1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Isaac Huang <isaac@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_LNET
46 #define SRPC_PEER_HASH_SIZE 101 /* # peer lists */
57 struct smoketest_rpc {
58 spinlock_t rpc_glock; /* global lock */
59 srpc_service_t *rpc_services[SRPC_SERVICE_MAX_ID + 1];
60 struct list_head *rpc_peers; /* hash table of known peers */
61 lnet_handle_eq_t rpc_lnet_eq; /* _the_ LNet event queue */
62 srpc_state_t rpc_state;
63 srpc_counters_t rpc_counters;
64 __u64 rpc_matchbits; /* matchbits counter */
67 static int srpc_peer_credits = 16;
68 CFS_MODULE_PARM(srpc_peer_credits, "i", int, 0444,
69 "# in-flight RPCs per peer (16 by default)");
72 int srpc_handle_rpc (swi_workitem_t *wi);
74 void srpc_get_counters (srpc_counters_t *cnt)
76 spin_lock(&srpc_data.rpc_glock);
77 *cnt = srpc_data.rpc_counters;
78 spin_unlock(&srpc_data.rpc_glock);
81 void srpc_set_counters (const srpc_counters_t *cnt)
83 spin_lock(&srpc_data.rpc_glock);
84 srpc_data.rpc_counters = *cnt;
85 spin_unlock(&srpc_data.rpc_glock);
89 srpc_add_bulk_page (srpc_bulk_t *bk, cfs_page_t *pg, int i)
91 LASSERT (i >= 0 && i < bk->bk_niov);
94 bk->bk_iovs[i].kiov_offset = 0;
95 bk->bk_iovs[i].kiov_page = pg;
96 bk->bk_iovs[i].kiov_len = CFS_PAGE_SIZE;
98 LASSERT (bk->bk_pages != NULL);
100 bk->bk_pages[i] = pg;
101 bk->bk_iovs[i].iov_len = CFS_PAGE_SIZE;
102 bk->bk_iovs[i].iov_base = cfs_page_address(pg);
108 srpc_free_bulk (srpc_bulk_t *bk)
113 LASSERT (bk != NULL);
115 LASSERT (bk->bk_pages != NULL);
118 for (i = 0; i < bk->bk_niov; i++) {
120 pg = bk->bk_iovs[i].kiov_page;
122 pg = bk->bk_pages[i];
124 if (pg == NULL) break;
130 LIBCFS_FREE(bk->bk_pages, sizeof(cfs_page_t *) * bk->bk_niov);
132 LIBCFS_FREE(bk, offsetof(srpc_bulk_t, bk_iovs[bk->bk_niov]));
137 srpc_alloc_bulk (int npages, int sink)
143 LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
145 LIBCFS_ALLOC(bk, offsetof(srpc_bulk_t, bk_iovs[npages]));
147 CERROR ("Can't allocate descriptor for %d pages\n", npages);
151 memset(bk, 0, offsetof(srpc_bulk_t, bk_iovs[npages]));
153 bk->bk_niov = npages;
154 bk->bk_len = npages * CFS_PAGE_SIZE;
156 LIBCFS_ALLOC(pages, sizeof(cfs_page_t *) * npages);
158 LIBCFS_FREE(bk, offsetof(srpc_bulk_t, bk_iovs[npages]));
159 CERROR ("Can't allocate page array for %d pages\n", npages);
163 memset(pages, 0, sizeof(cfs_page_t *) * npages);
164 bk->bk_pages = pages;
169 for (i = 0; i < npages; i++) {
170 cfs_page_t *pg = cfs_alloc_page(CFS_ALLOC_STD);
173 CERROR ("Can't allocate page %d of %d\n", i, npages);
178 srpc_add_bulk_page(bk, pg, i);
185 static inline struct list_head *
186 srpc_nid2peerlist (lnet_nid_t nid)
188 unsigned int hash = ((unsigned int)nid) % SRPC_PEER_HASH_SIZE;
190 return &srpc_data.rpc_peers[hash];
193 static inline srpc_peer_t *
194 srpc_create_peer (lnet_nid_t nid)
198 LASSERT (nid != LNET_NID_ANY);
200 LIBCFS_ALLOC(peer, sizeof(srpc_peer_t));
202 CERROR ("Failed to allocate peer structure for %s\n",
203 libcfs_nid2str(nid));
207 memset(peer, 0, sizeof(srpc_peer_t));
209 peer->stp_credits = srpc_peer_credits;
211 spin_lock_init(&peer->stp_lock);
212 CFS_INIT_LIST_HEAD(&peer->stp_rpcq);
213 CFS_INIT_LIST_HEAD(&peer->stp_ctl_rpcq);
218 srpc_find_peer_locked (lnet_nid_t nid)
220 struct list_head *peer_list = srpc_nid2peerlist(nid);
223 LASSERT (nid != LNET_NID_ANY);
225 cfs_list_for_each_entry_typed (peer, peer_list,
226 srpc_peer_t, stp_list) {
227 if (peer->stp_nid == nid)
235 srpc_nid2peer (lnet_nid_t nid)
238 srpc_peer_t *new_peer;
240 spin_lock(&srpc_data.rpc_glock);
241 peer = srpc_find_peer_locked(nid);
242 spin_unlock(&srpc_data.rpc_glock);
247 new_peer = srpc_create_peer(nid);
249 spin_lock(&srpc_data.rpc_glock);
251 peer = srpc_find_peer_locked(nid);
253 spin_unlock(&srpc_data.rpc_glock);
254 if (new_peer != NULL)
255 LIBCFS_FREE(new_peer, sizeof(srpc_peer_t));
260 if (new_peer == NULL) {
261 spin_unlock(&srpc_data.rpc_glock);
265 list_add_tail(&new_peer->stp_list, srpc_nid2peerlist(nid));
266 spin_unlock(&srpc_data.rpc_glock);
275 spin_lock(&srpc_data.rpc_glock);
276 id = srpc_data.rpc_matchbits++;
277 spin_unlock(&srpc_data.rpc_glock);
282 srpc_init_server_rpc (srpc_server_rpc_t *rpc,
283 srpc_service_t *sv, srpc_buffer_t *buffer)
285 memset(rpc, 0, sizeof(*rpc));
286 swi_init_workitem(&rpc->srpc_wi, rpc, srpc_handle_rpc);
288 rpc->srpc_ev.ev_fired = 1; /* no event expected now */
290 rpc->srpc_service = sv;
291 rpc->srpc_reqstbuf = buffer;
292 rpc->srpc_peer = buffer->buf_peer;
293 rpc->srpc_self = buffer->buf_self;
294 LNetInvalidateHandle(&rpc->srpc_replymdh);
298 srpc_add_service (srpc_service_t *sv)
302 srpc_server_rpc_t *rpc;
304 LASSERT (sv->sv_concur > 0);
305 LASSERT (0 <= id && id <= SRPC_SERVICE_MAX_ID);
307 spin_lock(&srpc_data.rpc_glock);
309 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
311 if (srpc_data.rpc_services[id] != NULL) {
312 spin_unlock(&srpc_data.rpc_glock);
316 srpc_data.rpc_services[id] = sv;
317 spin_unlock(&srpc_data.rpc_glock);
320 sv->sv_nposted_msg = 0;
321 sv->sv_shuttingdown = 0;
322 spin_lock_init(&sv->sv_lock);
323 CFS_INIT_LIST_HEAD(&sv->sv_free_rpcq);
324 CFS_INIT_LIST_HEAD(&sv->sv_active_rpcq);
325 CFS_INIT_LIST_HEAD(&sv->sv_posted_msgq);
326 CFS_INIT_LIST_HEAD(&sv->sv_blocked_msgq);
328 sv->sv_ev.ev_data = sv;
329 sv->sv_ev.ev_type = SRPC_REQUEST_RCVD;
331 for (i = 0; i < sv->sv_concur; i++) {
332 LIBCFS_ALLOC(rpc, sizeof(*rpc));
333 if (rpc == NULL) goto enomem;
335 list_add(&rpc->srpc_list, &sv->sv_free_rpcq);
338 CDEBUG (D_NET, "Adding service: id %d, name %s, concurrency %d\n",
339 id, sv->sv_name, sv->sv_concur);
343 while (!list_empty(&sv->sv_free_rpcq)) {
344 rpc = list_entry(sv->sv_free_rpcq.next,
345 srpc_server_rpc_t, srpc_list);
346 list_del(&rpc->srpc_list);
347 LIBCFS_FREE(rpc, sizeof(*rpc));
350 spin_lock(&srpc_data.rpc_glock);
351 srpc_data.rpc_services[id] = NULL;
352 spin_unlock(&srpc_data.rpc_glock);
357 srpc_remove_service (srpc_service_t *sv)
361 spin_lock(&srpc_data.rpc_glock);
363 if (srpc_data.rpc_services[id] != sv) {
364 spin_unlock(&srpc_data.rpc_glock);
368 srpc_data.rpc_services[id] = NULL;
369 spin_unlock(&srpc_data.rpc_glock);
374 srpc_post_passive_rdma(int portal, __u64 matchbits, void *buf,
375 int len, int options, lnet_process_id_t peer,
376 lnet_handle_md_t *mdh, srpc_event_t *ev)
380 lnet_handle_me_t meh;
382 rc = LNetMEAttach(portal, peer, matchbits, 0,
383 LNET_UNLINK, LNET_INS_AFTER, &meh);
385 CERROR ("LNetMEAttach failed: %d\n", rc);
386 LASSERT (rc == -ENOMEM);
394 md.options = options;
395 md.eq_handle = srpc_data.rpc_lnet_eq;
397 rc = LNetMDAttach(meh, md, LNET_UNLINK, mdh);
399 CERROR ("LNetMDAttach failed: %d\n", rc);
400 LASSERT (rc == -ENOMEM);
402 rc = LNetMEUnlink(meh);
408 "Posted passive RDMA: peer %s, portal %d, matchbits "LPX64"\n",
409 libcfs_id2str(peer), portal, matchbits);
414 srpc_post_active_rdma(int portal, __u64 matchbits, void *buf, int len,
415 int options, lnet_process_id_t peer, lnet_nid_t self,
416 lnet_handle_md_t *mdh, srpc_event_t *ev)
424 md.eq_handle = srpc_data.rpc_lnet_eq;
425 md.threshold = ((options & LNET_MD_OP_GET) != 0) ? 2 : 1;
426 md.options = options & ~(LNET_MD_OP_PUT | LNET_MD_OP_GET);
428 rc = LNetMDBind(md, LNET_UNLINK, mdh);
430 CERROR ("LNetMDBind failed: %d\n", rc);
431 LASSERT (rc == -ENOMEM);
435 /* this is kind of an abuse of the LNET_MD_OP_{PUT,GET} options.
436 * they're only meaningful for MDs attached to an ME (i.e. passive
438 if ((options & LNET_MD_OP_PUT) != 0) {
439 rc = LNetPut(self, *mdh, LNET_NOACK_REQ, peer,
440 portal, matchbits, 0, 0);
442 LASSERT ((options & LNET_MD_OP_GET) != 0);
444 rc = LNetGet(self, *mdh, peer, portal, matchbits, 0);
448 CERROR ("LNet%s(%s, %d, "LPD64") failed: %d\n",
449 ((options & LNET_MD_OP_PUT) != 0) ? "Put" : "Get",
450 libcfs_id2str(peer), portal, matchbits, rc);
452 /* The forthcoming unlink event will complete this operation
453 * with failure, so fall through and return success here.
455 rc = LNetMDUnlink(*mdh);
459 "Posted active RDMA: peer %s, portal %u, matchbits "LPX64"\n",
460 libcfs_id2str(peer), portal, matchbits);
466 srpc_post_active_rqtbuf(lnet_process_id_t peer, int service, void *buf,
467 int len, lnet_handle_md_t *mdh, srpc_event_t *ev)
472 if (service > SRPC_FRAMEWORK_SERVICE_MAX_ID)
473 portal = SRPC_REQUEST_PORTAL;
475 portal = SRPC_FRAMEWORK_REQUEST_PORTAL;
477 rc = srpc_post_active_rdma(portal, service, buf, len,
478 LNET_MD_OP_PUT, peer,
479 LNET_NID_ANY, mdh, ev);
484 srpc_post_passive_rqtbuf(int service, void *buf, int len,
485 lnet_handle_md_t *mdh, srpc_event_t *ev)
489 lnet_process_id_t any = {0};
491 any.nid = LNET_NID_ANY;
492 any.pid = LNET_PID_ANY;
494 if (service > SRPC_FRAMEWORK_SERVICE_MAX_ID)
495 portal = SRPC_REQUEST_PORTAL;
497 portal = SRPC_FRAMEWORK_REQUEST_PORTAL;
499 rc = srpc_post_passive_rdma(portal, service, buf, len,
500 LNET_MD_OP_PUT, any, mdh, ev);
505 srpc_service_post_buffer (srpc_service_t *sv, srpc_buffer_t *buf)
507 srpc_msg_t *msg = &buf->buf_msg;
510 LASSERT (!sv->sv_shuttingdown);
512 LNetInvalidateHandle(&buf->buf_mdh);
513 list_add(&buf->buf_list, &sv->sv_posted_msgq);
514 sv->sv_nposted_msg++;
515 spin_unlock(&sv->sv_lock);
517 rc = srpc_post_passive_rqtbuf(sv->sv_id, msg, sizeof(*msg),
518 &buf->buf_mdh, &sv->sv_ev);
520 /* At this point, a RPC (new or delayed) may have arrived in
521 * msg and its event handler has been called. So we must add
522 * buf to sv_posted_msgq _before_ dropping sv_lock */
524 spin_lock(&sv->sv_lock);
527 if (sv->sv_shuttingdown) {
528 spin_unlock(&sv->sv_lock);
530 /* srpc_shutdown_service might have tried to unlink me
531 * when my buf_mdh was still invalid */
532 LNetMDUnlink(buf->buf_mdh);
534 spin_lock(&sv->sv_lock);
539 sv->sv_nposted_msg--;
540 if (sv->sv_shuttingdown) return rc;
542 list_del(&buf->buf_list);
544 spin_unlock(&sv->sv_lock);
545 LIBCFS_FREE(buf, sizeof(*buf));
546 spin_lock(&sv->sv_lock);
551 srpc_service_add_buffers (srpc_service_t *sv, int nbuffer)
557 LASSERTF (nbuffer > 0,
558 "nbuffer must be positive: %d\n", nbuffer);
560 for (posted = 0; posted < nbuffer; posted++) {
561 LIBCFS_ALLOC(buf, sizeof(*buf));
562 if (buf == NULL) break;
564 spin_lock(&sv->sv_lock);
565 rc = srpc_service_post_buffer(sv, buf);
566 spin_unlock(&sv->sv_lock);
575 srpc_service_remove_buffers (srpc_service_t *sv, int nbuffer)
577 LASSERTF (nbuffer > 0,
578 "nbuffer must be positive: %d\n", nbuffer);
580 spin_lock(&sv->sv_lock);
582 LASSERT (sv->sv_nprune >= 0);
583 LASSERT (!sv->sv_shuttingdown);
585 sv->sv_nprune += nbuffer;
587 spin_unlock(&sv->sv_lock);
591 /* returns 1 if sv has finished, otherwise 0 */
593 srpc_finish_service (srpc_service_t *sv)
595 srpc_server_rpc_t *rpc;
598 spin_lock(&sv->sv_lock);
600 LASSERT (sv->sv_shuttingdown); /* srpc_shutdown_service called */
602 if (sv->sv_nposted_msg != 0 || !list_empty(&sv->sv_active_rpcq)) {
604 "waiting for %d posted buffers to unlink and "
605 "in-flight RPCs to die.\n",
608 if (!list_empty(&sv->sv_active_rpcq)) {
609 rpc = list_entry(sv->sv_active_rpcq.next,
610 srpc_server_rpc_t, srpc_list);
612 "Active RPC on shutdown: sv %s, peer %s, "
613 "wi %s scheduled %d running %d, "
614 "ev fired %d type %d status %d lnet %d\n",
615 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
616 swi_state2str(rpc->srpc_wi.wi_state),
617 rpc->srpc_wi.wi_scheduled,
618 rpc->srpc_wi.wi_running,
619 rpc->srpc_ev.ev_fired,
620 rpc->srpc_ev.ev_type,
621 rpc->srpc_ev.ev_status,
622 rpc->srpc_ev.ev_lnet);
625 spin_unlock(&sv->sv_lock);
629 spin_unlock(&sv->sv_lock); /* no lock needed from now on */
634 if (!list_empty(&sv->sv_posted_msgq))
635 q = &sv->sv_posted_msgq;
636 else if (!list_empty(&sv->sv_blocked_msgq))
637 q = &sv->sv_blocked_msgq;
641 buf = list_entry(q->next, srpc_buffer_t, buf_list);
642 list_del(&buf->buf_list);
644 LIBCFS_FREE(buf, sizeof(*buf));
647 while (!list_empty(&sv->sv_free_rpcq)) {
648 rpc = list_entry(sv->sv_free_rpcq.next,
649 srpc_server_rpc_t, srpc_list);
650 list_del(&rpc->srpc_list);
651 LIBCFS_FREE(rpc, sizeof(*rpc));
657 /* called with sv->sv_lock held */
659 srpc_service_recycle_buffer (srpc_service_t *sv, srpc_buffer_t *buf)
661 if (sv->sv_shuttingdown) goto free;
663 if (sv->sv_nprune == 0) {
664 if (srpc_service_post_buffer(sv, buf) != 0)
665 CWARN ("Failed to post %s buffer\n", sv->sv_name);
671 spin_unlock(&sv->sv_lock);
672 LIBCFS_FREE(buf, sizeof(*buf));
673 spin_lock(&sv->sv_lock);
677 srpc_shutdown_service (srpc_service_t *sv)
679 srpc_server_rpc_t *rpc;
682 spin_lock(&sv->sv_lock);
684 CDEBUG (D_NET, "Shutting down service: id %d, name %s\n",
685 sv->sv_id, sv->sv_name);
687 sv->sv_shuttingdown = 1; /* i.e. no new active RPC */
689 /* schedule in-flight RPCs to notice the shutdown */
690 cfs_list_for_each_entry_typed (rpc, &sv->sv_active_rpcq,
691 srpc_server_rpc_t, srpc_list) {
692 swi_schedule_workitem(&rpc->srpc_wi);
695 spin_unlock(&sv->sv_lock);
697 /* OK to traverse sv_posted_msgq without lock, since no one
698 * touches sv_posted_msgq now */
699 cfs_list_for_each_entry_typed (buf, &sv->sv_posted_msgq,
700 srpc_buffer_t, buf_list)
701 LNetMDUnlink(buf->buf_mdh);
707 srpc_send_request (srpc_client_rpc_t *rpc)
709 srpc_event_t *ev = &rpc->crpc_reqstev;
714 ev->ev_type = SRPC_REQUEST_SENT;
716 rc = srpc_post_active_rqtbuf(rpc->crpc_dest, rpc->crpc_service,
717 &rpc->crpc_reqstmsg, sizeof(srpc_msg_t),
718 &rpc->crpc_reqstmdh, ev);
720 LASSERT (rc == -ENOMEM);
721 ev->ev_fired = 1; /* no more event expected */
727 srpc_prepare_reply (srpc_client_rpc_t *rpc)
729 srpc_event_t *ev = &rpc->crpc_replyev;
730 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.rpyid;
735 ev->ev_type = SRPC_REPLY_RCVD;
737 *id = srpc_next_id();
739 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, *id,
740 &rpc->crpc_replymsg, sizeof(srpc_msg_t),
741 LNET_MD_OP_PUT, rpc->crpc_dest,
742 &rpc->crpc_replymdh, ev);
744 LASSERT (rc == -ENOMEM);
745 ev->ev_fired = 1; /* no more event expected */
751 srpc_prepare_bulk (srpc_client_rpc_t *rpc)
753 srpc_bulk_t *bk = &rpc->crpc_bulk;
754 srpc_event_t *ev = &rpc->crpc_bulkev;
755 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.bulkid;
759 LASSERT (bk->bk_niov <= LNET_MAX_IOV);
761 if (bk->bk_niov == 0) return 0; /* nothing to do */
763 opt = bk->bk_sink ? LNET_MD_OP_PUT : LNET_MD_OP_GET;
767 opt |= LNET_MD_IOVEC;
772 ev->ev_type = SRPC_BULK_REQ_RCVD;
774 *id = srpc_next_id();
776 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, *id,
777 &bk->bk_iovs[0], bk->bk_niov, opt,
778 rpc->crpc_dest, &bk->bk_mdh, ev);
780 LASSERT (rc == -ENOMEM);
781 ev->ev_fired = 1; /* no more event expected */
787 srpc_do_bulk (srpc_server_rpc_t *rpc)
789 srpc_event_t *ev = &rpc->srpc_ev;
790 srpc_bulk_t *bk = rpc->srpc_bulk;
791 __u64 id = rpc->srpc_reqstbuf->buf_msg.msg_body.reqst.bulkid;
795 LASSERT (bk != NULL);
797 opt = bk->bk_sink ? LNET_MD_OP_GET : LNET_MD_OP_PUT;
801 opt |= LNET_MD_IOVEC;
806 ev->ev_type = bk->bk_sink ? SRPC_BULK_GET_RPLD : SRPC_BULK_PUT_SENT;
808 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, id,
809 &bk->bk_iovs[0], bk->bk_niov, opt,
810 rpc->srpc_peer, rpc->srpc_self,
813 ev->ev_fired = 1; /* no more event expected */
817 /* called with srpc_service_t::sv_lock held */
819 srpc_schedule_server_rpc (srpc_server_rpc_t *rpc)
821 srpc_service_t *sv = rpc->srpc_service;
823 if (sv->sv_id > SRPC_FRAMEWORK_SERVICE_MAX_ID)
824 swi_schedule_workitem(&rpc->srpc_wi);
825 else /* framework RPCs are handled one by one */
826 swi_schedule_serial_workitem(&rpc->srpc_wi);
831 /* only called from srpc_handle_rpc */
833 srpc_server_rpc_done (srpc_server_rpc_t *rpc, int status)
835 srpc_service_t *sv = rpc->srpc_service;
836 srpc_buffer_t *buffer;
838 LASSERT (status != 0 || rpc->srpc_wi.wi_state == SWI_STATE_DONE);
840 rpc->srpc_status = status;
842 CDEBUG (status == 0 ? D_NET : D_NETERROR,
843 "Server RPC done: service %s, peer %s, status %s:%d\n",
844 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
845 swi_state2str(rpc->srpc_wi.wi_state), status);
848 spin_lock(&srpc_data.rpc_glock);
849 srpc_data.rpc_counters.rpcs_dropped++;
850 spin_unlock(&srpc_data.rpc_glock);
853 if (rpc->srpc_done != NULL)
854 (*rpc->srpc_done) (rpc);
855 LASSERT (rpc->srpc_bulk == NULL);
857 spin_lock(&sv->sv_lock);
859 if (rpc->srpc_reqstbuf != NULL) {
860 /* NB might drop sv_lock in srpc_service_recycle_buffer, but
861 * sv won't go away for sv_active_rpcq must not be empty */
862 srpc_service_recycle_buffer(sv, rpc->srpc_reqstbuf);
863 rpc->srpc_reqstbuf = NULL;
866 list_del(&rpc->srpc_list); /* from sv->sv_active_rpcq */
869 * No one can schedule me now since:
870 * - I'm not on sv_active_rpcq.
871 * - all LNet events have been fired.
872 * Cancel pending schedules and prevent future schedule attempts:
874 LASSERT (rpc->srpc_ev.ev_fired);
875 swi_kill_workitem(&rpc->srpc_wi);
877 if (!sv->sv_shuttingdown && !list_empty(&sv->sv_blocked_msgq)) {
878 buffer = list_entry(sv->sv_blocked_msgq.next,
879 srpc_buffer_t, buf_list);
880 list_del(&buffer->buf_list);
882 srpc_init_server_rpc(rpc, sv, buffer);
883 list_add_tail(&rpc->srpc_list, &sv->sv_active_rpcq);
884 srpc_schedule_server_rpc(rpc);
886 list_add(&rpc->srpc_list, &sv->sv_free_rpcq);
889 spin_unlock(&sv->sv_lock);
893 /* handles an incoming RPC */
895 srpc_handle_rpc (swi_workitem_t *wi)
897 srpc_server_rpc_t *rpc = wi->wi_data;
898 srpc_service_t *sv = rpc->srpc_service;
899 srpc_event_t *ev = &rpc->srpc_ev;
902 LASSERT (wi == &rpc->srpc_wi);
904 spin_lock(&sv->sv_lock);
906 if (sv->sv_shuttingdown) {
907 spin_unlock(&sv->sv_lock);
909 if (rpc->srpc_bulk != NULL)
910 LNetMDUnlink(rpc->srpc_bulk->bk_mdh);
911 LNetMDUnlink(rpc->srpc_replymdh);
913 if (ev->ev_fired) { /* no more event, OK to finish */
914 srpc_server_rpc_done(rpc, -ESHUTDOWN);
920 spin_unlock(&sv->sv_lock);
922 switch (wi->wi_state) {
925 case SWI_STATE_NEWBORN: {
927 srpc_generic_reply_t *reply;
929 msg = &rpc->srpc_reqstbuf->buf_msg;
930 reply = &rpc->srpc_replymsg.msg_body.reply;
932 if (msg->msg_magic == 0) {
933 /* moaned already in srpc_lnet_ev_handler */
935 } else if (msg->msg_version != SRPC_MSG_VERSION &&
936 msg->msg_version != __swab32(SRPC_MSG_VERSION)) {
937 CWARN ("Version mismatch: %u, %u expected, from %s\n",
938 msg->msg_version, SRPC_MSG_VERSION,
939 libcfs_id2str(rpc->srpc_peer));
940 reply->status = EPROTO;
943 rc = (*sv->sv_handler) (rpc);
944 LASSERT (reply->status == 0 || !rpc->srpc_bulk);
948 srpc_server_rpc_done(rpc, rc);
952 wi->wi_state = SWI_STATE_BULK_STARTED;
954 if (rpc->srpc_bulk != NULL) {
955 rc = srpc_do_bulk(rpc);
957 return 0; /* wait for bulk */
959 LASSERT (ev->ev_fired);
963 case SWI_STATE_BULK_STARTED:
964 /* we cannot LASSERT ev_fired right here because it
965 * may be set only upon an event with unlinked==1 */
967 if (rpc->srpc_bulk != NULL) {
970 if (sv->sv_bulk_ready != NULL)
971 rc = (*sv->sv_bulk_ready) (rpc, rc);
975 srpc_server_rpc_done(rpc, rc);
979 rpc->srpc_status = rc;
980 wi->wi_state = SWI_STATE_BULK_ERRORED;
981 LNetMDUnlink(rpc->srpc_bulk->bk_mdh);
982 return 0; /* wait for UNLINK event */
986 LASSERT (rpc->srpc_bulk == NULL || ev->ev_fired);
988 wi->wi_state = SWI_STATE_REPLY_SUBMITTED;
989 rc = srpc_send_reply(rpc);
991 return 0; /* wait for reply */
992 srpc_server_rpc_done(rpc, rc);
995 case SWI_STATE_REPLY_SUBMITTED:
996 LASSERT (ev->ev_fired);
998 wi->wi_state = SWI_STATE_DONE;
999 srpc_server_rpc_done(rpc, ev->ev_status);
1002 case SWI_STATE_BULK_ERRORED:
1003 LASSERT (rpc->srpc_bulk != NULL && ev->ev_fired);
1004 LASSERT (rpc->srpc_status != 0);
1006 srpc_server_rpc_done(rpc, rpc->srpc_status);
1014 srpc_client_rpc_expired (void *data)
1016 srpc_client_rpc_t *rpc = data;
1018 CWARN ("Client RPC expired: service %d, peer %s, timeout %d.\n",
1019 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1022 spin_lock(&rpc->crpc_lock);
1024 rpc->crpc_timeout = 0;
1025 srpc_abort_rpc(rpc, -ETIMEDOUT);
1027 spin_unlock(&rpc->crpc_lock);
1029 spin_lock(&srpc_data.rpc_glock);
1030 srpc_data.rpc_counters.rpcs_expired++;
1031 spin_unlock(&srpc_data.rpc_glock);
1036 srpc_add_client_rpc_timer (srpc_client_rpc_t *rpc)
1038 stt_timer_t *timer = &rpc->crpc_timer;
1040 if (rpc->crpc_timeout == 0) return;
1042 CFS_INIT_LIST_HEAD(&timer->stt_list);
1043 timer->stt_data = rpc;
1044 timer->stt_func = srpc_client_rpc_expired;
1045 timer->stt_expires = cfs_time_add(rpc->crpc_timeout,
1046 cfs_time_current_sec());
1047 stt_add_timer(timer);
1052 * Called with rpc->crpc_lock held.
1054 * Upon exit the RPC expiry timer is not queued and the handler is not
1055 * running on any CPU. */
1057 srpc_del_client_rpc_timer (srpc_client_rpc_t *rpc)
1059 /* timer not planted or already exploded */
1060 if (rpc->crpc_timeout == 0) return;
1062 /* timer sucessfully defused */
1063 if (stt_del_timer(&rpc->crpc_timer)) return;
1066 /* timer detonated, wait for it to explode */
1067 while (rpc->crpc_timeout != 0) {
1068 spin_unlock(&rpc->crpc_lock);
1072 spin_lock(&rpc->crpc_lock);
1075 LBUG(); /* impossible in single-threaded runtime */
1081 srpc_check_sends (srpc_peer_t *peer, int credits)
1083 struct list_head *q;
1084 srpc_client_rpc_t *rpc;
1086 LASSERT (credits >= 0);
1087 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
1089 spin_lock(&peer->stp_lock);
1090 peer->stp_credits += credits;
1092 while (peer->stp_credits) {
1093 if (!list_empty(&peer->stp_ctl_rpcq))
1094 q = &peer->stp_ctl_rpcq;
1095 else if (!list_empty(&peer->stp_rpcq))
1096 q = &peer->stp_rpcq;
1100 peer->stp_credits--;
1102 rpc = list_entry(q->next, srpc_client_rpc_t, crpc_privl);
1103 list_del_init(&rpc->crpc_privl);
1104 srpc_client_rpc_decref(rpc); /* --ref for peer->*rpcq */
1106 swi_schedule_workitem(&rpc->crpc_wi);
1109 spin_unlock(&peer->stp_lock);
1114 srpc_client_rpc_done (srpc_client_rpc_t *rpc, int status)
1116 swi_workitem_t *wi = &rpc->crpc_wi;
1117 srpc_peer_t *peer = rpc->crpc_peer;
1119 LASSERT (status != 0 || wi->wi_state == SWI_STATE_DONE);
1121 spin_lock(&rpc->crpc_lock);
1123 rpc->crpc_closed = 1;
1124 if (rpc->crpc_status == 0)
1125 rpc->crpc_status = status;
1127 srpc_del_client_rpc_timer(rpc);
1129 CDEBUG ((status == 0) ? D_NET : D_NETERROR,
1130 "Client RPC done: service %d, peer %s, status %s:%d:%d\n",
1131 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1132 swi_state2str(wi->wi_state), rpc->crpc_aborted, status);
1135 * No one can schedule me now since:
1136 * - RPC timer has been defused.
1137 * - all LNet events have been fired.
1138 * - crpc_closed has been set, preventing srpc_abort_rpc from
1140 * Cancel pending schedules and prevent future schedule attempts:
1142 LASSERT (!srpc_event_pending(rpc));
1143 swi_kill_workitem(wi);
1145 spin_unlock(&rpc->crpc_lock);
1147 (*rpc->crpc_done) (rpc);
1150 srpc_check_sends(peer, 1);
1154 /* sends an outgoing RPC */
1156 srpc_send_rpc (swi_workitem_t *wi)
1159 srpc_client_rpc_t *rpc = wi->wi_data;
1160 srpc_msg_t *reply = &rpc->crpc_replymsg;
1161 int do_bulk = rpc->crpc_bulk.bk_niov > 0;
1163 LASSERT (rpc != NULL);
1164 LASSERT (wi == &rpc->crpc_wi);
1166 spin_lock(&rpc->crpc_lock);
1168 if (rpc->crpc_aborted) {
1169 spin_unlock(&rpc->crpc_lock);
1173 spin_unlock(&rpc->crpc_lock);
1175 switch (wi->wi_state) {
1178 case SWI_STATE_NEWBORN:
1179 LASSERT (!srpc_event_pending(rpc));
1181 rc = srpc_prepare_reply(rpc);
1183 srpc_client_rpc_done(rpc, rc);
1187 rc = srpc_prepare_bulk(rpc);
1190 wi->wi_state = SWI_STATE_REQUEST_SUBMITTED;
1191 rc = srpc_send_request(rpc);
1194 case SWI_STATE_REQUEST_SUBMITTED:
1195 /* CAVEAT EMPTOR: rqtev, rpyev, and bulkev may come in any
1196 * order; however, they're processed in a strict order:
1197 * rqt, rpy, and bulk. */
1198 if (!rpc->crpc_reqstev.ev_fired) break;
1200 rc = rpc->crpc_reqstev.ev_status;
1203 wi->wi_state = SWI_STATE_REQUEST_SENT;
1204 /* perhaps more events, fall thru */
1205 case SWI_STATE_REQUEST_SENT: {
1206 srpc_msg_type_t type = srpc_service2reply(rpc->crpc_service);
1208 if (!rpc->crpc_replyev.ev_fired) break;
1210 rc = rpc->crpc_replyev.ev_status;
1213 if ((reply->msg_type != type &&
1214 reply->msg_type != __swab32(type)) ||
1215 (reply->msg_magic != SRPC_MSG_MAGIC &&
1216 reply->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1217 CWARN ("Bad message from %s: type %u (%d expected),"
1218 " magic %u (%d expected).\n",
1219 libcfs_id2str(rpc->crpc_dest),
1220 reply->msg_type, type,
1221 reply->msg_magic, SRPC_MSG_MAGIC);
1226 if (do_bulk && reply->msg_body.reply.status != 0) {
1227 CWARN ("Remote error %d at %s, unlink bulk buffer in "
1228 "case peer didn't initiate bulk transfer\n",
1229 reply->msg_body.reply.status,
1230 libcfs_id2str(rpc->crpc_dest));
1231 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1234 wi->wi_state = SWI_STATE_REPLY_RECEIVED;
1236 case SWI_STATE_REPLY_RECEIVED:
1237 if (do_bulk && !rpc->crpc_bulkev.ev_fired) break;
1239 rc = do_bulk ? rpc->crpc_bulkev.ev_status : 0;
1241 /* Bulk buffer was unlinked due to remote error. Clear error
1242 * since reply buffer still contains valid data.
1243 * NB rpc->crpc_done shouldn't look into bulk data in case of
1245 if (do_bulk && rpc->crpc_bulkev.ev_lnet == LNET_EVENT_UNLINK &&
1246 rpc->crpc_status == 0 && reply->msg_body.reply.status != 0)
1249 wi->wi_state = SWI_STATE_DONE;
1250 srpc_client_rpc_done(rpc, rc);
1255 spin_lock(&rpc->crpc_lock);
1256 srpc_abort_rpc(rpc, rc);
1257 spin_unlock(&rpc->crpc_lock);
1261 if (rpc->crpc_aborted) {
1262 LNetMDUnlink(rpc->crpc_reqstmdh);
1263 LNetMDUnlink(rpc->crpc_replymdh);
1264 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1266 if (!srpc_event_pending(rpc)) {
1267 srpc_client_rpc_done(rpc, -EINTR);
1275 srpc_create_client_rpc (lnet_process_id_t peer, int service,
1276 int nbulkiov, int bulklen,
1277 void (*rpc_done)(srpc_client_rpc_t *),
1278 void (*rpc_fini)(srpc_client_rpc_t *), void *priv)
1280 srpc_client_rpc_t *rpc;
1282 LIBCFS_ALLOC(rpc, offsetof(srpc_client_rpc_t,
1283 crpc_bulk.bk_iovs[nbulkiov]));
1287 srpc_init_client_rpc(rpc, peer, service, nbulkiov,
1288 bulklen, rpc_done, rpc_fini, priv);
1292 /* called with rpc->crpc_lock held */
1294 srpc_queue_rpc (srpc_peer_t *peer, srpc_client_rpc_t *rpc)
1296 int service = rpc->crpc_service;
1298 LASSERT (peer->stp_nid == rpc->crpc_dest.nid);
1299 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
1301 rpc->crpc_peer = peer;
1303 spin_lock(&peer->stp_lock);
1305 /* Framework RPCs that alter session state shall take precedence
1306 * over test RPCs and framework query RPCs */
1307 if (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID &&
1308 service != SRPC_SERVICE_DEBUG &&
1309 service != SRPC_SERVICE_QUERY_STAT)
1310 list_add_tail(&rpc->crpc_privl, &peer->stp_ctl_rpcq);
1312 list_add_tail(&rpc->crpc_privl, &peer->stp_rpcq);
1314 srpc_client_rpc_addref(rpc); /* ++ref for peer->*rpcq */
1315 spin_unlock(&peer->stp_lock);
1319 /* called with rpc->crpc_lock held */
1321 srpc_abort_rpc (srpc_client_rpc_t *rpc, int why)
1323 srpc_peer_t *peer = rpc->crpc_peer;
1327 if (rpc->crpc_aborted || /* already aborted */
1328 rpc->crpc_closed) /* callback imminent */
1332 "Aborting RPC: service %d, peer %s, state %s, why %d\n",
1333 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1334 swi_state2str(rpc->crpc_wi.wi_state), why);
1336 rpc->crpc_aborted = 1;
1337 rpc->crpc_status = why;
1340 spin_lock(&peer->stp_lock);
1342 if (!list_empty(&rpc->crpc_privl)) { /* still queued */
1343 list_del_init(&rpc->crpc_privl);
1344 srpc_client_rpc_decref(rpc); /* --ref for peer->*rpcq */
1345 rpc->crpc_peer = NULL; /* no credit taken */
1348 spin_unlock(&peer->stp_lock);
1351 swi_schedule_workitem(&rpc->crpc_wi);
1355 /* called with rpc->crpc_lock held */
1357 srpc_post_rpc (srpc_client_rpc_t *rpc)
1361 LASSERT (!rpc->crpc_aborted);
1362 LASSERT (rpc->crpc_peer == NULL);
1363 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
1364 LASSERT ((rpc->crpc_bulk.bk_len & ~CFS_PAGE_MASK) == 0);
1366 CDEBUG (D_NET, "Posting RPC: peer %s, service %d, timeout %d\n",
1367 libcfs_id2str(rpc->crpc_dest), rpc->crpc_service,
1370 srpc_add_client_rpc_timer(rpc);
1372 peer = srpc_nid2peer(rpc->crpc_dest.nid);
1374 srpc_abort_rpc(rpc, -ENOMEM);
1378 srpc_queue_rpc(peer, rpc);
1380 spin_unlock(&rpc->crpc_lock);
1381 srpc_check_sends(peer, 0);
1382 spin_lock(&rpc->crpc_lock);
1388 srpc_send_reply (srpc_server_rpc_t *rpc)
1390 srpc_event_t *ev = &rpc->srpc_ev;
1391 srpc_msg_t *msg = &rpc->srpc_replymsg;
1392 srpc_buffer_t *buffer = rpc->srpc_reqstbuf;
1393 srpc_service_t *sv = rpc->srpc_service;
1397 LASSERT (buffer != NULL);
1398 rpyid = buffer->buf_msg.msg_body.reqst.rpyid;
1400 spin_lock(&sv->sv_lock);
1402 if (!sv->sv_shuttingdown &&
1403 sv->sv_id > SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1404 /* Repost buffer before replying since test client
1405 * might send me another RPC once it gets the reply */
1406 if (srpc_service_post_buffer(sv, buffer) != 0)
1407 CWARN ("Failed to repost %s buffer\n", sv->sv_name);
1408 rpc->srpc_reqstbuf = NULL;
1411 spin_unlock(&sv->sv_lock);
1415 ev->ev_type = SRPC_REPLY_SENT;
1417 msg->msg_magic = SRPC_MSG_MAGIC;
1418 msg->msg_version = SRPC_MSG_VERSION;
1419 msg->msg_type = srpc_service2reply(sv->sv_id);
1421 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, rpyid, msg,
1422 sizeof(*msg), LNET_MD_OP_PUT,
1423 rpc->srpc_peer, rpc->srpc_self,
1424 &rpc->srpc_replymdh, ev);
1426 ev->ev_fired = 1; /* no more event expected */
1430 /* when in kernel always called with LNET_LOCK() held, and in thread context */
1432 srpc_lnet_ev_handler (lnet_event_t *ev)
1434 srpc_event_t *rpcev = ev->md.user_ptr;
1435 srpc_client_rpc_t *crpc;
1436 srpc_server_rpc_t *srpc;
1437 srpc_buffer_t *buffer;
1440 srpc_msg_type_t type;
1443 LASSERT (!in_interrupt());
1445 if (ev->status != 0) {
1446 spin_lock(&srpc_data.rpc_glock);
1447 srpc_data.rpc_counters.errors++;
1448 spin_unlock(&srpc_data.rpc_glock);
1451 rpcev->ev_lnet = ev->type;
1453 switch (rpcev->ev_type) {
1456 case SRPC_REQUEST_SENT:
1457 if (ev->status == 0 && ev->type != LNET_EVENT_UNLINK) {
1458 spin_lock(&srpc_data.rpc_glock);
1459 srpc_data.rpc_counters.rpcs_sent++;
1460 spin_unlock(&srpc_data.rpc_glock);
1462 case SRPC_REPLY_RCVD:
1463 case SRPC_BULK_REQ_RCVD:
1464 crpc = rpcev->ev_data;
1466 LASSERT (rpcev == &crpc->crpc_reqstev ||
1467 rpcev == &crpc->crpc_replyev ||
1468 rpcev == &crpc->crpc_bulkev);
1470 spin_lock(&crpc->crpc_lock);
1472 LASSERT (rpcev->ev_fired == 0);
1473 rpcev->ev_fired = 1;
1474 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1475 -EINTR : ev->status;
1476 swi_schedule_workitem(&crpc->crpc_wi);
1478 spin_unlock(&crpc->crpc_lock);
1481 case SRPC_REQUEST_RCVD:
1482 sv = rpcev->ev_data;
1484 LASSERT (rpcev == &sv->sv_ev);
1486 spin_lock(&sv->sv_lock);
1488 LASSERT (ev->unlinked);
1489 LASSERT (ev->type == LNET_EVENT_PUT ||
1490 ev->type == LNET_EVENT_UNLINK);
1491 LASSERT (ev->type != LNET_EVENT_UNLINK ||
1492 sv->sv_shuttingdown);
1494 buffer = container_of(ev->md.start, srpc_buffer_t, buf_msg);
1495 buffer->buf_peer = ev->initiator;
1496 buffer->buf_self = ev->target.nid;
1498 sv->sv_nposted_msg--;
1499 LASSERT (sv->sv_nposted_msg >= 0);
1501 if (sv->sv_shuttingdown) {
1502 /* Leave buffer on sv->sv_posted_msgq since
1503 * srpc_finish_service needs to traverse it. */
1504 spin_unlock(&sv->sv_lock);
1508 list_del(&buffer->buf_list); /* from sv->sv_posted_msgq */
1509 msg = &buffer->buf_msg;
1510 type = srpc_service2request(sv->sv_id);
1512 if (ev->status != 0 || ev->mlength != sizeof(*msg) ||
1513 (msg->msg_type != type &&
1514 msg->msg_type != __swab32(type)) ||
1515 (msg->msg_magic != SRPC_MSG_MAGIC &&
1516 msg->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1517 CERROR ("Dropping RPC (%s) from %s: "
1518 "status %d mlength %d type %u magic %u.\n",
1519 sv->sv_name, libcfs_id2str(ev->initiator),
1520 ev->status, ev->mlength,
1521 msg->msg_type, msg->msg_magic);
1523 /* NB can't call srpc_service_recycle_buffer here since
1524 * it may call LNetM[DE]Attach. The invalid magic tells
1525 * srpc_handle_rpc to drop this RPC */
1529 if (!list_empty(&sv->sv_free_rpcq)) {
1530 srpc = list_entry(sv->sv_free_rpcq.next,
1531 srpc_server_rpc_t, srpc_list);
1532 list_del(&srpc->srpc_list);
1534 srpc_init_server_rpc(srpc, sv, buffer);
1535 list_add_tail(&srpc->srpc_list, &sv->sv_active_rpcq);
1536 srpc_schedule_server_rpc(srpc);
1538 list_add_tail(&buffer->buf_list, &sv->sv_blocked_msgq);
1541 spin_unlock(&sv->sv_lock);
1543 spin_lock(&srpc_data.rpc_glock);
1544 srpc_data.rpc_counters.rpcs_rcvd++;
1545 spin_unlock(&srpc_data.rpc_glock);
1548 case SRPC_BULK_GET_RPLD:
1549 LASSERT (ev->type == LNET_EVENT_SEND ||
1550 ev->type == LNET_EVENT_REPLY ||
1551 ev->type == LNET_EVENT_UNLINK);
1553 if (ev->type == LNET_EVENT_SEND && !ev->unlinked) {
1554 if (ev->status == 0)
1555 break; /* wait for the final LNET_EVENT_REPLY */
1557 fired_flag = 0; /* LNET_EVENT_REPLY may arrive
1558 (optimized GET case) */
1560 case SRPC_BULK_PUT_SENT:
1561 if (ev->status == 0 && ev->type != LNET_EVENT_UNLINK) {
1562 spin_lock(&srpc_data.rpc_glock);
1564 if (rpcev->ev_type == SRPC_BULK_GET_RPLD)
1565 srpc_data.rpc_counters.bulk_get += ev->mlength;
1567 srpc_data.rpc_counters.bulk_put += ev->mlength;
1569 spin_unlock(&srpc_data.rpc_glock);
1571 case SRPC_REPLY_SENT:
1572 srpc = rpcev->ev_data;
1573 sv = srpc->srpc_service;
1575 LASSERT (rpcev == &srpc->srpc_ev);
1577 spin_lock(&sv->sv_lock);
1579 rpcev->ev_fired = 1;
1581 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1582 -EINTR : ev->status;
1584 srpc_schedule_server_rpc(srpc);
1585 spin_unlock(&sv->sv_lock);
1595 srpc_check_event (int timeout)
1601 rc = LNetEQPoll(&srpc_data.rpc_lnet_eq, 1,
1602 timeout * 1000, &ev, &i);
1603 if (rc == 0) return 0;
1605 LASSERT (rc == -EOVERFLOW || rc == 1);
1607 /* We can't affort to miss any events... */
1608 if (rc == -EOVERFLOW) {
1609 CERROR ("Dropped an event!!!\n");
1613 srpc_lnet_ev_handler(&ev);
1628 s = getenv("SRPC_PEER_CREDITS");
1629 srpc_peer_credits = (s != NULL) ? atoi(s) : srpc_peer_credits;
1632 if (srpc_peer_credits <= 0) {
1633 CERROR("Peer credits must be positive: %d\n", srpc_peer_credits);
1637 memset(&srpc_data, 0, sizeof(struct smoketest_rpc));
1638 spin_lock_init(&srpc_data.rpc_glock);
1640 /* 1 second pause to avoid timestamp reuse */
1641 cfs_pause(cfs_time_seconds(1));
1642 srpc_data.rpc_matchbits = ((__u64) cfs_time_current_sec()) << 48;
1644 srpc_data.rpc_state = SRPC_STATE_NONE;
1646 LIBCFS_ALLOC(srpc_data.rpc_peers,
1647 sizeof(struct list_head) * SRPC_PEER_HASH_SIZE);
1648 if (srpc_data.rpc_peers == NULL) {
1649 CERROR ("Failed to alloc peer hash.\n");
1653 for (i = 0; i < SRPC_PEER_HASH_SIZE; i++)
1654 CFS_INIT_LIST_HEAD(&srpc_data.rpc_peers[i]);
1657 rc = LNetNIInit(LUSTRE_SRV_LNET_PID);
1659 if (the_lnet.ln_server_mode_flag)
1660 rc = LNetNIInit(LUSTRE_SRV_LNET_PID);
1662 rc = LNetNIInit(getpid() | LNET_PID_USERFLAG);
1665 CERROR ("LNetNIInit() has failed: %d\n", rc);
1666 LIBCFS_FREE(srpc_data.rpc_peers,
1667 sizeof(struct list_head) * SRPC_PEER_HASH_SIZE);
1671 srpc_data.rpc_state = SRPC_STATE_NI_INIT;
1673 LNetInvalidateHandle(&srpc_data.rpc_lnet_eq);
1675 rc = LNetEQAlloc(16, srpc_lnet_ev_handler, &srpc_data.rpc_lnet_eq);
1677 rc = LNetEQAlloc(10240, LNET_EQ_HANDLER_NONE, &srpc_data.rpc_lnet_eq);
1680 CERROR("LNetEQAlloc() has failed: %d\n", rc);
1684 rc = LNetSetLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1687 srpc_data.rpc_state = SRPC_STATE_EQ_INIT;
1693 srpc_data.rpc_state = SRPC_STATE_WI_INIT;
1701 srpc_data.rpc_state = SRPC_STATE_RUNNING;
1707 srpc_shutdown (void)
1713 state = srpc_data.rpc_state;
1714 srpc_data.rpc_state = SRPC_STATE_STOPPING;
1719 case SRPC_STATE_RUNNING:
1720 spin_lock(&srpc_data.rpc_glock);
1722 for (i = 0; i <= SRPC_SERVICE_MAX_ID; i++) {
1723 srpc_service_t *sv = srpc_data.rpc_services[i];
1725 LASSERTF (sv == NULL,
1726 "service not empty: id %d, name %s\n",
1730 spin_unlock(&srpc_data.rpc_glock);
1734 case SRPC_STATE_WI_INIT:
1737 case SRPC_STATE_EQ_INIT:
1738 rc = LNetClearLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1740 rc = LNetEQFree(srpc_data.rpc_lnet_eq);
1741 LASSERT (rc == 0); /* the EQ should have no user by now */
1743 case SRPC_STATE_NI_INIT:
1748 /* srpc_peer_t's are kept in hash until shutdown */
1749 for (i = 0; i < SRPC_PEER_HASH_SIZE; i++) {
1752 while (!list_empty(&srpc_data.rpc_peers[i])) {
1753 peer = list_entry(srpc_data.rpc_peers[i].next,
1754 srpc_peer_t, stp_list);
1755 list_del(&peer->stp_list);
1757 LASSERT (list_empty(&peer->stp_rpcq));
1758 LASSERT (list_empty(&peer->stp_ctl_rpcq));
1759 LASSERT (peer->stp_credits == srpc_peer_credits);
1761 LIBCFS_FREE(peer, sizeof(srpc_peer_t));
1765 LIBCFS_FREE(srpc_data.rpc_peers,
1766 sizeof(struct list_head) * SRPC_PEER_HASH_SIZE);