, int status)
1144 {
1145 struct swi_workitem *wi = &rpc->crpc_wi;
1146
1147 LASSERT(status || wi->swi_state == SWI_STATE_DONE);
1148
1149 spin_lock(&rpc->crpc_lock);
1150
1151 rpc->crpc_closed = 1;
1152 if (!rpc->crpc_status)
1153 rpc->crpc_status = status;
1154
1155 srpc_del_client_rpc_timer(rpc);
1156
1157 CDEBUG_LIMIT(!status ? D_NET : D_NETERROR,
1158 "Client RPC done: service %d, peer %s, status %s:%d:%d\n",
1159 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1160 swi_state2str(wi->swi_state), rpc->crpc_aborted, status);
1161
1162 /*
1163 * No one can schedule me now since:
1164 * - RPC timer has been defused.
1165 * - all LNet events have been fired.
1166 * - crpc_closed has been set, preventing srpc_abort_rpc from
1167 * scheduling me.
1168 * Cancel pending schedules and prevent future schedule attempts:
1169 */
1170 LASSERT(!srpc_event_pending(rpc));
1171 swi_exit_workitem(wi);
1172
1173 spin_unlock(&rpc->crpc_lock);
1174
1175 (*rpc->crpc_done)(rpc);
1176 }
1177
1178 /* sends an outgoing RPC */
1179 int
srpc_send_rpc(struct swi_workitem * wi)1180 srpc_send_rpc(struct swi_workitem *wi)
1181 {
1182 int rc = 0;
1183 struct srpc_client_rpc *rpc;
1184 struct srpc_msg *reply;
1185 int do_bulk;
1186
1187 LASSERT(wi);
1188
1189 rpc = wi->swi_workitem.wi_data;
1190
1191 LASSERT(rpc);
1192 LASSERT(wi == &rpc->crpc_wi);
1193
1194 reply = &rpc->crpc_replymsg;
1195 do_bulk = rpc->crpc_bulk.bk_niov > 0;
1196
1197 spin_lock(&rpc->crpc_lock);
1198
1199 if (rpc->crpc_aborted) {
1200 spin_unlock(&rpc->crpc_lock);
1201 goto abort;
1202 }
1203
1204 spin_unlock(&rpc->crpc_lock);
1205
1206 switch (wi->swi_state) {
1207 default:
1208 LBUG();
1209 case SWI_STATE_NEWBORN:
1210 LASSERT(!srpc_event_pending(rpc));
1211
1212 rc = srpc_prepare_reply(rpc);
1213 if (rc) {
1214 srpc_client_rpc_done(rpc, rc);
1215 return 1;
1216 }
1217
1218 rc = srpc_prepare_bulk(rpc);
1219 if (rc)
1220 break;
1221
1222 wi->swi_state = SWI_STATE_REQUEST_SUBMITTED;
1223 rc = srpc_send_request(rpc);
1224 break;
1225
1226 case SWI_STATE_REQUEST_SUBMITTED:
1227 /*
1228 * CAVEAT EMPTOR: rqtev, rpyev, and bulkev may come in any
1229 * order; however, they're processed in a strict order:
1230 * rqt, rpy, and bulk.
1231 */
1232 if (!rpc->crpc_reqstev.ev_fired)
1233 break;
1234
1235 rc = rpc->crpc_reqstev.ev_status;
1236 if (rc)
1237 break;
1238
1239 wi->swi_state = SWI_STATE_REQUEST_SENT;
1240 /* perhaps more events, fall thru */
1241 case SWI_STATE_REQUEST_SENT: {
1242 enum srpc_msg_type type = srpc_service2reply(rpc->crpc_service);
1243
1244 if (!rpc->crpc_replyev.ev_fired)
1245 break;
1246
1247 rc = rpc->crpc_replyev.ev_status;
1248 if (rc)
1249 break;
1250
1251 srpc_unpack_msg_hdr(reply);
1252 if (reply->msg_type != type ||
1253 (reply->msg_magic != SRPC_MSG_MAGIC &&
1254 reply->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1255 CWARN("Bad message from %s: type %u (%d expected), magic %u (%d expected).\n",
1256 libcfs_id2str(rpc->crpc_dest),
1257 reply->msg_type, type,
1258 reply->msg_magic, SRPC_MSG_MAGIC);
1259 rc = -EBADMSG;
1260 break;
1261 }
1262
1263 if (do_bulk && reply->msg_body.reply.status) {
1264 CWARN("Remote error %d at %s, unlink bulk buffer in case peer didn't initiate bulk transfer\n",
1265 reply->msg_body.reply.status,
1266 libcfs_id2str(rpc->crpc_dest));
1267 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1268 }
1269
1270 wi->swi_state = SWI_STATE_REPLY_RECEIVED;
1271 }
1272 case SWI_STATE_REPLY_RECEIVED:
1273 if (do_bulk && !rpc->crpc_bulkev.ev_fired)
1274 break;
1275
1276 rc = do_bulk ? rpc->crpc_bulkev.ev_status : 0;
1277
1278 /*
1279 * Bulk buffer was unlinked due to remote error. Clear error
1280 * since reply buffer still contains valid data.
1281 * NB rpc->crpc_done shouldn't look into bulk data in case of
1282 * remote error.
1283 */
1284 if (do_bulk && rpc->crpc_bulkev.ev_lnet == LNET_EVENT_UNLINK &&
1285 !rpc->crpc_status && reply->msg_body.reply.status)
1286 rc = 0;
1287
1288 wi->swi_state = SWI_STATE_DONE;
1289 srpc_client_rpc_done(rpc, rc);
1290 return 1;
1291 }
1292
1293 if (rc) {
1294 spin_lock(&rpc->crpc_lock);
1295 srpc_abort_rpc(rpc, rc);
1296 spin_unlock(&rpc->crpc_lock);
1297 }
1298
1299 abort:
1300 if (rpc->crpc_aborted) {
1301 LNetMDUnlink(rpc->crpc_reqstmdh);
1302 LNetMDUnlink(rpc->crpc_replymdh);
1303 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1304
1305 if (!srpc_event_pending(rpc)) {
1306 srpc_client_rpc_done(rpc, -EINTR);
1307 return 1;
1308 }
1309 }
1310 return 0;
1311 }
1312
1313 struct srpc_client_rpc *
srpc_create_client_rpc(struct lnet_process_id peer,int service,int nbulkiov,int bulklen,void (* rpc_done)(struct srpc_client_rpc *),void (* rpc_fini)(struct srpc_client_rpc *),void * priv)1314 srpc_create_client_rpc(struct lnet_process_id peer, int service,
1315 int nbulkiov, int bulklen,
1316 void (*rpc_done)(struct srpc_client_rpc *),
1317 void (*rpc_fini)(struct srpc_client_rpc *), void *priv)
1318 {
1319 struct srpc_client_rpc *rpc;
1320
1321 LIBCFS_ALLOC(rpc, offsetof(struct srpc_client_rpc,
1322 crpc_bulk.bk_iovs[nbulkiov]));
1323 if (!rpc)
1324 return NULL;
1325
1326 srpc_init_client_rpc(rpc, peer, service, nbulkiov,
1327 bulklen, rpc_done, rpc_fini, priv);
1328 return rpc;
1329 }
1330
1331 /* called with rpc->crpc_lock held */
1332 void
srpc_abort_rpc(struct srpc_client_rpc * rpc,int why)1333 srpc_abort_rpc(struct srpc_client_rpc *rpc, int why)
1334 {
1335 LASSERT(why);
1336
1337 if (rpc->crpc_aborted || /* already aborted */
1338 rpc->crpc_closed) /* callback imminent */
1339 return;
1340
1341 CDEBUG(D_NET, "Aborting RPC: service %d, peer %s, state %s, why %d\n",
1342 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1343 swi_state2str(rpc->crpc_wi.swi_state), why);
1344
1345 rpc->crpc_aborted = 1;
1346 rpc->crpc_status = why;
1347 swi_schedule_workitem(&rpc->crpc_wi);
1348 }
1349
1350 /* called with rpc->crpc_lock held */
1351 void
srpc_post_rpc(struct srpc_client_rpc * rpc)1352 srpc_post_rpc(struct srpc_client_rpc *rpc)
1353 {
1354 LASSERT(!rpc->crpc_aborted);
1355 LASSERT(srpc_data.rpc_state == SRPC_STATE_RUNNING);
1356
1357 CDEBUG(D_NET, "Posting RPC: peer %s, service %d, timeout %d\n",
1358 libcfs_id2str(rpc->crpc_dest), rpc->crpc_service,
1359 rpc->crpc_timeout);
1360
1361 srpc_add_client_rpc_timer(rpc);
1362 swi_schedule_workitem(&rpc->crpc_wi);
1363 }
1364
1365 int
srpc_send_reply(struct srpc_server_rpc * rpc)1366 srpc_send_reply(struct srpc_server_rpc *rpc)
1367 {
1368 struct srpc_event *ev = &rpc->srpc_ev;
1369 struct srpc_msg *msg = &rpc->srpc_replymsg;
1370 struct srpc_buffer *buffer = rpc->srpc_reqstbuf;
1371 struct srpc_service_cd *scd = rpc->srpc_scd;
1372 struct srpc_service *sv = scd->scd_svc;
1373 __u64 rpyid;
1374 int rc;
1375
1376 LASSERT(buffer);
1377 rpyid = buffer->buf_msg.msg_body.reqst.rpyid;
1378
1379 spin_lock(&scd->scd_lock);
1380
1381 if (!sv->sv_shuttingdown && !srpc_serv_is_framework(sv)) {
1382 /*
1383 * Repost buffer before replying since test client
1384 * might send me another RPC once it gets the reply
1385 */
1386 if (srpc_service_post_buffer(scd, buffer))
1387 CWARN("Failed to repost %s buffer\n", sv->sv_name);
1388 rpc->srpc_reqstbuf = NULL;
1389 }
1390
1391 spin_unlock(&scd->scd_lock);
1392
1393 ev->ev_fired = 0;
1394 ev->ev_data = rpc;
1395 ev->ev_type = SRPC_REPLY_SENT;
1396
1397 msg->msg_magic = SRPC_MSG_MAGIC;
1398 msg->msg_version = SRPC_MSG_VERSION;
1399 msg->msg_type = srpc_service2reply(sv->sv_id);
1400
1401 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, rpyid, msg,
1402 sizeof(*msg), LNET_MD_OP_PUT,
1403 rpc->srpc_peer, rpc->srpc_self,
1404 &rpc->srpc_replymdh, ev);
1405 if (rc)
1406 ev->ev_fired = 1; /* no more event expected */
1407 return rc;
1408 }
1409
1410 /* when in kernel always called with LNET_LOCK() held, and in thread context */
1411 static void
srpc_lnet_ev_handler(struct lnet_event * ev)1412 srpc_lnet_ev_handler(struct lnet_event *ev)
1413 {
1414 struct srpc_service_cd *scd;
1415 struct srpc_event *rpcev = ev->md.user_ptr;
1416 struct srpc_client_rpc *crpc;
1417 struct srpc_server_rpc *srpc;
1418 struct srpc_buffer *buffer;
1419 struct srpc_service *sv;
1420 struct srpc_msg *msg;
1421 enum srpc_msg_type type;
1422
1423 LASSERT(!in_interrupt());
1424
1425 if (ev->status) {
1426 __u32 errors;
1427
1428 spin_lock(&srpc_data.rpc_glock);
1429 if (ev->status != -ECANCELED) /* cancellation is not error */
1430 srpc_data.rpc_counters.errors++;
1431 errors = srpc_data.rpc_counters.errors;
1432 spin_unlock(&srpc_data.rpc_glock);
1433
1434 CNETERR("LNet event status %d type %d, RPC errors %u\n",
1435 ev->status, ev->type, errors);
1436 }
1437
1438 rpcev->ev_lnet = ev->type;
1439
1440 switch (rpcev->ev_type) {
1441 default:
1442 CERROR("Unknown event: status %d, type %d, lnet %d\n",
1443 rpcev->ev_status, rpcev->ev_type, rpcev->ev_lnet);
1444 LBUG();
1445 case SRPC_REQUEST_SENT:
1446 if (!ev->status && ev->type != LNET_EVENT_UNLINK) {
1447 spin_lock(&srpc_data.rpc_glock);
1448 srpc_data.rpc_counters.rpcs_sent++;
1449 spin_unlock(&srpc_data.rpc_glock);
1450 }
1451 case SRPC_REPLY_RCVD:
1452 case SRPC_BULK_REQ_RCVD:
1453 crpc = rpcev->ev_data;
1454
1455 if (rpcev != &crpc->crpc_reqstev &&
1456 rpcev != &crpc->crpc_replyev &&
1457 rpcev != &crpc->crpc_bulkev) {
1458 CERROR("rpcev %p, crpc %p, reqstev %p, replyev %p, bulkev %p\n",
1459 rpcev, crpc, &crpc->crpc_reqstev,
1460 &crpc->crpc_replyev, &crpc->crpc_bulkev);
1461 CERROR("Bad event: status %d, type %d, lnet %d\n",
1462 rpcev->ev_status, rpcev->ev_type, rpcev->ev_lnet);
1463 LBUG();
1464 }
1465
1466 spin_lock(&crpc->crpc_lock);
1467
1468 LASSERT(!rpcev->ev_fired);
1469 rpcev->ev_fired = 1;
1470 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1471 -EINTR : ev->status;
1472 swi_schedule_workitem(&crpc->crpc_wi);
1473
1474 spin_unlock(&crpc->crpc_lock);
1475 break;
1476
1477 case SRPC_REQUEST_RCVD:
1478 scd = rpcev->ev_data;
1479 sv = scd->scd_svc;
1480
1481 LASSERT(rpcev == &scd->scd_ev);
1482
1483 spin_lock(&scd->scd_lock);
1484
1485 LASSERT(ev->unlinked);
1486 LASSERT(ev->type == LNET_EVENT_PUT ||
1487 ev->type == LNET_EVENT_UNLINK);
1488 LASSERT(ev->type != LNET_EVENT_UNLINK ||
1489 sv->sv_shuttingdown);
1490
1491 buffer = container_of(ev->md.start, struct srpc_buffer, buf_msg);
1492 buffer->buf_peer = ev->initiator;
1493 buffer->buf_self = ev->target.nid;
1494
1495 LASSERT(scd->scd_buf_nposted > 0);
1496 scd->scd_buf_nposted--;
1497
1498 if (sv->sv_shuttingdown) {
1499 /*
1500 * Leave buffer on scd->scd_buf_nposted since
1501 * srpc_finish_service needs to traverse it.
1502 */
1503 spin_unlock(&scd->scd_lock);
1504 break;
1505 }
1506
1507 if (scd->scd_buf_err_stamp &&
1508 scd->scd_buf_err_stamp < ktime_get_real_seconds()) {
1509 /* re-enable adding buffer */
1510 scd->scd_buf_err_stamp = 0;
1511 scd->scd_buf_err = 0;
1512 }
1513
1514 if (!scd->scd_buf_err && /* adding buffer is enabled */
1515 !scd->scd_buf_adjust &&
1516 scd->scd_buf_nposted < scd->scd_buf_low) {
1517 scd->scd_buf_adjust = max(scd->scd_buf_total / 2,
1518 SFW_TEST_WI_MIN);
1519 swi_schedule_workitem(&scd->scd_buf_wi);
1520 }
1521
1522 list_del(&buffer->buf_list); /* from scd->scd_buf_posted */
1523 msg = &buffer->buf_msg;
1524 type = srpc_service2request(sv->sv_id);
1525
1526 if (ev->status || ev->mlength != sizeof(*msg) ||
1527 (msg->msg_type != type &&
1528 msg->msg_type != __swab32(type)) ||
1529 (msg->msg_magic != SRPC_MSG_MAGIC &&
1530 msg->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1531 CERROR("Dropping RPC (%s) from %s: status %d mlength %d type %u magic %u.\n",
1532 sv->sv_name, libcfs_id2str(ev->initiator),
1533 ev->status, ev->mlength,
1534 msg->msg_type, msg->msg_magic);
1535
1536 /*
1537 * NB can't call srpc_service_recycle_buffer here since
1538 * it may call LNetM[DE]Attach. The invalid magic tells
1539 * srpc_handle_rpc to drop this RPC
1540 */
1541 msg->msg_magic = 0;
1542 }
1543
1544 if (!list_empty(&scd->scd_rpc_free)) {
1545 srpc = list_entry(scd->scd_rpc_free.next,
1546 struct srpc_server_rpc,
1547 srpc_list);
1548 list_del(&srpc->srpc_list);
1549
1550 srpc_init_server_rpc(srpc, scd, buffer);
1551 list_add_tail(&srpc->srpc_list,
1552 &scd->scd_rpc_active);
1553 swi_schedule_workitem(&srpc->srpc_wi);
1554 } else {
1555 list_add_tail(&buffer->buf_list,
1556 &scd->scd_buf_blocked);
1557 }
1558
1559 spin_unlock(&scd->scd_lock);
1560
1561 spin_lock(&srpc_data.rpc_glock);
1562 srpc_data.rpc_counters.rpcs_rcvd++;
1563 spin_unlock(&srpc_data.rpc_glock);
1564 break;
1565
1566 case SRPC_BULK_GET_RPLD:
1567 LASSERT(ev->type == LNET_EVENT_SEND ||
1568 ev->type == LNET_EVENT_REPLY ||
1569 ev->type == LNET_EVENT_UNLINK);
1570
1571 if (!ev->unlinked)
1572 break; /* wait for final event */
1573
1574 case SRPC_BULK_PUT_SENT:
1575 if (!ev->status && ev->type != LNET_EVENT_UNLINK) {
1576 spin_lock(&srpc_data.rpc_glock);
1577
1578 if (rpcev->ev_type == SRPC_BULK_GET_RPLD)
1579 srpc_data.rpc_counters.bulk_get += ev->mlength;
1580 else
1581 srpc_data.rpc_counters.bulk_put += ev->mlength;
1582
1583 spin_unlock(&srpc_data.rpc_glock);
1584 }
1585 case SRPC_REPLY_SENT:
1586 srpc = rpcev->ev_data;
1587 scd = srpc->srpc_scd;
1588
1589 LASSERT(rpcev == &srpc->srpc_ev);
1590
1591 spin_lock(&scd->scd_lock);
1592
1593 rpcev->ev_fired = 1;
1594 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1595 -EINTR : ev->status;
1596 swi_schedule_workitem(&srpc->srpc_wi);
1597
1598 spin_unlock(&scd->scd_lock);
1599 break;
1600 }
1601 }
1602
1603 int
srpc_startup(void)1604 srpc_startup(void)
1605 {
1606 int rc;
1607
1608 memset(&srpc_data, 0, sizeof(struct smoketest_rpc));
1609 spin_lock_init(&srpc_data.rpc_glock);
1610
1611 /* 1 second pause to avoid timestamp reuse */
1612 set_current_state(TASK_UNINTERRUPTIBLE);
1613 schedule_timeout(cfs_time_seconds(1));
1614 srpc_data.rpc_matchbits = ((__u64)ktime_get_real_seconds()) << 48;
1615
1616 srpc_data.rpc_state = SRPC_STATE_NONE;
1617
1618 rc = LNetNIInit(LNET_PID_LUSTRE);
1619 if (rc < 0) {
1620 CERROR("LNetNIInit() has failed: %d\n", rc);
1621 return rc;
1622 }
1623
1624 srpc_data.rpc_state = SRPC_STATE_NI_INIT;
1625
1626 LNetInvalidateEQHandle(&srpc_data.rpc_lnet_eq);
1627 rc = LNetEQAlloc(0, srpc_lnet_ev_handler, &srpc_data.rpc_lnet_eq);
1628 if (rc) {
1629 CERROR("LNetEQAlloc() has failed: %d\n", rc);
1630 goto bail;
1631 }
1632
1633 rc = LNetSetLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1634 LASSERT(!rc);
1635 rc = LNetSetLazyPortal(SRPC_REQUEST_PORTAL);
1636 LASSERT(!rc);
1637
1638 srpc_data.rpc_state = SRPC_STATE_EQ_INIT;
1639
1640 rc = stt_startup();
1641
1642 bail:
1643 if (rc)
1644 srpc_shutdown();
1645 else
1646 srpc_data.rpc_state = SRPC_STATE_RUNNING;
1647
1648 return rc;
1649 }
1650
1651 void
srpc_shutdown(void)1652 srpc_shutdown(void)
1653 {
1654 int i;
1655 int rc;
1656 int state;
1657
1658 state = srpc_data.rpc_state;
1659 srpc_data.rpc_state = SRPC_STATE_STOPPING;
1660
1661 switch (state) {
1662 default:
1663 LBUG();
1664 case SRPC_STATE_RUNNING:
1665 spin_lock(&srpc_data.rpc_glock);
1666
1667 for (i = 0; i <= SRPC_SERVICE_MAX_ID; i++) {
1668 struct srpc_service *sv = srpc_data.rpc_services[i];
1669
1670 LASSERTF(!sv, "service not empty: id %d, name %s\n",
1671 i, sv->sv_name);
1672 }
1673
1674 spin_unlock(&srpc_data.rpc_glock);
1675
1676 stt_shutdown();
1677
1678 case SRPC_STATE_EQ_INIT:
1679 rc = LNetClearLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1680 rc = LNetClearLazyPortal(SRPC_REQUEST_PORTAL);
1681 LASSERT(!rc);
1682 rc = LNetEQFree(srpc_data.rpc_lnet_eq);
1683 LASSERT(!rc); /* the EQ should have no user by now */
1684
1685 case SRPC_STATE_NI_INIT:
1686 LNetNIFini();
1687 }
1688 }
1689