1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
5 * Author: Isaac Huang <isaac@clusterfs.com>
9 #define DEBUG_SUBSYSTEM S_LNET
11 #include <libcfs/kp30.h>
12 #include <libcfs/libcfs.h>
13 #include <lnet/lib-lnet.h>
27 #define SRPC_PEER_HASH_SIZE 101 /* # peer lists */
28 #define SRPC_PEER_CREDITS 16 /* >= most LND's default peer credit */
30 struct smoketest_rpc {
31 spinlock_t rpc_glock; /* global lock */
32 srpc_service_t *rpc_services[SRPC_SERVICE_MAX_ID + 1];
33 struct list_head *rpc_peers; /* hash table of known peers */
34 lnet_handle_eq_t rpc_lnet_eq; /* _the_ LNet event queue */
35 srpc_state_t rpc_state;
36 srpc_counters_t rpc_counters;
37 __u64 rpc_matchbits; /* matchbits counter */
41 int srpc_handle_rpc (swi_workitem_t *wi);
43 void srpc_get_counters (srpc_counters_t *cnt)
45 spin_lock(&srpc_data.rpc_glock);
46 *cnt = srpc_data.rpc_counters;
47 spin_unlock(&srpc_data.rpc_glock);
50 void srpc_set_counters (const srpc_counters_t *cnt)
52 spin_lock(&srpc_data.rpc_glock);
53 srpc_data.rpc_counters = *cnt;
54 spin_unlock(&srpc_data.rpc_glock);
58 srpc_add_bulk_page (srpc_bulk_t *bk, cfs_page_t *pg, int i)
60 LASSERT (i >= 0 && i < bk->bk_niov);
63 bk->bk_iovs[i].kiov_offset = 0;
64 bk->bk_iovs[i].kiov_page = pg;
65 bk->bk_iovs[i].kiov_len = CFS_PAGE_SIZE;
67 LASSERT (bk->bk_pages != NULL);
70 bk->bk_iovs[i].iov_len = CFS_PAGE_SIZE;
71 bk->bk_iovs[i].iov_base = cfs_page_address(pg);
77 srpc_free_bulk (srpc_bulk_t *bk)
84 LASSERT (bk->bk_pages != NULL);
87 for (i = 0; i < bk->bk_niov; i++) {
89 pg = bk->bk_iovs[i].kiov_page;
93 if (pg == NULL) break;
99 LIBCFS_FREE(bk->bk_pages, sizeof(cfs_page_t *) * bk->bk_niov);
101 LIBCFS_FREE(bk, offsetof(srpc_bulk_t, bk_iovs[bk->bk_niov]));
106 srpc_alloc_bulk (int npages, int sink)
112 LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
114 LIBCFS_ALLOC(bk, offsetof(srpc_bulk_t, bk_iovs[npages]));
116 CERROR ("Can't allocate descriptor for %d pages\n", npages);
120 memset(bk, 0, offsetof(srpc_bulk_t, bk_iovs[npages]));
122 bk->bk_niov = npages;
123 bk->bk_len = npages * CFS_PAGE_SIZE;
125 LIBCFS_ALLOC(pages, sizeof(cfs_page_t *) * npages);
127 LIBCFS_FREE(bk, offsetof(srpc_bulk_t, bk_iovs[npages]));
128 CERROR ("Can't allocate page array for %d pages\n", npages);
132 memset(pages, 0, sizeof(cfs_page_t *) * npages);
133 bk->bk_pages = pages;
138 for (i = 0; i < npages; i++) {
139 cfs_page_t *pg = cfs_alloc_page(CFS_ALLOC_STD);
142 CERROR ("Can't allocate page %d of %d\n", i, npages);
147 srpc_add_bulk_page(bk, pg, i);
154 static inline struct list_head *
155 srpc_nid2peerlist (lnet_nid_t nid)
157 unsigned int hash = ((unsigned int)nid) % SRPC_PEER_HASH_SIZE;
159 return &srpc_data.rpc_peers[hash];
162 static inline srpc_peer_t *
163 srpc_create_peer (lnet_nid_t nid)
167 LASSERT (nid != LNET_NID_ANY);
169 LIBCFS_ALLOC(peer, sizeof(srpc_peer_t));
171 CERROR ("Failed to allocate peer structure for %s\n",
172 libcfs_nid2str(nid));
176 memset(peer, 0, sizeof(srpc_peer_t));
178 peer->stp_credits = SRPC_PEER_CREDITS;
180 spin_lock_init(&peer->stp_lock);
181 CFS_INIT_LIST_HEAD(&peer->stp_rpcq);
182 CFS_INIT_LIST_HEAD(&peer->stp_ctl_rpcq);
187 srpc_find_peer_locked (lnet_nid_t nid)
189 struct list_head *peer_list = srpc_nid2peerlist(nid);
192 LASSERT (nid != LNET_NID_ANY);
194 list_for_each_entry (peer, peer_list, stp_list) {
195 if (peer->stp_nid == nid)
203 srpc_nid2peer (lnet_nid_t nid)
206 srpc_peer_t *new_peer;
208 spin_lock(&srpc_data.rpc_glock);
209 peer = srpc_find_peer_locked(nid);
210 spin_unlock(&srpc_data.rpc_glock);
215 new_peer = srpc_create_peer(nid);
217 spin_lock(&srpc_data.rpc_glock);
219 peer = srpc_find_peer_locked(nid);
221 spin_unlock(&srpc_data.rpc_glock);
222 if (new_peer != NULL)
223 LIBCFS_FREE(new_peer, sizeof(srpc_peer_t));
228 if (new_peer == NULL) {
229 spin_unlock(&srpc_data.rpc_glock);
233 list_add_tail(&new_peer->stp_list, srpc_nid2peerlist(nid));
234 spin_unlock(&srpc_data.rpc_glock);
243 spin_lock(&srpc_data.rpc_glock);
244 id = srpc_data.rpc_matchbits++;
245 spin_unlock(&srpc_data.rpc_glock);
250 srpc_init_server_rpc (srpc_server_rpc_t *rpc,
251 srpc_service_t *sv, srpc_buffer_t *buffer)
253 memset(rpc, 0, sizeof(*rpc));
254 swi_init_workitem(&rpc->srpc_wi, rpc, srpc_handle_rpc);
256 rpc->srpc_ev.ev_fired = 1; /* no event expected now */
258 rpc->srpc_service = sv;
259 rpc->srpc_reqstbuf = buffer;
260 rpc->srpc_peer = buffer->buf_peer;
261 rpc->srpc_self = buffer->buf_self;
262 rpc->srpc_replymdh = LNET_INVALID_HANDLE;
266 srpc_add_service (srpc_service_t *sv)
270 srpc_server_rpc_t *rpc;
272 LASSERT (sv->sv_concur > 0);
273 LASSERT (0 <= id && id <= SRPC_SERVICE_MAX_ID);
275 spin_lock(&srpc_data.rpc_glock);
277 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
279 if (srpc_data.rpc_services[id] != NULL) {
280 spin_unlock(&srpc_data.rpc_glock);
284 srpc_data.rpc_services[id] = sv;
285 spin_unlock(&srpc_data.rpc_glock);
288 sv->sv_nposted_msg = 0;
289 sv->sv_shuttingdown = 0;
290 spin_lock_init(&sv->sv_lock);
291 CFS_INIT_LIST_HEAD(&sv->sv_free_rpcq);
292 CFS_INIT_LIST_HEAD(&sv->sv_active_rpcq);
293 CFS_INIT_LIST_HEAD(&sv->sv_posted_msgq);
294 CFS_INIT_LIST_HEAD(&sv->sv_blocked_msgq);
296 sv->sv_ev.ev_data = sv;
297 sv->sv_ev.ev_type = SRPC_REQUEST_RCVD;
299 for (i = 0; i < sv->sv_concur; i++) {
300 LIBCFS_ALLOC(rpc, sizeof(*rpc));
301 if (rpc == NULL) goto enomem;
303 list_add(&rpc->srpc_list, &sv->sv_free_rpcq);
306 CDEBUG (D_NET, "Adding service: id %d, name %s, concurrency %d\n",
307 id, sv->sv_name, sv->sv_concur);
311 while (!list_empty(&sv->sv_free_rpcq)) {
312 rpc = list_entry(sv->sv_free_rpcq.next,
313 srpc_server_rpc_t, srpc_list);
314 list_del(&rpc->srpc_list);
315 LIBCFS_FREE(rpc, sizeof(*rpc));
318 spin_lock(&srpc_data.rpc_glock);
319 srpc_data.rpc_services[id] = NULL;
320 spin_unlock(&srpc_data.rpc_glock);
325 srpc_remove_service (srpc_service_t *sv)
329 spin_lock(&srpc_data.rpc_glock);
331 if (srpc_data.rpc_services[id] != sv) {
332 spin_unlock(&srpc_data.rpc_glock);
336 srpc_data.rpc_services[id] = NULL;
337 spin_unlock(&srpc_data.rpc_glock);
342 srpc_post_passive_rdma(int portal, __u64 matchbits, void *buf,
343 int len, int options, lnet_process_id_t peer,
344 lnet_handle_md_t *mdh, srpc_event_t *ev)
348 lnet_handle_me_t meh;
350 rc = LNetMEAttach(portal, peer, matchbits, 0,
351 LNET_UNLINK, LNET_INS_AFTER, &meh);
353 CERROR ("LNetMEAttach failed: %d\n", rc);
354 LASSERT (rc == -ENOMEM);
362 md.options = options;
363 md.eq_handle = srpc_data.rpc_lnet_eq;
365 rc = LNetMDAttach(meh, md, LNET_UNLINK, mdh);
367 CERROR ("LNetMDAttach failed: %d\n", rc);
368 LASSERT (rc == -ENOMEM);
370 rc = LNetMEUnlink(meh);
376 "Posted passive RDMA: peer %s, portal %d, matchbits "LPX64"\n",
377 libcfs_id2str(peer), portal, matchbits);
382 srpc_post_active_rdma(int portal, __u64 matchbits, void *buf, int len,
383 int options, lnet_process_id_t peer, lnet_nid_t self,
384 lnet_handle_md_t *mdh, srpc_event_t *ev)
392 md.eq_handle = srpc_data.rpc_lnet_eq;
393 md.threshold = ((options & LNET_MD_OP_GET) != 0) ? 2 : 1;
394 md.options = options & ~(LNET_MD_OP_PUT | LNET_MD_OP_GET);
396 rc = LNetMDBind(md, LNET_UNLINK, mdh);
398 CERROR ("LNetMDBind failed: %d\n", rc);
399 LASSERT (rc == -ENOMEM);
403 /* this is kind of an abuse of the LNET_MD_OP_{PUT,GET} options.
404 * they're only meaningful for MDs attached to an ME (i.e. passive
406 if ((options & LNET_MD_OP_PUT) != 0) {
407 rc = LNetPut(self, *mdh, LNET_NOACK_REQ, peer,
408 portal, matchbits, 0, 0);
410 LASSERT ((options & LNET_MD_OP_GET) != 0);
412 rc = LNetGet(self, *mdh, peer, portal, matchbits, 0);
416 CERROR ("LNet%s(%s, %d, "LPD64") failed: %d\n",
417 ((options & LNET_MD_OP_PUT) != 0) ? "Put" : "Get",
418 libcfs_id2str(peer), portal, matchbits, rc);
420 /* The forthcoming unlink event will complete this operation
421 * with failure, so fall through and return success here.
423 rc = LNetMDUnlink(*mdh);
427 "Posted active RDMA: peer %s, portal %u, matchbits "LPX64"\n",
428 libcfs_id2str(peer), portal, matchbits);
434 srpc_post_active_rqtbuf(lnet_process_id_t peer, int service, void *buf,
435 int len, lnet_handle_md_t *mdh, srpc_event_t *ev)
440 if (service > SRPC_FRAMEWORK_SERVICE_MAX_ID)
441 portal = SRPC_REQUEST_PORTAL;
443 portal = SRPC_FRAMEWORK_REQUEST_PORTAL;
445 rc = srpc_post_active_rdma(portal, service, buf, len,
446 LNET_MD_OP_PUT, peer,
447 LNET_NID_ANY, mdh, ev);
452 srpc_post_passive_rqtbuf(int service, void *buf, int len,
453 lnet_handle_md_t *mdh, srpc_event_t *ev)
457 lnet_process_id_t any = {.nid = LNET_NID_ANY,
458 .pid = LNET_PID_ANY};
460 if (service > SRPC_FRAMEWORK_SERVICE_MAX_ID)
461 portal = SRPC_REQUEST_PORTAL;
463 portal = SRPC_FRAMEWORK_REQUEST_PORTAL;
465 rc = srpc_post_passive_rdma(portal, service, buf, len,
466 LNET_MD_OP_PUT, any, mdh, ev);
471 srpc_service_post_buffer (srpc_service_t *sv, srpc_buffer_t *buf)
473 srpc_msg_t *msg = &buf->buf_msg;
476 LASSERT (!sv->sv_shuttingdown);
478 buf->buf_mdh = LNET_INVALID_HANDLE;
479 list_add(&buf->buf_list, &sv->sv_posted_msgq);
480 sv->sv_nposted_msg++;
481 spin_unlock(&sv->sv_lock);
483 rc = srpc_post_passive_rqtbuf(sv->sv_id, msg, sizeof(*msg),
484 &buf->buf_mdh, &sv->sv_ev);
486 /* At this point, a RPC (new or delayed) may have arrived in
487 * msg and its event handler has been called. So we must add
488 * buf to sv_posted_msgq _before_ dropping sv_lock */
490 spin_lock(&sv->sv_lock);
493 if (sv->sv_shuttingdown) {
494 spin_unlock(&sv->sv_lock);
496 /* srpc_shutdown_service might have tried to unlink me
497 * when my buf_mdh was still invalid */
498 LNetMDUnlink(buf->buf_mdh);
500 spin_lock(&sv->sv_lock);
505 sv->sv_nposted_msg--;
506 if (sv->sv_shuttingdown) return rc;
508 list_del(&buf->buf_list);
510 spin_unlock(&sv->sv_lock);
511 LIBCFS_FREE(buf, sizeof(*buf));
512 spin_lock(&sv->sv_lock);
517 srpc_service_add_buffers (srpc_service_t *sv, int nbuffer)
523 LASSERTF (nbuffer > 0,
524 "nbuffer must be positive: %d\n", nbuffer);
526 for (posted = 0; posted < nbuffer; posted++) {
527 LIBCFS_ALLOC(buf, sizeof(*buf));
528 if (buf == NULL) break;
530 spin_lock(&sv->sv_lock);
531 rc = srpc_service_post_buffer(sv, buf);
532 spin_unlock(&sv->sv_lock);
541 srpc_service_remove_buffers (srpc_service_t *sv, int nbuffer)
543 LASSERTF (nbuffer > 0,
544 "nbuffer must be positive: %d\n", nbuffer);
546 spin_lock(&sv->sv_lock);
548 LASSERT (sv->sv_nprune >= 0);
549 LASSERT (!sv->sv_shuttingdown);
551 sv->sv_nprune += nbuffer;
553 spin_unlock(&sv->sv_lock);
557 /* returns 1 if sv has finished, otherwise 0 */
559 srpc_finish_service (srpc_service_t *sv)
561 srpc_server_rpc_t *rpc;
564 spin_lock(&sv->sv_lock);
566 LASSERT (sv->sv_shuttingdown); /* srpc_shutdown_service called */
568 if (sv->sv_nposted_msg != 0 || !list_empty(&sv->sv_active_rpcq)) {
570 "waiting for %d posted buffers to unlink and "
571 "in-flight RPCs to die.\n",
574 if (!list_empty(&sv->sv_active_rpcq)) {
575 rpc = list_entry(sv->sv_active_rpcq.next,
576 srpc_server_rpc_t, srpc_list);
578 "Active RPC on shutdown: sv %s, peer %s, "
579 "wi %s scheduled %d running %d, "
580 "ev fired %d type %d status %d lnet %d\n",
581 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
582 swi_state2str(rpc->srpc_wi.wi_state),
583 rpc->srpc_wi.wi_scheduled,
584 rpc->srpc_wi.wi_running,
585 rpc->srpc_ev.ev_fired,
586 rpc->srpc_ev.ev_type,
587 rpc->srpc_ev.ev_status,
588 rpc->srpc_ev.ev_lnet);
591 spin_unlock(&sv->sv_lock);
595 spin_unlock(&sv->sv_lock); /* no lock needed from now on */
600 if (!list_empty(&sv->sv_posted_msgq))
601 q = &sv->sv_posted_msgq;
602 else if (!list_empty(&sv->sv_blocked_msgq))
603 q = &sv->sv_blocked_msgq;
607 buf = list_entry(q->next, srpc_buffer_t, buf_list);
608 list_del(&buf->buf_list);
610 LIBCFS_FREE(buf, sizeof(*buf));
613 while (!list_empty(&sv->sv_free_rpcq)) {
614 rpc = list_entry(sv->sv_free_rpcq.next,
615 srpc_server_rpc_t, srpc_list);
616 list_del(&rpc->srpc_list);
617 LIBCFS_FREE(rpc, sizeof(*rpc));
623 /* called with sv->sv_lock held */
625 srpc_service_recycle_buffer (srpc_service_t *sv, srpc_buffer_t *buf)
627 if (sv->sv_shuttingdown)
630 if (sv->sv_nprune == 0) {
631 if (srpc_service_post_buffer(sv, buf) != 0)
632 CWARN ("Failed to post %s buffer\n", sv->sv_name);
638 spin_unlock(&sv->sv_lock);
639 LIBCFS_FREE(buf, sizeof(*buf));
640 spin_lock(&sv->sv_lock);
644 srpc_shutdown_service (srpc_service_t *sv)
646 srpc_server_rpc_t *rpc;
649 spin_lock(&sv->sv_lock);
651 CDEBUG (D_NET, "Shutting down service: id %d, name %s\n",
652 sv->sv_id, sv->sv_name);
654 sv->sv_shuttingdown = 1; /* i.e. no new active RPC */
656 /* schedule in-flight RPCs to notice the shutdown */
657 list_for_each_entry (rpc, &sv->sv_active_rpcq, srpc_list) {
658 swi_schedule_workitem(&rpc->srpc_wi);
661 spin_unlock(&sv->sv_lock);
663 /* OK to traverse sv_posted_msgq without lock, since no one
664 * touches sv_posted_msgq now */
665 list_for_each_entry (buf, &sv->sv_posted_msgq, buf_list)
666 LNetMDUnlink(buf->buf_mdh);
672 srpc_send_request (srpc_client_rpc_t *rpc)
674 srpc_event_t *ev = &rpc->crpc_reqstev;
679 ev->ev_type = SRPC_REQUEST_SENT;
681 rc = srpc_post_active_rqtbuf(rpc->crpc_dest, rpc->crpc_service,
682 &rpc->crpc_reqstmsg, sizeof(srpc_msg_t),
683 &rpc->crpc_reqstmdh, ev);
685 LASSERT (rc == -ENOMEM);
686 ev->ev_fired = 1; /* no more event expected */
692 srpc_prepare_reply (srpc_client_rpc_t *rpc)
694 srpc_event_t *ev = &rpc->crpc_replyev;
695 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.rpyid;
700 ev->ev_type = SRPC_REPLY_RCVD;
702 *id = srpc_next_id();
704 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, *id,
705 &rpc->crpc_replymsg, sizeof(srpc_msg_t),
706 LNET_MD_OP_PUT, rpc->crpc_dest,
707 &rpc->crpc_replymdh, ev);
709 LASSERT (rc == -ENOMEM);
710 ev->ev_fired = 1; /* no more event expected */
716 srpc_prepare_bulk (srpc_client_rpc_t *rpc)
718 srpc_bulk_t *bk = &rpc->crpc_bulk;
719 srpc_event_t *ev = &rpc->crpc_bulkev;
720 __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.bulkid;
724 LASSERT (bk->bk_niov <= LNET_MAX_IOV);
726 if (bk->bk_niov == 0) return 0; /* nothing to do */
728 opt = bk->bk_sink ? LNET_MD_OP_PUT : LNET_MD_OP_GET;
732 opt |= LNET_MD_IOVEC;
737 ev->ev_type = SRPC_BULK_REQ_RCVD;
739 *id = srpc_next_id();
741 rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, *id,
742 &bk->bk_iovs[0], bk->bk_niov, opt,
743 rpc->crpc_dest, &bk->bk_mdh, ev);
745 LASSERT (rc == -ENOMEM);
746 ev->ev_fired = 1; /* no more event expected */
752 srpc_do_bulk (srpc_server_rpc_t *rpc)
754 srpc_event_t *ev = &rpc->srpc_ev;
755 srpc_bulk_t *bk = rpc->srpc_bulk;
756 __u64 id = rpc->srpc_reqstbuf->buf_msg.msg_body.reqst.bulkid;
760 LASSERT (bk != NULL);
762 opt = bk->bk_sink ? LNET_MD_OP_GET : LNET_MD_OP_PUT;
766 opt |= LNET_MD_IOVEC;
771 ev->ev_type = bk->bk_sink ? SRPC_BULK_GET_RPLD : SRPC_BULK_PUT_SENT;
773 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, id,
774 &bk->bk_iovs[0], bk->bk_niov, opt,
775 rpc->srpc_peer, rpc->srpc_self,
778 ev->ev_fired = 1; /* no more event expected */
782 /* called with srpc_service_t::sv_lock held */
784 srpc_schedule_server_rpc (srpc_server_rpc_t *rpc)
786 srpc_service_t *sv = rpc->srpc_service;
788 if (sv->sv_id > SRPC_FRAMEWORK_SERVICE_MAX_ID)
789 swi_schedule_workitem(&rpc->srpc_wi);
790 else /* framework RPCs are handled one by one */
791 swi_schedule_serial_workitem(&rpc->srpc_wi);
796 /* only called from srpc_handle_rpc */
798 srpc_server_rpc_done (srpc_server_rpc_t *rpc, int status)
800 srpc_service_t *sv = rpc->srpc_service;
801 srpc_buffer_t *buffer;
803 LASSERT (status != 0 || rpc->srpc_wi.wi_state == SWI_STATE_DONE);
805 rpc->srpc_status = status;
807 CDEBUG (status == 0 ? D_NET : D_NETERROR,
808 "Server RPC done: service %s, peer %s, status %s:%d\n",
809 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
810 swi_state2str(rpc->srpc_wi.wi_state), status);
813 spin_lock(&srpc_data.rpc_glock);
814 srpc_data.rpc_counters.rpcs_dropped++;
815 spin_unlock(&srpc_data.rpc_glock);
818 if (rpc->srpc_done != NULL)
819 (*rpc->srpc_done) (rpc);
820 LASSERT (rpc->srpc_bulk == NULL);
822 spin_lock(&sv->sv_lock);
824 if (rpc->srpc_reqstbuf != NULL) {
825 /* NB might drop sv_lock in srpc_service_recycle_buffer, but
826 * sv won't go away for sv_active_rpcq must not be empty */
827 srpc_service_recycle_buffer(sv, rpc->srpc_reqstbuf);
828 rpc->srpc_reqstbuf = NULL;
831 list_del(&rpc->srpc_list); /* from sv->sv_active_rpcq */
834 * No one can schedule me now since:
835 * - I'm not on sv_active_rpcq.
836 * - all LNet events have been fired.
837 * Cancel pending schedules and prevent future schedule attempts:
839 LASSERT (rpc->srpc_ev.ev_fired);
840 swi_kill_workitem(&rpc->srpc_wi);
842 if (!sv->sv_shuttingdown && !list_empty(&sv->sv_blocked_msgq)) {
843 buffer = list_entry(sv->sv_blocked_msgq.next,
844 srpc_buffer_t, buf_list);
845 list_del(&buffer->buf_list);
847 srpc_init_server_rpc(rpc, sv, buffer);
848 list_add_tail(&rpc->srpc_list, &sv->sv_active_rpcq);
849 srpc_schedule_server_rpc(rpc);
851 list_add(&rpc->srpc_list, &sv->sv_free_rpcq);
854 spin_unlock(&sv->sv_lock);
858 /* handles an incoming RPC */
860 srpc_handle_rpc (swi_workitem_t *wi)
862 srpc_server_rpc_t *rpc = wi->wi_data;
863 srpc_service_t *sv = rpc->srpc_service;
864 srpc_event_t *ev = &rpc->srpc_ev;
867 LASSERT (wi == &rpc->srpc_wi);
869 spin_lock(&sv->sv_lock);
871 if (sv->sv_shuttingdown) {
872 spin_unlock(&sv->sv_lock);
874 if (rpc->srpc_bulk != NULL)
875 LNetMDUnlink(rpc->srpc_bulk->bk_mdh);
876 LNetMDUnlink(rpc->srpc_replymdh);
878 if (ev->ev_fired) { /* no more event, OK to finish */
879 srpc_server_rpc_done(rpc, -ESHUTDOWN);
885 spin_unlock(&sv->sv_lock);
887 switch (wi->wi_state) {
890 case SWI_STATE_NEWBORN: {
892 srpc_generic_reply_t *reply;
894 msg = &rpc->srpc_reqstbuf->buf_msg;
895 reply = &rpc->srpc_replymsg.msg_body.reply;
897 if (msg->msg_version != SRPC_MSG_VERSION &&
898 msg->msg_version != __swab32(SRPC_MSG_VERSION)) {
899 CWARN ("Version mismatch: %u, %u expected, from %s\n",
900 msg->msg_version, SRPC_MSG_VERSION,
901 libcfs_id2str(rpc->srpc_peer));
902 reply->status = EPROTO;
905 rc = (*sv->sv_handler) (rpc);
906 LASSERT (reply->status == 0 || !rpc->srpc_bulk);
910 srpc_server_rpc_done(rpc, rc);
914 wi->wi_state = SWI_STATE_BULK_STARTED;
916 if (rpc->srpc_bulk != NULL) {
917 rc = srpc_do_bulk(rpc);
919 return 0; /* wait for bulk */
921 LASSERT (ev->ev_fired);
925 case SWI_STATE_BULK_STARTED:
926 LASSERT (rpc->srpc_bulk == NULL || ev->ev_fired);
928 if (rpc->srpc_bulk != NULL) {
931 if (sv->sv_bulk_ready != NULL)
932 rc = (*sv->sv_bulk_ready) (rpc, rc);
935 srpc_server_rpc_done(rpc, rc);
940 wi->wi_state = SWI_STATE_REPLY_SUBMITTED;
941 rc = srpc_send_reply(rpc);
943 return 0; /* wait for reply */
944 srpc_server_rpc_done(rpc, rc);
947 case SWI_STATE_REPLY_SUBMITTED:
948 LASSERT (ev->ev_fired);
950 wi->wi_state = SWI_STATE_DONE;
951 srpc_server_rpc_done(rpc, ev->ev_status);
959 srpc_client_rpc_expired (void *data)
961 srpc_client_rpc_t *rpc = data;
963 CWARN ("Client RPC expired: service %d, peer %s, timeout %d.\n",
964 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
967 spin_lock(&rpc->crpc_lock);
969 rpc->crpc_timeout = 0;
970 srpc_abort_rpc(rpc, -ETIMEDOUT);
972 spin_unlock(&rpc->crpc_lock);
974 spin_lock(&srpc_data.rpc_glock);
975 srpc_data.rpc_counters.rpcs_expired++;
976 spin_unlock(&srpc_data.rpc_glock);
981 srpc_add_client_rpc_timer (srpc_client_rpc_t *rpc)
983 stt_timer_t *timer = &rpc->crpc_timer;
985 if (rpc->crpc_timeout == 0) return;
987 CFS_INIT_LIST_HEAD(&timer->stt_list);
988 timer->stt_data = rpc;
989 timer->stt_func = srpc_client_rpc_expired;
990 timer->stt_expires = cfs_time_add(rpc->crpc_timeout,
991 cfs_time_current_sec());
992 stt_add_timer(timer);
997 * Called with rpc->crpc_lock held.
999 * Upon exit the RPC expiry timer is not queued and the handler is not
1000 * running on any CPU. */
1002 srpc_del_client_rpc_timer (srpc_client_rpc_t *rpc)
1004 /* timer not planted or already exploded */
1005 if (rpc->crpc_timeout == 0) return;
1007 /* timer sucessfully defused */
1008 if (stt_del_timer(&rpc->crpc_timer)) return;
1011 /* timer detonated, wait for it to explode */
1012 while (rpc->crpc_timeout != 0) {
1013 spin_unlock(&rpc->crpc_lock);
1017 spin_lock(&rpc->crpc_lock);
1020 LBUG(); /* impossible in single-threaded runtime */
1026 srpc_check_sends (srpc_peer_t *peer, int credits)
1028 struct list_head *q;
1029 srpc_client_rpc_t *rpc;
1031 LASSERT (credits >= 0);
1032 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
1034 spin_lock(&peer->stp_lock);
1035 peer->stp_credits += credits;
1037 while (peer->stp_credits) {
1038 if (!list_empty(&peer->stp_ctl_rpcq))
1039 q = &peer->stp_ctl_rpcq;
1040 else if (!list_empty(&peer->stp_rpcq))
1041 q = &peer->stp_rpcq;
1045 peer->stp_credits--;
1047 rpc = list_entry(q->next, srpc_client_rpc_t, crpc_privl);
1048 list_del_init(&rpc->crpc_privl);
1049 srpc_client_rpc_decref(rpc); /* --ref for peer->*rpcq */
1051 swi_schedule_workitem(&rpc->crpc_wi);
1054 spin_unlock(&peer->stp_lock);
1059 srpc_client_rpc_done (srpc_client_rpc_t *rpc, int status)
1061 swi_workitem_t *wi = &rpc->crpc_wi;
1062 srpc_peer_t *peer = rpc->crpc_peer;
1064 LASSERT (status != 0 || wi->wi_state == SWI_STATE_DONE);
1066 spin_lock(&rpc->crpc_lock);
1068 rpc->crpc_closed = 1;
1069 if (rpc->crpc_status == 0)
1070 rpc->crpc_status = status;
1072 srpc_del_client_rpc_timer(rpc);
1074 CDEBUG ((status == 0) ? D_NET : D_NETERROR,
1075 "Client RPC done: service %d, peer %s, status %s:%d:%d\n",
1076 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1077 swi_state2str(wi->wi_state), rpc->crpc_aborted, status);
1080 * No one can schedule me now since:
1081 * - RPC timer has been defused.
1082 * - all LNet events have been fired.
1083 * - crpc_closed has been set, preventing srpc_abort_rpc from
1085 * Cancel pending schedules and prevent future schedule attempts:
1087 LASSERT (!srpc_event_pending(rpc));
1088 swi_kill_workitem(wi);
1090 spin_unlock(&rpc->crpc_lock);
1092 (*rpc->crpc_done) (rpc);
1095 srpc_check_sends(peer, 1);
1099 /* sends an outgoing RPC */
1101 srpc_send_rpc (swi_workitem_t *wi)
1104 srpc_client_rpc_t *rpc = wi->wi_data;
1105 srpc_msg_t *reply = &rpc->crpc_replymsg;
1106 int do_bulk = rpc->crpc_bulk.bk_niov > 0;
1108 LASSERT (rpc != NULL);
1109 LASSERT (wi == &rpc->crpc_wi);
1111 spin_lock(&rpc->crpc_lock);
1113 if (rpc->crpc_aborted) {
1114 spin_unlock(&rpc->crpc_lock);
1118 spin_unlock(&rpc->crpc_lock);
1120 switch (wi->wi_state) {
1123 case SWI_STATE_NEWBORN:
1124 LASSERT (!srpc_event_pending(rpc));
1126 rc = srpc_prepare_reply(rpc);
1128 srpc_client_rpc_done(rpc, rc);
1132 rc = srpc_prepare_bulk(rpc);
1135 wi->wi_state = SWI_STATE_REQUEST_SUBMITTED;
1136 rc = srpc_send_request(rpc);
1139 case SWI_STATE_REQUEST_SUBMITTED:
1140 /* CAVEAT EMPTOR: rqtev, rpyev, and bulkev may come in any
1141 * order; however, they're processed in a strict order:
1142 * rqt, rpy, and bulk. */
1143 if (!rpc->crpc_reqstev.ev_fired) break;
1145 rc = rpc->crpc_reqstev.ev_status;
1148 wi->wi_state = SWI_STATE_REQUEST_SENT;
1149 /* perhaps more events, fall thru */
1150 case SWI_STATE_REQUEST_SENT: {
1151 srpc_msg_type_t type = srpc_service2reply(rpc->crpc_service);
1153 if (!rpc->crpc_replyev.ev_fired) break;
1155 rc = rpc->crpc_replyev.ev_status;
1158 if ((reply->msg_type != type &&
1159 reply->msg_type != __swab32(type)) ||
1160 (reply->msg_magic != SRPC_MSG_MAGIC &&
1161 reply->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1162 CWARN ("Bad message from %s: type %u (%d expected),"
1163 " magic %u (%d expected).\n",
1164 libcfs_id2str(rpc->crpc_dest),
1165 reply->msg_type, type,
1166 reply->msg_magic, SRPC_MSG_MAGIC);
1171 if (do_bulk && reply->msg_body.reply.status != 0) {
1172 CWARN ("Remote error %d at %s, unlink bulk buffer in "
1173 "case peer didn't initiate bulk transfer\n",
1174 reply->msg_body.reply.status,
1175 libcfs_id2str(rpc->crpc_dest));
1176 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1179 wi->wi_state = SWI_STATE_REPLY_RECEIVED;
1181 case SWI_STATE_REPLY_RECEIVED:
1182 if (do_bulk && !rpc->crpc_bulkev.ev_fired) break;
1184 rc = do_bulk ? rpc->crpc_bulkev.ev_status : 0;
1186 /* Bulk buffer was unlinked due to remote error. Clear error
1187 * since reply buffer still contains valid data.
1188 * NB rpc->crpc_done shouldn't look into bulk data in case of
1190 if (do_bulk && rpc->crpc_bulkev.ev_lnet == LNET_EVENT_UNLINK &&
1191 rpc->crpc_status == 0 && reply->msg_body.reply.status != 0)
1194 wi->wi_state = SWI_STATE_DONE;
1195 srpc_client_rpc_done(rpc, rc);
1200 spin_lock(&rpc->crpc_lock);
1201 srpc_abort_rpc(rpc, rc);
1202 spin_unlock(&rpc->crpc_lock);
1206 if (rpc->crpc_aborted) {
1207 LNetMDUnlink(rpc->crpc_reqstmdh);
1208 LNetMDUnlink(rpc->crpc_replymdh);
1209 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1211 if (!srpc_event_pending(rpc)) {
1212 srpc_client_rpc_done(rpc, -EINTR);
1220 srpc_create_client_rpc (lnet_process_id_t peer, int service,
1221 int nbulkiov, int bulklen,
1222 void (*rpc_done)(srpc_client_rpc_t *),
1223 void (*rpc_fini)(srpc_client_rpc_t *), void *priv)
1225 srpc_client_rpc_t *rpc;
1227 LIBCFS_ALLOC(rpc, offsetof(srpc_client_rpc_t,
1228 crpc_bulk.bk_iovs[nbulkiov]));
1232 srpc_init_client_rpc(rpc, peer, service, nbulkiov,
1233 bulklen, rpc_done, rpc_fini, priv);
1237 /* called with rpc->crpc_lock held */
1239 srpc_queue_rpc (srpc_peer_t *peer, srpc_client_rpc_t *rpc)
1241 int service = rpc->crpc_service;
1243 LASSERT (peer->stp_nid == rpc->crpc_dest.nid);
1244 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
1246 rpc->crpc_peer = peer;
1248 spin_lock(&peer->stp_lock);
1250 /* Framework RPCs that alter session state shall take precedence
1251 * over test RPCs and framework query RPCs */
1252 if (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID &&
1253 service != SRPC_SERVICE_DEBUG &&
1254 service != SRPC_SERVICE_QUERY_STAT)
1255 list_add_tail(&rpc->crpc_privl, &peer->stp_ctl_rpcq);
1257 list_add_tail(&rpc->crpc_privl, &peer->stp_rpcq);
1259 srpc_client_rpc_addref(rpc); /* ++ref for peer->*rpcq */
1260 spin_unlock(&peer->stp_lock);
1264 /* called with rpc->crpc_lock held */
1266 srpc_abort_rpc (srpc_client_rpc_t *rpc, int why)
1268 srpc_peer_t *peer = rpc->crpc_peer;
1272 if (rpc->crpc_aborted || /* already aborted */
1273 rpc->crpc_closed) /* callback imminent */
1277 "Aborting RPC: service %d, peer %s, state %s, why %d\n",
1278 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1279 swi_state2str(rpc->crpc_wi.wi_state), why);
1281 rpc->crpc_aborted = 1;
1282 rpc->crpc_status = why;
1285 spin_lock(&peer->stp_lock);
1287 if (!list_empty(&rpc->crpc_privl)) { /* still queued */
1288 list_del_init(&rpc->crpc_privl);
1289 srpc_client_rpc_decref(rpc); /* --ref for peer->*rpcq */
1290 rpc->crpc_peer = NULL; /* no credit taken */
1293 spin_unlock(&peer->stp_lock);
1296 swi_schedule_workitem(&rpc->crpc_wi);
1300 /* called with rpc->crpc_lock held */
1302 srpc_post_rpc (srpc_client_rpc_t *rpc)
1306 LASSERT (!rpc->crpc_aborted);
1307 LASSERT (rpc->crpc_peer == NULL);
1308 LASSERT (srpc_data.rpc_state == SRPC_STATE_RUNNING);
1309 LASSERT ((rpc->crpc_bulk.bk_len & ~CFS_PAGE_MASK) == 0);
1311 CDEBUG (D_NET, "Posting RPC: peer %s, service %d, timeout %d\n",
1312 libcfs_id2str(rpc->crpc_dest), rpc->crpc_service,
1315 srpc_add_client_rpc_timer(rpc);
1317 peer = srpc_nid2peer(rpc->crpc_dest.nid);
1319 srpc_abort_rpc(rpc, -ENOMEM);
1323 srpc_queue_rpc(peer, rpc);
1325 spin_unlock(&rpc->crpc_lock);
1326 srpc_check_sends(peer, 0);
1327 spin_lock(&rpc->crpc_lock);
1333 srpc_send_reply (srpc_server_rpc_t *rpc)
1335 srpc_event_t *ev = &rpc->srpc_ev;
1336 srpc_msg_t *msg = &rpc->srpc_replymsg;
1337 srpc_buffer_t *buffer = rpc->srpc_reqstbuf;
1338 srpc_service_t *sv = rpc->srpc_service;
1342 LASSERT (buffer != NULL);
1343 rpyid = buffer->buf_msg.msg_body.reqst.rpyid;
1345 spin_lock(&sv->sv_lock);
1347 if (!sv->sv_shuttingdown &&
1348 sv->sv_id > SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1349 /* Repost buffer before replying since test client
1350 * might send me another RPC once it gets the reply */
1351 if (srpc_service_post_buffer(sv, buffer) != 0)
1352 CWARN ("Failed to repost %s buffer\n", sv->sv_name);
1353 rpc->srpc_reqstbuf = NULL;
1356 spin_unlock(&sv->sv_lock);
1360 ev->ev_type = SRPC_REPLY_SENT;
1362 msg->msg_magic = SRPC_MSG_MAGIC;
1363 msg->msg_version = SRPC_MSG_VERSION;
1364 msg->msg_type = srpc_service2reply(sv->sv_id);
1366 rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, rpyid, msg,
1367 sizeof(*msg), LNET_MD_OP_PUT,
1368 rpc->srpc_peer, rpc->srpc_self,
1369 &rpc->srpc_replymdh, ev);
1371 ev->ev_fired = 1; /* no more event expected */
1375 /* always called with LNET_LOCK() held, and in thread context */
1377 srpc_lnet_ev_handler (lnet_event_t *ev)
1379 srpc_event_t *rpcev = ev->md.user_ptr;
1380 srpc_client_rpc_t *crpc;
1381 srpc_server_rpc_t *srpc;
1382 srpc_buffer_t *buffer;
1385 srpc_msg_type_t type;
1387 LASSERT (!in_interrupt());
1389 if (ev->status != 0) {
1390 spin_lock(&srpc_data.rpc_glock);
1391 srpc_data.rpc_counters.errors++;
1392 spin_unlock(&srpc_data.rpc_glock);
1395 rpcev->ev_lnet = ev->type;
1397 switch (rpcev->ev_type) {
1400 case SRPC_REQUEST_SENT:
1401 if (ev->status == 0 && ev->type != LNET_EVENT_UNLINK) {
1402 spin_lock(&srpc_data.rpc_glock);
1403 srpc_data.rpc_counters.rpcs_sent++;
1404 spin_unlock(&srpc_data.rpc_glock);
1406 case SRPC_REPLY_RCVD:
1407 case SRPC_BULK_REQ_RCVD:
1408 crpc = rpcev->ev_data;
1410 LASSERT (rpcev == &crpc->crpc_reqstev ||
1411 rpcev == &crpc->crpc_replyev ||
1412 rpcev == &crpc->crpc_bulkev);
1414 spin_lock(&crpc->crpc_lock);
1416 LASSERT (rpcev->ev_fired == 0);
1417 rpcev->ev_fired = 1;
1418 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1419 -EINTR : ev->status;
1420 swi_schedule_workitem(&crpc->crpc_wi);
1422 spin_unlock(&crpc->crpc_lock);
1425 case SRPC_REQUEST_RCVD:
1426 sv = rpcev->ev_data;
1428 LASSERT (rpcev == &sv->sv_ev);
1430 spin_lock(&sv->sv_lock);
1432 LASSERT (ev->unlinked);
1433 LASSERT (ev->type == LNET_EVENT_PUT ||
1434 ev->type == LNET_EVENT_UNLINK);
1435 LASSERT (ev->type != LNET_EVENT_UNLINK ||
1436 sv->sv_shuttingdown);
1438 buffer = container_of(ev->md.start, srpc_buffer_t, buf_msg);
1439 buffer->buf_peer = ev->initiator;
1440 buffer->buf_self = ev->target.nid;
1442 sv->sv_nposted_msg--;
1443 LASSERT (sv->sv_nposted_msg >= 0);
1445 if (sv->sv_shuttingdown) {
1446 /* Leave buffer on sv->sv_posted_msgq since
1447 * srpc_finish_service needs to traverse it. */
1448 spin_unlock(&sv->sv_lock);
1452 list_del(&buffer->buf_list); /* from sv->sv_posted_msgq */
1453 msg = &buffer->buf_msg;
1454 type = srpc_service2request(sv->sv_id);
1456 if (ev->status != 0 || ev->mlength != sizeof(*msg) ||
1457 (msg->msg_type != type &&
1458 msg->msg_type != __swab32(type)) ||
1459 (msg->msg_magic != SRPC_MSG_MAGIC &&
1460 msg->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1461 CERROR ("Dropping RPC (%s) from %s: "
1462 "status %d mlength %d type %u magic %u.\n",
1463 sv->sv_name, libcfs_id2str(ev->initiator),
1464 ev->status, ev->mlength,
1465 msg->msg_type, msg->msg_magic);
1467 /* NB might drop sv_lock in srpc_service_recycle_buffer,
1468 * sv_nposted_msg++ as an implicit reference to prevent
1469 * sv from disappearing under me */
1470 sv->sv_nposted_msg++;
1471 srpc_service_recycle_buffer(sv, buffer);
1472 sv->sv_nposted_msg--;
1473 spin_unlock(&sv->sv_lock);
1475 if (ev->status == 0) { /* status!=0 counted already */
1476 spin_lock(&srpc_data.rpc_glock);
1477 srpc_data.rpc_counters.errors++;
1478 spin_unlock(&srpc_data.rpc_glock);
1483 if (!list_empty(&sv->sv_free_rpcq)) {
1484 srpc = list_entry(sv->sv_free_rpcq.next,
1485 srpc_server_rpc_t, srpc_list);
1486 list_del(&srpc->srpc_list);
1488 srpc_init_server_rpc(srpc, sv, buffer);
1489 list_add_tail(&srpc->srpc_list, &sv->sv_active_rpcq);
1490 srpc_schedule_server_rpc(srpc);
1492 list_add_tail(&buffer->buf_list, &sv->sv_blocked_msgq);
1495 spin_unlock(&sv->sv_lock);
1497 spin_lock(&srpc_data.rpc_glock);
1498 srpc_data.rpc_counters.rpcs_rcvd++;
1499 spin_unlock(&srpc_data.rpc_glock);
1502 case SRPC_BULK_GET_RPLD:
1503 LASSERT (ev->type == LNET_EVENT_SEND ||
1504 ev->type == LNET_EVENT_REPLY ||
1505 ev->type == LNET_EVENT_UNLINK);
1507 if (ev->type == LNET_EVENT_SEND &&
1508 ev->status == 0 && !ev->unlinked)
1509 break; /* wait for the final LNET_EVENT_REPLY */
1511 case SRPC_BULK_PUT_SENT:
1512 if (ev->status == 0 && ev->type != LNET_EVENT_UNLINK) {
1513 spin_lock(&srpc_data.rpc_glock);
1515 if (rpcev->ev_type == SRPC_BULK_GET_RPLD)
1516 srpc_data.rpc_counters.bulk_get += ev->mlength;
1518 srpc_data.rpc_counters.bulk_put += ev->mlength;
1520 spin_unlock(&srpc_data.rpc_glock);
1522 case SRPC_REPLY_SENT:
1523 srpc = rpcev->ev_data;
1524 sv = srpc->srpc_service;
1526 LASSERT (rpcev == &srpc->srpc_ev);
1528 spin_lock(&sv->sv_lock);
1529 rpcev->ev_fired = 1;
1530 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1531 -EINTR : ev->status;
1532 srpc_schedule_server_rpc(srpc);
1533 spin_unlock(&sv->sv_lock);
1543 srpc_check_event (int timeout)
1549 rc = LNetEQPoll(&srpc_data.rpc_lnet_eq, 1,
1550 timeout * 1000, &ev, &i);
1554 LASSERT (rc == -EOVERFLOW || rc == 1);
1556 /* We can't affort to miss any events... */
1557 if (rc == -EOVERFLOW) {
1558 CERROR ("Dropped an event!!!\n");
1562 srpc_lnet_ev_handler(&ev);
1574 memset(&srpc_data, 0, sizeof(struct smoketest_rpc));
1575 spin_lock_init(&srpc_data.rpc_glock);
1577 /* 1 second pause to avoid timestamp reuse */
1578 cfs_pause(cfs_time_seconds(1));
1579 srpc_data.rpc_matchbits = ((__u64) cfs_time_current_sec()) << 48;
1581 srpc_data.rpc_state = SRPC_STATE_NONE;
1583 LIBCFS_ALLOC(srpc_data.rpc_peers,
1584 sizeof(struct list_head) * SRPC_PEER_HASH_SIZE);
1585 if (srpc_data.rpc_peers == NULL) {
1586 CERROR ("Failed to alloc peer hash.\n");
1590 for (i = 0; i < SRPC_PEER_HASH_SIZE; i++)
1591 CFS_INIT_LIST_HEAD(&srpc_data.rpc_peers[i]);
1594 rc = LNetNIInit(LUSTRE_SRV_LNET_PID);
1596 rc = LNetNIInit(getpid());
1599 CERROR ("LNetNIInit() has failed: %d\n", rc);
1600 LIBCFS_FREE(srpc_data.rpc_peers,
1601 sizeof(struct list_head) * SRPC_PEER_HASH_SIZE);
1605 srpc_data.rpc_state = SRPC_STATE_NI_INIT;
1607 srpc_data.rpc_lnet_eq = LNET_EQ_NONE;
1609 rc = LNetEQAlloc(16, srpc_lnet_ev_handler, &srpc_data.rpc_lnet_eq);
1611 rc = LNetEQAlloc(10240, LNET_EQ_HANDLER_NONE, &srpc_data.rpc_lnet_eq);
1614 CERROR("LNetEQAlloc() has failed: %d\n", rc);
1618 rc = LNetSetLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1621 srpc_data.rpc_state = SRPC_STATE_EQ_INIT;
1627 srpc_data.rpc_state = SRPC_STATE_WI_INIT;
1635 srpc_data.rpc_state = SRPC_STATE_RUNNING;
1641 srpc_shutdown (void)
1647 state = srpc_data.rpc_state;
1648 srpc_data.rpc_state = SRPC_STATE_STOPPING;
1653 case SRPC_STATE_RUNNING:
1654 spin_lock(&srpc_data.rpc_glock);
1656 for (i = 0; i <= SRPC_SERVICE_MAX_ID; i++) {
1657 srpc_service_t *sv = srpc_data.rpc_services[i];
1659 LASSERTF (sv == NULL,
1660 "service not empty: id %d, name %s\n",
1664 spin_unlock(&srpc_data.rpc_glock);
1668 case SRPC_STATE_WI_INIT:
1671 case SRPC_STATE_EQ_INIT:
1672 rc = LNetClearLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1674 rc = LNetEQFree(srpc_data.rpc_lnet_eq);
1675 LASSERT (rc == 0); /* the EQ should have no user by now */
1677 case SRPC_STATE_NI_INIT:
1682 /* srpc_peer_t's are kept in hash until shutdown */
1683 for (i = 0; i < SRPC_PEER_HASH_SIZE; i++) {
1686 while (!list_empty(&srpc_data.rpc_peers[i])) {
1687 peer = list_entry(srpc_data.rpc_peers[i].next,
1688 srpc_peer_t, stp_list);
1689 list_del(&peer->stp_list);
1691 LASSERT (list_empty(&peer->stp_rpcq));
1692 LASSERT (list_empty(&peer->stp_ctl_rpcq));
1693 LASSERT (peer->stp_credits == SRPC_PEER_CREDITS);
1695 LIBCFS_FREE(peer, sizeof(srpc_peer_t));
1699 LIBCFS_FREE(srpc_data.rpc_peers,
1700 sizeof(struct list_head) * SRPC_PEER_HASH_SIZE);