drivers/staging/lustre/lnet/selftest/rpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lnet/selftest/rpc.c
  37  *
  38  * Author: Isaac Huang <isaac@clusterfs.com>
  39  *
  40  * 2012-05-13: Liang Zhen <liang@whamcloud.com>
  41  * - percpt data for service to improve smp performance
  42  * - code cleanup
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_LNET
  46
  47 #include "selftest.h"
  48
  49 typedef enum {
  50         SRPC_STATE_NONE,
  51         SRPC_STATE_NI_INIT,
  52         SRPC_STATE_EQ_INIT,
  53         SRPC_STATE_RUNNING,
  54         SRPC_STATE_STOPPING,
  55 } srpc_state_t;
  56
  57 static struct smoketest_rpc {
  58         spinlock_t       rpc_glock;     /* global lock */
  59         srpc_service_t  *rpc_services[SRPC_SERVICE_MAX_ID + 1];
  60         lnet_handle_eq_t rpc_lnet_eq;   /* _the_ LNet event queue */
  61         srpc_state_t     rpc_state;
  62         srpc_counters_t  rpc_counters;
  63         __u64            rpc_matchbits; /* matchbits counter */
  64 } srpc_data;
  65
  66 static inline int
  67 srpc_serv_portal(int svc_id)
  68 {
  69         return svc_id < SRPC_FRAMEWORK_SERVICE_MAX_ID ?
  70                SRPC_FRAMEWORK_REQUEST_PORTAL : SRPC_REQUEST_PORTAL;
  71 }
  72
  73 /* forward ref's */
  74 int srpc_handle_rpc(swi_workitem_t *wi);
  75
  76 void srpc_get_counters(srpc_counters_t *cnt)
  77 {
  78         spin_lock(&srpc_data.rpc_glock);
  79         *cnt = srpc_data.rpc_counters;
  80         spin_unlock(&srpc_data.rpc_glock);
  81 }
  82
  83 void srpc_set_counters(const srpc_counters_t *cnt)
  84 {
  85         spin_lock(&srpc_data.rpc_glock);
  86         srpc_data.rpc_counters = *cnt;
  87         spin_unlock(&srpc_data.rpc_glock);
  88 }
  89
  90 static int
  91 srpc_add_bulk_page(srpc_bulk_t *bk, struct page *pg, int i, int nob)
  92 {
  93         nob = min_t(int, nob, PAGE_SIZE);
  94
  95         LASSERT(nob > 0);
  96         LASSERT(i >= 0 && i < bk->bk_niov);
  97
  98         bk->bk_iovs[i].kiov_offset = 0;
  99         bk->bk_iovs[i].kiov_page = pg;
 100         bk->bk_iovs[i].kiov_len = nob;
 101         return nob;
 102 }
 103
 104 void
 105 srpc_free_bulk(srpc_bulk_t *bk)
 106 {
 107         int i;
 108         struct page *pg;
 109
 110         LASSERT(bk);
 111
 112         for (i = 0; i < bk->bk_niov; i++) {
 113                 pg = bk->bk_iovs[i].kiov_page;
 114                 if (!pg)
 115                         break;
 116
 117                 __free_page(pg);
 118         }
 119
 120         LIBCFS_FREE(bk, offsetof(srpc_bulk_t, bk_iovs[bk->bk_niov]));
 121 }
 122
 123 srpc_bulk_t *
 124 srpc_alloc_bulk(int cpt, unsigned bulk_npg, unsigned bulk_len, int sink)
 125 {
 126         srpc_bulk_t *bk;
 127         int i;
 128
 129         LASSERT(bulk_npg > 0 && bulk_npg <= LNET_MAX_IOV);
 130
 131         LIBCFS_CPT_ALLOC(bk, lnet_cpt_table(), cpt,
 132                          offsetof(srpc_bulk_t, bk_iovs[bulk_npg]));
 133         if (!bk) {
 134                 CERROR("Can't allocate descriptor for %d pages\n", bulk_npg);
 135                 return NULL;
 136         }
 137
 138         memset(bk, 0, offsetof(srpc_bulk_t, bk_iovs[bulk_npg]));
 139         bk->bk_sink = sink;
 140         bk->bk_len = bulk_len;
 141         bk->bk_niov = bulk_npg;
 142
 143         for (i = 0; i < bulk_npg; i++) {
 144                 struct page *pg;
 145                 int nob;
 146
 147                 pg = alloc_pages_node(cfs_cpt_spread_node(lnet_cpt_table(), cpt),
 148                                       GFP_KERNEL, 0);
 149                 if (!pg) {
 150                         CERROR("Can't allocate page %d of %d\n", i, bulk_npg);
 151                         srpc_free_bulk(bk);
 152                         return NULL;
 153                 }
 154
 155                 nob = srpc_add_bulk_page(bk, pg, i, bulk_len);
 156                 bulk_len -= nob;
 157         }
 158
 159         return bk;
 160 }
 161
 162 static inline __u64
 163 srpc_next_id(void)
 164 {
 165         __u64 id;
 166
 167         spin_lock(&srpc_data.rpc_glock);
 168         id = srpc_data.rpc_matchbits++;
 169         spin_unlock(&srpc_data.rpc_glock);
 170         return id;
 171 }
 172
 173 static void
 174 srpc_init_server_rpc(struct srpc_server_rpc *rpc,
 175                      struct srpc_service_cd *scd,
 176                      struct srpc_buffer *buffer)
 177 {
 178         memset(rpc, 0, sizeof(*rpc));
 179         swi_init_workitem(&rpc->srpc_wi, rpc, srpc_handle_rpc,
 180                           srpc_serv_is_framework(scd->scd_svc) ?
 181                           lst_sched_serial : lst_sched_test[scd->scd_cpt]);
 182
 183         rpc->srpc_ev.ev_fired = 1; /* no event expected now */
 184
 185         rpc->srpc_scd = scd;
 186         rpc->srpc_reqstbuf = buffer;
 187         rpc->srpc_peer = buffer->buf_peer;
 188         rpc->srpc_self = buffer->buf_self;
 189         LNetInvalidateHandle(&rpc->srpc_replymdh);
 190 }
 191
 192 static void
 193 srpc_service_fini(struct srpc_service *svc)
 194 {
 195         struct srpc_service_cd *scd;
 196         struct srpc_server_rpc *rpc;
 197         struct srpc_buffer *buf;
 198         struct list_head *q;
 199         int i;
 200
 201         if (!svc->sv_cpt_data)
 202                 return;
 203
 204         cfs_percpt_for_each(scd, i, svc->sv_cpt_data) {
 205                 while (1) {
 206                         if (!list_empty(&scd->scd_buf_posted))
 207                                 q = &scd->scd_buf_posted;
 208                         else if (!list_empty(&scd->scd_buf_blocked))
 209                                 q = &scd->scd_buf_blocked;
 210                         else
 211                                 break;
 212
 213                         while (!list_empty(q)) {
 214                                 buf = list_entry(q->next, struct srpc_buffer,
 215                                                  buf_list);
 216                                 list_del(&buf->buf_list);
 217                                 LIBCFS_FREE(buf, sizeof(*buf));
 218                         }
 219                 }
 220
 221                 LASSERT(list_empty(&scd->scd_rpc_active));
 222
 223                 while (!list_empty(&scd->scd_rpc_free)) {
 224                         rpc = list_entry(scd->scd_rpc_free.next,
 225                                          struct srpc_server_rpc,
 226                                          srpc_list);
 227                         list_del(&rpc->srpc_list);
 228                         LIBCFS_FREE(rpc, sizeof(*rpc));
 229                 }
 230         }
 231
 232         cfs_percpt_free(svc->sv_cpt_data);
 233         svc->sv_cpt_data = NULL;
 234 }
 235
 236 static int
 237 srpc_service_nrpcs(struct srpc_service *svc)
 238 {
 239         int nrpcs = svc->sv_wi_total / svc->sv_ncpts;
 240
 241         return srpc_serv_is_framework(svc) ?
 242                max(nrpcs, SFW_FRWK_WI_MIN) : max(nrpcs, SFW_TEST_WI_MIN);
 243 }
 244
 245 int srpc_add_buffer(struct swi_workitem *wi);
 246
 247 static int
 248 srpc_service_init(struct srpc_service *svc)
 249 {
 250         struct srpc_service_cd *scd;
 251         struct srpc_server_rpc *rpc;
 252         int nrpcs;
 253         int i;
 254         int j;
 255
 256         svc->sv_shuttingdown = 0;
 257
 258         svc->sv_cpt_data = cfs_percpt_alloc(lnet_cpt_table(),
 259                                             sizeof(struct srpc_service_cd));
 260         if (!svc->sv_cpt_data)
 261                 return -ENOMEM;
 262
 263         svc->sv_ncpts = srpc_serv_is_framework(svc) ?
 264                         1 : cfs_cpt_number(lnet_cpt_table());
 265         nrpcs = srpc_service_nrpcs(svc);
 266
 267         cfs_percpt_for_each(scd, i, svc->sv_cpt_data) {
 268                 scd->scd_cpt = i;
 269                 scd->scd_svc = svc;
 270                 spin_lock_init(&scd->scd_lock);
 271                 INIT_LIST_HEAD(&scd->scd_rpc_free);
 272                 INIT_LIST_HEAD(&scd->scd_rpc_active);
 273                 INIT_LIST_HEAD(&scd->scd_buf_posted);
 274                 INIT_LIST_HEAD(&scd->scd_buf_blocked);
 275
 276                 scd->scd_ev.ev_data = scd;
 277                 scd->scd_ev.ev_type = SRPC_REQUEST_RCVD;
 278
 279                 /*
 280                  * NB: don't use lst_sched_serial for adding buffer,
 281                  * see details in srpc_service_add_buffers()
 282                  */
 283                 swi_init_workitem(&scd->scd_buf_wi, scd,
 284                                   srpc_add_buffer, lst_sched_test[i]);
 285
 286                 if (i && srpc_serv_is_framework(svc)) {
 287                         /*
 288                          * NB: framework service only needs srpc_service_cd for
 289                          * one partition, but we allocate for all to make
 290                          * it easier to implement, it will waste a little
 291                          * memory but nobody should care about this
 292                          */
 293                         continue;
 294                 }
 295
 296                 for (j = 0; j < nrpcs; j++) {
 297                         LIBCFS_CPT_ALLOC(rpc, lnet_cpt_table(),
 298                                          i, sizeof(*rpc));
 299                         if (!rpc) {
 300                                 srpc_service_fini(svc);
 301                                 return -ENOMEM;
 302                         }
 303                         list_add(&rpc->srpc_list, &scd->scd_rpc_free);
 304                 }
 305         }
 306
 307         return 0;
 308 }
 309
 310 int
 311 srpc_add_service(struct srpc_service *sv)
 312 {
 313         int id = sv->sv_id;
 314
 315         LASSERT(0 <= id && id <= SRPC_SERVICE_MAX_ID);
 316
 317         if (srpc_service_init(sv))
 318                 return -ENOMEM;
 319
 320         spin_lock(&srpc_data.rpc_glock);
 321
 322         LASSERT(srpc_data.rpc_state == SRPC_STATE_RUNNING);
 323
 324         if (srpc_data.rpc_services[id]) {
 325                 spin_unlock(&srpc_data.rpc_glock);
 326                 goto failed;
 327         }
 328
 329         srpc_data.rpc_services[id] = sv;
 330         spin_unlock(&srpc_data.rpc_glock);
 331
 332         CDEBUG(D_NET, "Adding service: id %d, name %s\n", id, sv->sv_name);
 333         return 0;
 334
 335  failed:
 336         srpc_service_fini(sv);
 337         return -EBUSY;
 338 }
 339
 340 int
 341 srpc_remove_service(srpc_service_t *sv)
 342 {
 343         int id = sv->sv_id;
 344
 345         spin_lock(&srpc_data.rpc_glock);
 346
 347         if (srpc_data.rpc_services[id] != sv) {
 348                 spin_unlock(&srpc_data.rpc_glock);
 349                 return -ENOENT;
 350         }
 351
 352         srpc_data.rpc_services[id] = NULL;
 353         spin_unlock(&srpc_data.rpc_glock);
 354         return 0;
 355 }
 356
 357 static int
 358 srpc_post_passive_rdma(int portal, int local, __u64 matchbits, void *buf,
 359                        int len, int options, lnet_process_id_t peer,
 360                        lnet_handle_md_t *mdh, srpc_event_t *ev)
 361 {
 362         int rc;
 363         lnet_md_t md;
 364         lnet_handle_me_t meh;
 365
 366         rc = LNetMEAttach(portal, peer, matchbits, 0, LNET_UNLINK,
 367                           local ? LNET_INS_LOCAL : LNET_INS_AFTER, &meh);
 368         if (rc) {
 369                 CERROR("LNetMEAttach failed: %d\n", rc);
 370                 LASSERT(rc == -ENOMEM);
 371                 return -ENOMEM;
 372         }
 373
 374         md.threshold = 1;
 375         md.user_ptr = ev;
 376         md.start = buf;
 377         md.length = len;
 378         md.options = options;
 379         md.eq_handle = srpc_data.rpc_lnet_eq;
 380
 381         rc = LNetMDAttach(meh, md, LNET_UNLINK, mdh);
 382         if (rc) {
 383                 CERROR("LNetMDAttach failed: %d\n", rc);
 384                 LASSERT(rc == -ENOMEM);
 385
 386                 rc = LNetMEUnlink(meh);
 387                 LASSERT(!rc);
 388                 return -ENOMEM;
 389         }
 390
 391         CDEBUG(D_NET, "Posted passive RDMA: peer %s, portal %d, matchbits %#llx\n",
 392                libcfs_id2str(peer), portal, matchbits);
 393         return 0;
 394 }
 395
 396 static int
 397 srpc_post_active_rdma(int portal, __u64 matchbits, void *buf, int len,
 398                       int options, lnet_process_id_t peer, lnet_nid_t self,
 399                       lnet_handle_md_t *mdh, srpc_event_t *ev)
 400 {
 401         int rc;
 402         lnet_md_t 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
 451 srpc_post_passive_rqtbuf(int service, int local, void *buf, int len,
 452                          lnet_handle_md_t *mdh, srpc_event_t *ev)
 453 {
 454         lnet_process_id_t 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
 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         LNetInvalidateHandle(&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
 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
 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
 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
 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
 700 srpc_service_recycle_buffer(struct srpc_service_cd *scd, srpc_buffer_t *buf)
 701 __must_hold(&scd->scd_lock)
 702 {
 703         if (!scd->scd_svc->sv_shuttingdown && scd->scd_buf_adjust >= 0) {
 704                 if (srpc_service_post_buffer(scd, buf)) {
 705                         CWARN("Failed to post %s buffer\n",
 706                               scd->scd_svc->sv_name);
 707                 }
 708                 return;
 709         }
 710
 711         /* service is shutting down, or we want to recycle some buffers */
 712         scd->scd_buf_total--;
 713
 714         if (scd->scd_buf_adjust < 0) {
 715                 scd->scd_buf_adjust++;
 716                 if (scd->scd_buf_adjust < 0 &&
 717                     !scd->scd_buf_total && !scd->scd_buf_posting) {
 718                         CDEBUG(D_INFO,
 719                                "Try to recycle %d buffers but nothing left\n",
 720                                scd->scd_buf_adjust);
 721                         scd->scd_buf_adjust = 0;
 722                 }
 723         }
 724
 725         spin_unlock(&scd->scd_lock);
 726         LIBCFS_FREE(buf, sizeof(*buf));
 727         spin_lock(&scd->scd_lock);
 728 }
 729
 730 void
 731 srpc_abort_service(struct srpc_service *sv)
 732 {
 733         struct srpc_service_cd *scd;
 734         struct srpc_server_rpc *rpc;
 735         int i;
 736
 737         CDEBUG(D_NET, "Aborting service: id %d, name %s\n",
 738                sv->sv_id, sv->sv_name);
 739
 740         cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
 741                 spin_lock(&scd->scd_lock);
 742
 743                 /*
 744                  * schedule in-flight RPCs to notice the abort, NB:
 745                  * racing with incoming RPCs; complete fix should make test
 746                  * RPCs carry session ID in its headers
 747                  */
 748                 list_for_each_entry(rpc, &scd->scd_rpc_active, srpc_list) {
 749                         rpc->srpc_aborted = 1;
 750                         swi_schedule_workitem(&rpc->srpc_wi);
 751                 }
 752
 753                 spin_unlock(&scd->scd_lock);
 754         }
 755 }
 756
 757 void
 758 srpc_shutdown_service(srpc_service_t *sv)
 759 {
 760         struct srpc_service_cd *scd;
 761         struct srpc_server_rpc *rpc;
 762         srpc_buffer_t *buf;
 763         int i;
 764
 765         CDEBUG(D_NET, "Shutting down service: id %d, name %s\n",
 766                sv->sv_id, sv->sv_name);
 767
 768         cfs_percpt_for_each(scd, i, sv->sv_cpt_data)
 769                 spin_lock(&scd->scd_lock);
 770
 771         sv->sv_shuttingdown = 1; /* i.e. no new active RPC */
 772
 773         cfs_percpt_for_each(scd, i, sv->sv_cpt_data)
 774                 spin_unlock(&scd->scd_lock);
 775
 776         cfs_percpt_for_each(scd, i, sv->sv_cpt_data) {
 777                 spin_lock(&scd->scd_lock);
 778
 779                 /* schedule in-flight RPCs to notice the shutdown */
 780                 list_for_each_entry(rpc, &scd->scd_rpc_active, srpc_list)
 781                         swi_schedule_workitem(&rpc->srpc_wi);
 782
 783                 spin_unlock(&scd->scd_lock);
 784
 785                 /*
 786                  * OK to traverse scd_buf_posted without lock, since no one
 787                  * touches scd_buf_posted now
 788                  */
 789                 list_for_each_entry(buf, &scd->scd_buf_posted, buf_list)
 790                         LNetMDUnlink(buf->buf_mdh);
 791         }
 792 }
 793
 794 static int
 795 srpc_send_request(srpc_client_rpc_t *rpc)
 796 {
 797         srpc_event_t *ev = &rpc->crpc_reqstev;
 798         int rc;
 799
 800         ev->ev_fired = 0;
 801         ev->ev_data = rpc;
 802         ev->ev_type = SRPC_REQUEST_SENT;
 803
 804          rc = srpc_post_active_rdma(srpc_serv_portal(rpc->crpc_service),
 805                                     rpc->crpc_service, &rpc->crpc_reqstmsg,
 806                                     sizeof(srpc_msg_t), LNET_MD_OP_PUT,
 807                                     rpc->crpc_dest, LNET_NID_ANY,
 808                                     &rpc->crpc_reqstmdh, ev);
 809         if (rc) {
 810                 LASSERT(rc == -ENOMEM);
 811                 ev->ev_fired = 1;  /* no more event expected */
 812         }
 813         return rc;
 814 }
 815
 816 static int
 817 srpc_prepare_reply(srpc_client_rpc_t *rpc)
 818 {
 819         srpc_event_t *ev = &rpc->crpc_replyev;
 820         __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.rpyid;
 821         int rc;
 822
 823         ev->ev_fired = 0;
 824         ev->ev_data = rpc;
 825         ev->ev_type = SRPC_REPLY_RCVD;
 826
 827         *id = srpc_next_id();
 828
 829         rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, 0, *id,
 830                                     &rpc->crpc_replymsg, sizeof(srpc_msg_t),
 831                                     LNET_MD_OP_PUT, rpc->crpc_dest,
 832                                     &rpc->crpc_replymdh, ev);
 833         if (rc) {
 834                 LASSERT(rc == -ENOMEM);
 835                 ev->ev_fired = 1;  /* no more event expected */
 836         }
 837         return rc;
 838 }
 839
 840 static int
 841 srpc_prepare_bulk(srpc_client_rpc_t *rpc)
 842 {
 843         srpc_bulk_t *bk = &rpc->crpc_bulk;
 844         srpc_event_t *ev = &rpc->crpc_bulkev;
 845         __u64 *id = &rpc->crpc_reqstmsg.msg_body.reqst.bulkid;
 846         int rc;
 847         int opt;
 848
 849         LASSERT(bk->bk_niov <= LNET_MAX_IOV);
 850
 851         if (!bk->bk_niov)
 852                 return 0; /* nothing to do */
 853
 854         opt = bk->bk_sink ? LNET_MD_OP_PUT : LNET_MD_OP_GET;
 855         opt |= LNET_MD_KIOV;
 856
 857         ev->ev_fired = 0;
 858         ev->ev_data = rpc;
 859         ev->ev_type = SRPC_BULK_REQ_RCVD;
 860
 861         *id = srpc_next_id();
 862
 863         rc = srpc_post_passive_rdma(SRPC_RDMA_PORTAL, 0, *id,
 864                                     &bk->bk_iovs[0], bk->bk_niov, opt,
 865                                     rpc->crpc_dest, &bk->bk_mdh, ev);
 866         if (rc) {
 867                 LASSERT(rc == -ENOMEM);
 868                 ev->ev_fired = 1;  /* no more event expected */
 869         }
 870         return rc;
 871 }
 872
 873 static int
 874 srpc_do_bulk(struct srpc_server_rpc *rpc)
 875 {
 876         srpc_event_t *ev = &rpc->srpc_ev;
 877         srpc_bulk_t *bk = rpc->srpc_bulk;
 878         __u64 id = rpc->srpc_reqstbuf->buf_msg.msg_body.reqst.bulkid;
 879         int rc;
 880         int opt;
 881
 882         LASSERT(bk);
 883
 884         opt = bk->bk_sink ? LNET_MD_OP_GET : LNET_MD_OP_PUT;
 885         opt |= LNET_MD_KIOV;
 886
 887         ev->ev_fired = 0;
 888         ev->ev_data = rpc;
 889         ev->ev_type = bk->bk_sink ? SRPC_BULK_GET_RPLD : SRPC_BULK_PUT_SENT;
 890
 891         rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, id,
 892                                    &bk->bk_iovs[0], bk->bk_niov, opt,
 893                                    rpc->srpc_peer, rpc->srpc_self,
 894                                    &bk->bk_mdh, ev);
 895         if (rc)
 896                 ev->ev_fired = 1;  /* no more event expected */
 897         return rc;
 898 }
 899
 900 /* only called from srpc_handle_rpc */
 901 static void
 902 srpc_server_rpc_done(struct srpc_server_rpc *rpc, int status)
 903 {
 904         struct srpc_service_cd *scd = rpc->srpc_scd;
 905         struct srpc_service *sv = scd->scd_svc;
 906         srpc_buffer_t *buffer;
 907
 908         LASSERT(status || rpc->srpc_wi.swi_state == SWI_STATE_DONE);
 909
 910         rpc->srpc_status = status;
 911
 912         CDEBUG_LIMIT(!status ? D_NET : D_NETERROR,
 913                      "Server RPC %p done: service %s, peer %s, status %s:%d\n",
 914                      rpc, sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 915                      swi_state2str(rpc->srpc_wi.swi_state), status);
 916
 917         if (status) {
 918                 spin_lock(&srpc_data.rpc_glock);
 919                 srpc_data.rpc_counters.rpcs_dropped++;
 920                 spin_unlock(&srpc_data.rpc_glock);
 921         }
 922
 923         if (rpc->srpc_done)
 924                 (*rpc->srpc_done) (rpc);
 925         LASSERT(!rpc->srpc_bulk);
 926
 927         spin_lock(&scd->scd_lock);
 928
 929         if (rpc->srpc_reqstbuf) {
 930                 /*
 931                  * NB might drop sv_lock in srpc_service_recycle_buffer, but
 932                  * sv won't go away for scd_rpc_active must not be empty
 933                  */
 934                 srpc_service_recycle_buffer(scd, rpc->srpc_reqstbuf);
 935                 rpc->srpc_reqstbuf = NULL;
 936         }
 937
 938         list_del(&rpc->srpc_list); /* from scd->scd_rpc_active */
 939
 940         /*
 941          * No one can schedule me now since:
 942          * - I'm not on scd_rpc_active.
 943          * - all LNet events have been fired.
 944          * Cancel pending schedules and prevent future schedule attempts:
 945          */
 946         LASSERT(rpc->srpc_ev.ev_fired);
 947         swi_exit_workitem(&rpc->srpc_wi);
 948
 949         if (!sv->sv_shuttingdown && !list_empty(&scd->scd_buf_blocked)) {
 950                 buffer = list_entry(scd->scd_buf_blocked.next,
 951                                     srpc_buffer_t, buf_list);
 952                 list_del(&buffer->buf_list);
 953
 954                 srpc_init_server_rpc(rpc, scd, buffer);
 955                 list_add_tail(&rpc->srpc_list, &scd->scd_rpc_active);
 956                 swi_schedule_workitem(&rpc->srpc_wi);
 957         } else {
 958                 list_add(&rpc->srpc_list, &scd->scd_rpc_free);
 959         }
 960
 961         spin_unlock(&scd->scd_lock);
 962 }
 963
 964 /* handles an incoming RPC */
 965 int
 966 srpc_handle_rpc(swi_workitem_t *wi)
 967 {
 968         struct srpc_server_rpc *rpc = wi->swi_workitem.wi_data;
 969         struct srpc_service_cd *scd = rpc->srpc_scd;
 970         struct srpc_service *sv = scd->scd_svc;
 971         srpc_event_t *ev = &rpc->srpc_ev;
 972         int rc = 0;
 973
 974         LASSERT(wi == &rpc->srpc_wi);
 975
 976         spin_lock(&scd->scd_lock);
 977
 978         if (sv->sv_shuttingdown || rpc->srpc_aborted) {
 979                 spin_unlock(&scd->scd_lock);
 980
 981                 if (rpc->srpc_bulk)
 982                         LNetMDUnlink(rpc->srpc_bulk->bk_mdh);
 983                 LNetMDUnlink(rpc->srpc_replymdh);
 984
 985                 if (ev->ev_fired) { /* no more event, OK to finish */
 986                         srpc_server_rpc_done(rpc, -ESHUTDOWN);
 987                         return 1;
 988                 }
 989                 return 0;
 990         }
 991
 992         spin_unlock(&scd->scd_lock);
 993
 994         switch (wi->swi_state) {
 995         default:
 996                 LBUG();
 997         case SWI_STATE_NEWBORN: {
 998                 srpc_msg_t *msg;
 999                 srpc_generic_reply_t *reply;
1000
1001                 msg = &rpc->srpc_reqstbuf->buf_msg;
1002                 reply = &rpc->srpc_replymsg.msg_body.reply;
1003
1004                 if (!msg->msg_magic) {
1005                         /* moaned already in srpc_lnet_ev_handler */
1006                         srpc_server_rpc_done(rpc, EBADMSG);
1007                         return 1;
1008                 }
1009
1010                 srpc_unpack_msg_hdr(msg);
1011                 if (msg->msg_version != SRPC_MSG_VERSION) {
1012                         CWARN("Version mismatch: %u, %u expected, from %s\n",
1013                               msg->msg_version, SRPC_MSG_VERSION,
1014                               libcfs_id2str(rpc->srpc_peer));
1015                         reply->status = EPROTO;
1016                         /* drop through and send reply */
1017                 } else {
1018                         reply->status = 0;
1019                         rc = (*sv->sv_handler)(rpc);
1020                         LASSERT(!reply->status || !rpc->srpc_bulk);
1021                         if (rc) {
1022                                 srpc_server_rpc_done(rpc, rc);
1023                                 return 1;
1024                         }
1025                 }
1026
1027                 wi->swi_state = SWI_STATE_BULK_STARTED;
1028
1029                 if (rpc->srpc_bulk) {
1030                         rc = srpc_do_bulk(rpc);
1031                         if (!rc)
1032                                 return 0; /* wait for bulk */
1033
1034                         LASSERT(ev->ev_fired);
1035                         ev->ev_status = rc;
1036                 }
1037         }
1038         case SWI_STATE_BULK_STARTED:
1039                 LASSERT(!rpc->srpc_bulk || ev->ev_fired);
1040
1041                 if (rpc->srpc_bulk) {
1042                         rc = ev->ev_status;
1043
1044                         if (sv->sv_bulk_ready)
1045                                 rc = (*sv->sv_bulk_ready) (rpc, rc);
1046
1047                         if (rc) {
1048                                 srpc_server_rpc_done(rpc, rc);
1049                                 return 1;
1050                         }
1051                 }
1052
1053                 wi->swi_state = SWI_STATE_REPLY_SUBMITTED;
1054                 rc = srpc_send_reply(rpc);
1055                 if (!rc)
1056                         return 0; /* wait for reply */
1057                 srpc_server_rpc_done(rpc, rc);
1058                 return 1;
1059
1060         case SWI_STATE_REPLY_SUBMITTED:
1061                 if (!ev->ev_fired) {
1062                         CERROR("RPC %p: bulk %p, service %d\n",
1063                                rpc, rpc->srpc_bulk, sv->sv_id);
1064                         CERROR("Event: status %d, type %d, lnet %d\n",
1065                                ev->ev_status, ev->ev_type, ev->ev_lnet);
1066                         LASSERT(ev->ev_fired);
1067                 }
1068
1069                 wi->swi_state = SWI_STATE_DONE;
1070                 srpc_server_rpc_done(rpc, ev->ev_status);
1071                 return 1;
1072         }
1073
1074         return 0;
1075 }
1076
1077 static void
1078 srpc_client_rpc_expired(void *data)
1079 {
1080         srpc_client_rpc_t *rpc = data;
1081
1082         CWARN("Client RPC expired: service %d, peer %s, timeout %d.\n",
1083               rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1084               rpc->crpc_timeout);
1085
1086         spin_lock(&rpc->crpc_lock);
1087
1088         rpc->crpc_timeout = 0;
1089         srpc_abort_rpc(rpc, -ETIMEDOUT);
1090
1091         spin_unlock(&rpc->crpc_lock);
1092
1093         spin_lock(&srpc_data.rpc_glock);
1094         srpc_data.rpc_counters.rpcs_expired++;
1095         spin_unlock(&srpc_data.rpc_glock);
1096 }
1097
1098 static void
1099 srpc_add_client_rpc_timer(srpc_client_rpc_t *rpc)
1100 {
1101         struct stt_timer *timer = &rpc->crpc_timer;
1102
1103         if (!rpc->crpc_timeout)
1104                 return;
1105
1106         INIT_LIST_HEAD(&timer->stt_list);
1107         timer->stt_data = rpc;
1108         timer->stt_func = srpc_client_rpc_expired;
1109         timer->stt_expires = ktime_get_real_seconds() + rpc->crpc_timeout;
1110         stt_add_timer(timer);
1111 }
1112
1113 /*
1114  * Called with rpc->crpc_lock held.
1115  *
1116  * Upon exit the RPC expiry timer is not queued and the handler is not
1117  * running on any CPU.
1118  */
1119 static void
1120 srpc_del_client_rpc_timer(srpc_client_rpc_t *rpc)
1121 {
1122         /* timer not planted or already exploded */
1123         if (!rpc->crpc_timeout)
1124                 return;
1125
1126         /* timer successfully defused */
1127         if (stt_del_timer(&rpc->crpc_timer))
1128                 return;
1129
1130         /* timer detonated, wait for it to explode */
1131         while (rpc->crpc_timeout) {
1132                 spin_unlock(&rpc->crpc_lock);
1133
1134                 schedule();
1135
1136                 spin_lock(&rpc->crpc_lock);
1137         }
1138 }
1139
1140 static void
1141 srpc_client_rpc_done(srpc_client_rpc_t *rpc, int status)
1142 {
1143         swi_workitem_t *wi = &rpc->crpc_wi;
1144
1145         LASSERT(status || wi->swi_state == SWI_STATE_DONE);
1146
1147         spin_lock(&rpc->crpc_lock);
1148
1149         rpc->crpc_closed = 1;
1150         if (!rpc->crpc_status)
1151                 rpc->crpc_status = status;
1152
1153         srpc_del_client_rpc_timer(rpc);
1154
1155         CDEBUG_LIMIT(!status ? D_NET : D_NETERROR,
1156                      "Client RPC done: service %d, peer %s, status %s:%d:%d\n",
1157                      rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1158                      swi_state2str(wi->swi_state), rpc->crpc_aborted, status);
1159
1160         /*
1161          * No one can schedule me now since:
1162          * - RPC timer has been defused.
1163          * - all LNet events have been fired.
1164          * - crpc_closed has been set, preventing srpc_abort_rpc from
1165          *   scheduling me.
1166          * Cancel pending schedules and prevent future schedule attempts:
1167          */
1168         LASSERT(!srpc_event_pending(rpc));
1169         swi_exit_workitem(wi);
1170
1171         spin_unlock(&rpc->crpc_lock);
1172
1173         (*rpc->crpc_done)(rpc);
1174 }
1175
1176 /* sends an outgoing RPC */
1177 int
1178 srpc_send_rpc(swi_workitem_t *wi)
1179 {
1180         int rc = 0;
1181         srpc_client_rpc_t *rpc;
1182         srpc_msg_t *reply;
1183         int do_bulk;
1184
1185         LASSERT(wi);
1186
1187         rpc = wi->swi_workitem.wi_data;
1188
1189         LASSERT(rpc);
1190         LASSERT(wi == &rpc->crpc_wi);
1191
1192         reply = &rpc->crpc_replymsg;
1193         do_bulk = rpc->crpc_bulk.bk_niov > 0;
1194
1195         spin_lock(&rpc->crpc_lock);
1196
1197         if (rpc->crpc_aborted) {
1198                 spin_unlock(&rpc->crpc_lock);
1199                 goto abort;
1200         }
1201
1202         spin_unlock(&rpc->crpc_lock);
1203
1204         switch (wi->swi_state) {
1205         default:
1206                 LBUG();
1207         case SWI_STATE_NEWBORN:
1208                 LASSERT(!srpc_event_pending(rpc));
1209
1210                 rc = srpc_prepare_reply(rpc);
1211                 if (rc) {
1212                         srpc_client_rpc_done(rpc, rc);
1213                         return 1;
1214                 }
1215
1216                 rc = srpc_prepare_bulk(rpc);
1217                 if (rc)
1218                         break;
1219
1220                 wi->swi_state = SWI_STATE_REQUEST_SUBMITTED;
1221                 rc = srpc_send_request(rpc);
1222                 break;
1223
1224         case SWI_STATE_REQUEST_SUBMITTED:
1225                 /*
1226                  * CAVEAT EMPTOR: rqtev, rpyev, and bulkev may come in any
1227                  * order; however, they're processed in a strict order:
1228                  * rqt, rpy, and bulk.
1229                  */
1230                 if (!rpc->crpc_reqstev.ev_fired)
1231                         break;
1232
1233                 rc = rpc->crpc_reqstev.ev_status;
1234                 if (rc)
1235                         break;
1236
1237                 wi->swi_state = SWI_STATE_REQUEST_SENT;
1238                 /* perhaps more events, fall thru */
1239         case SWI_STATE_REQUEST_SENT: {
1240                 srpc_msg_type_t type = srpc_service2reply(rpc->crpc_service);
1241
1242                 if (!rpc->crpc_replyev.ev_fired)
1243                         break;
1244
1245                 rc = rpc->crpc_replyev.ev_status;
1246                 if (rc)
1247                         break;
1248
1249                 srpc_unpack_msg_hdr(reply);
1250                 if (reply->msg_type != type ||
1251                     (reply->msg_magic != SRPC_MSG_MAGIC &&
1252                      reply->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1253                         CWARN("Bad message from %s: type %u (%d expected), magic %u (%d expected).\n",
1254                               libcfs_id2str(rpc->crpc_dest),
1255                               reply->msg_type, type,
1256                               reply->msg_magic, SRPC_MSG_MAGIC);
1257                         rc = -EBADMSG;
1258                         break;
1259                 }
1260
1261                 if (do_bulk && reply->msg_body.reply.status) {
1262                         CWARN("Remote error %d at %s, unlink bulk buffer in case peer didn't initiate bulk transfer\n",
1263                               reply->msg_body.reply.status,
1264                               libcfs_id2str(rpc->crpc_dest));
1265                         LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1266                 }
1267
1268                 wi->swi_state = SWI_STATE_REPLY_RECEIVED;
1269         }
1270         case SWI_STATE_REPLY_RECEIVED:
1271                 if (do_bulk && !rpc->crpc_bulkev.ev_fired)
1272                         break;
1273
1274                 rc = do_bulk ? rpc->crpc_bulkev.ev_status : 0;
1275
1276                 /*
1277                  * Bulk buffer was unlinked due to remote error. Clear error
1278                  * since reply buffer still contains valid data.
1279                  * NB rpc->crpc_done shouldn't look into bulk data in case of
1280                  * remote error.
1281                  */
1282                 if (do_bulk && rpc->crpc_bulkev.ev_lnet == LNET_EVENT_UNLINK &&
1283                     !rpc->crpc_status && reply->msg_body.reply.status)
1284                         rc = 0;
1285
1286                 wi->swi_state = SWI_STATE_DONE;
1287                 srpc_client_rpc_done(rpc, rc);
1288                 return 1;
1289         }
1290
1291         if (rc) {
1292                 spin_lock(&rpc->crpc_lock);
1293                 srpc_abort_rpc(rpc, rc);
1294                 spin_unlock(&rpc->crpc_lock);
1295         }
1296
1297 abort:
1298         if (rpc->crpc_aborted) {
1299                 LNetMDUnlink(rpc->crpc_reqstmdh);
1300                 LNetMDUnlink(rpc->crpc_replymdh);
1301                 LNetMDUnlink(rpc->crpc_bulk.bk_mdh);
1302
1303                 if (!srpc_event_pending(rpc)) {
1304                         srpc_client_rpc_done(rpc, -EINTR);
1305                         return 1;
1306                 }
1307         }
1308         return 0;
1309 }
1310
1311 srpc_client_rpc_t *
1312 srpc_create_client_rpc(lnet_process_id_t peer, int service,
1313                        int nbulkiov, int bulklen,
1314                        void (*rpc_done)(srpc_client_rpc_t *),
1315                        void (*rpc_fini)(srpc_client_rpc_t *), void *priv)
1316 {
1317         srpc_client_rpc_t *rpc;
1318
1319         LIBCFS_ALLOC(rpc, offsetof(srpc_client_rpc_t,
1320                                    crpc_bulk.bk_iovs[nbulkiov]));
1321         if (!rpc)
1322                 return NULL;
1323
1324         srpc_init_client_rpc(rpc, peer, service, nbulkiov,
1325                              bulklen, rpc_done, rpc_fini, priv);
1326         return rpc;
1327 }
1328
1329 /* called with rpc->crpc_lock held */
1330 void
1331 srpc_abort_rpc(srpc_client_rpc_t *rpc, int why)
1332 {
1333         LASSERT(why);
1334
1335         if (rpc->crpc_aborted || /* already aborted */
1336             rpc->crpc_closed)    /* callback imminent */
1337                 return;
1338
1339         CDEBUG(D_NET, "Aborting RPC: service %d, peer %s, state %s, why %d\n",
1340                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
1341                swi_state2str(rpc->crpc_wi.swi_state), why);
1342
1343         rpc->crpc_aborted = 1;
1344         rpc->crpc_status = why;
1345         swi_schedule_workitem(&rpc->crpc_wi);
1346 }
1347
1348 /* called with rpc->crpc_lock held */
1349 void
1350 srpc_post_rpc(srpc_client_rpc_t *rpc)
1351 {
1352         LASSERT(!rpc->crpc_aborted);
1353         LASSERT(srpc_data.rpc_state == SRPC_STATE_RUNNING);
1354
1355         CDEBUG(D_NET, "Posting RPC: peer %s, service %d, timeout %d\n",
1356                libcfs_id2str(rpc->crpc_dest), rpc->crpc_service,
1357                rpc->crpc_timeout);
1358
1359         srpc_add_client_rpc_timer(rpc);
1360         swi_schedule_workitem(&rpc->crpc_wi);
1361 }
1362
1363 int
1364 srpc_send_reply(struct srpc_server_rpc *rpc)
1365 {
1366         srpc_event_t *ev = &rpc->srpc_ev;
1367         struct srpc_msg *msg = &rpc->srpc_replymsg;
1368         struct srpc_buffer *buffer = rpc->srpc_reqstbuf;
1369         struct srpc_service_cd *scd = rpc->srpc_scd;
1370         struct srpc_service *sv = scd->scd_svc;
1371         __u64 rpyid;
1372         int rc;
1373
1374         LASSERT(buffer);
1375         rpyid = buffer->buf_msg.msg_body.reqst.rpyid;
1376
1377         spin_lock(&scd->scd_lock);
1378
1379         if (!sv->sv_shuttingdown && !srpc_serv_is_framework(sv)) {
1380                 /*
1381                  * Repost buffer before replying since test client
1382                  * might send me another RPC once it gets the reply
1383                  */
1384                 if (srpc_service_post_buffer(scd, buffer))
1385                         CWARN("Failed to repost %s buffer\n", sv->sv_name);
1386                 rpc->srpc_reqstbuf = NULL;
1387         }
1388
1389         spin_unlock(&scd->scd_lock);
1390
1391         ev->ev_fired = 0;
1392         ev->ev_data = rpc;
1393         ev->ev_type = SRPC_REPLY_SENT;
1394
1395         msg->msg_magic = SRPC_MSG_MAGIC;
1396         msg->msg_version = SRPC_MSG_VERSION;
1397         msg->msg_type = srpc_service2reply(sv->sv_id);
1398
1399         rc = srpc_post_active_rdma(SRPC_RDMA_PORTAL, rpyid, msg,
1400                                    sizeof(*msg), LNET_MD_OP_PUT,
1401                                    rpc->srpc_peer, rpc->srpc_self,
1402                                    &rpc->srpc_replymdh, ev);
1403         if (rc)
1404                 ev->ev_fired = 1;  /* no more event expected */
1405         return rc;
1406 }
1407
1408 /* when in kernel always called with LNET_LOCK() held, and in thread context */
1409 static void
1410 srpc_lnet_ev_handler(lnet_event_t *ev)
1411 {
1412         struct srpc_service_cd *scd;
1413         srpc_event_t *rpcev = ev->md.user_ptr;
1414         srpc_client_rpc_t *crpc;
1415         struct srpc_server_rpc *srpc;
1416         srpc_buffer_t *buffer;
1417         srpc_service_t *sv;
1418         srpc_msg_t *msg;
1419         srpc_msg_type_t type;
1420
1421         LASSERT(!in_interrupt());
1422
1423         if (ev->status) {
1424                 __u32 errors;
1425
1426                 spin_lock(&srpc_data.rpc_glock);
1427                 if (ev->status != -ECANCELED) /* cancellation is not error */
1428                         srpc_data.rpc_counters.errors++;
1429                 errors = srpc_data.rpc_counters.errors;
1430                 spin_unlock(&srpc_data.rpc_glock);
1431
1432                 CNETERR("LNet event status %d type %d, RPC errors %u\n",
1433                         ev->status, ev->type, errors);
1434         }
1435
1436         rpcev->ev_lnet = ev->type;
1437
1438         switch (rpcev->ev_type) {
1439         default:
1440                 CERROR("Unknown event: status %d, type %d, lnet %d\n",
1441                        rpcev->ev_status, rpcev->ev_type, rpcev->ev_lnet);
1442                 LBUG();
1443         case SRPC_REQUEST_SENT:
1444                 if (!ev->status && ev->type != LNET_EVENT_UNLINK) {
1445                         spin_lock(&srpc_data.rpc_glock);
1446                         srpc_data.rpc_counters.rpcs_sent++;
1447                         spin_unlock(&srpc_data.rpc_glock);
1448                 }
1449         case SRPC_REPLY_RCVD:
1450         case SRPC_BULK_REQ_RCVD:
1451                 crpc = rpcev->ev_data;
1452
1453                 if (rpcev != &crpc->crpc_reqstev &&
1454                     rpcev != &crpc->crpc_replyev &&
1455                     rpcev != &crpc->crpc_bulkev) {
1456                         CERROR("rpcev %p, crpc %p, reqstev %p, replyev %p, bulkev %p\n",
1457                                rpcev, crpc, &crpc->crpc_reqstev,
1458                                &crpc->crpc_replyev, &crpc->crpc_bulkev);
1459                         CERROR("Bad event: status %d, type %d, lnet %d\n",
1460                                rpcev->ev_status, rpcev->ev_type, rpcev->ev_lnet);
1461                         LBUG();
1462                 }
1463
1464                 spin_lock(&crpc->crpc_lock);
1465
1466                 LASSERT(!rpcev->ev_fired);
1467                 rpcev->ev_fired = 1;
1468                 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1469                                                 -EINTR : ev->status;
1470                 swi_schedule_workitem(&crpc->crpc_wi);
1471
1472                 spin_unlock(&crpc->crpc_lock);
1473                 break;
1474
1475         case SRPC_REQUEST_RCVD:
1476                 scd = rpcev->ev_data;
1477                 sv = scd->scd_svc;
1478
1479                 LASSERT(rpcev == &scd->scd_ev);
1480
1481                 spin_lock(&scd->scd_lock);
1482
1483                 LASSERT(ev->unlinked);
1484                 LASSERT(ev->type == LNET_EVENT_PUT ||
1485                         ev->type == LNET_EVENT_UNLINK);
1486                 LASSERT(ev->type != LNET_EVENT_UNLINK ||
1487                         sv->sv_shuttingdown);
1488
1489                 buffer = container_of(ev->md.start, srpc_buffer_t, buf_msg);
1490                 buffer->buf_peer = ev->initiator;
1491                 buffer->buf_self = ev->target.nid;
1492
1493                 LASSERT(scd->scd_buf_nposted > 0);
1494                 scd->scd_buf_nposted--;
1495
1496                 if (sv->sv_shuttingdown) {
1497                         /*
1498                          * Leave buffer on scd->scd_buf_nposted since
1499                          * srpc_finish_service needs to traverse it.
1500                          */
1501                         spin_unlock(&scd->scd_lock);
1502                         break;
1503                 }
1504
1505                 if (scd->scd_buf_err_stamp &&
1506                     scd->scd_buf_err_stamp < ktime_get_real_seconds()) {
1507                         /* re-enable adding buffer */
1508                         scd->scd_buf_err_stamp = 0;
1509                         scd->scd_buf_err = 0;
1510                 }
1511
1512                 if (!scd->scd_buf_err && /* adding buffer is enabled */
1513                     !scd->scd_buf_adjust &&
1514                     scd->scd_buf_nposted < scd->scd_buf_low) {
1515                         scd->scd_buf_adjust = max(scd->scd_buf_total / 2,
1516                                                   SFW_TEST_WI_MIN);
1517                         swi_schedule_workitem(&scd->scd_buf_wi);
1518                 }
1519
1520                 list_del(&buffer->buf_list); /* from scd->scd_buf_posted */
1521                 msg = &buffer->buf_msg;
1522                 type = srpc_service2request(sv->sv_id);
1523
1524                 if (ev->status || ev->mlength != sizeof(*msg) ||
1525                     (msg->msg_type != type &&
1526                      msg->msg_type != __swab32(type)) ||
1527                     (msg->msg_magic != SRPC_MSG_MAGIC &&
1528                      msg->msg_magic != __swab32(SRPC_MSG_MAGIC))) {
1529                         CERROR("Dropping RPC (%s) from %s: status %d mlength %d type %u magic %u.\n",
1530                                sv->sv_name, libcfs_id2str(ev->initiator),
1531                                ev->status, ev->mlength,
1532                                msg->msg_type, msg->msg_magic);
1533
1534                         /*
1535                          * NB can't call srpc_service_recycle_buffer here since
1536                          * it may call LNetM[DE]Attach. The invalid magic tells
1537                          * srpc_handle_rpc to drop this RPC
1538                          */
1539                         msg->msg_magic = 0;
1540                 }
1541
1542                 if (!list_empty(&scd->scd_rpc_free)) {
1543                         srpc = list_entry(scd->scd_rpc_free.next,
1544                                           struct srpc_server_rpc,
1545                                           srpc_list);
1546                         list_del(&srpc->srpc_list);
1547
1548                         srpc_init_server_rpc(srpc, scd, buffer);
1549                         list_add_tail(&srpc->srpc_list,
1550                                       &scd->scd_rpc_active);
1551                         swi_schedule_workitem(&srpc->srpc_wi);
1552                 } else {
1553                         list_add_tail(&buffer->buf_list,
1554                                       &scd->scd_buf_blocked);
1555                 }
1556
1557                 spin_unlock(&scd->scd_lock);
1558
1559                 spin_lock(&srpc_data.rpc_glock);
1560                 srpc_data.rpc_counters.rpcs_rcvd++;
1561                 spin_unlock(&srpc_data.rpc_glock);
1562                 break;
1563
1564         case SRPC_BULK_GET_RPLD:
1565                 LASSERT(ev->type == LNET_EVENT_SEND ||
1566                         ev->type == LNET_EVENT_REPLY ||
1567                         ev->type == LNET_EVENT_UNLINK);
1568
1569                 if (!ev->unlinked)
1570                         break; /* wait for final event */
1571
1572         case SRPC_BULK_PUT_SENT:
1573                 if (!ev->status && ev->type != LNET_EVENT_UNLINK) {
1574                         spin_lock(&srpc_data.rpc_glock);
1575
1576                         if (rpcev->ev_type == SRPC_BULK_GET_RPLD)
1577                                 srpc_data.rpc_counters.bulk_get += ev->mlength;
1578                         else
1579                                 srpc_data.rpc_counters.bulk_put += ev->mlength;
1580
1581                         spin_unlock(&srpc_data.rpc_glock);
1582                 }
1583         case SRPC_REPLY_SENT:
1584                 srpc = rpcev->ev_data;
1585                 scd = srpc->srpc_scd;
1586
1587                 LASSERT(rpcev == &srpc->srpc_ev);
1588
1589                 spin_lock(&scd->scd_lock);
1590
1591                 rpcev->ev_fired = 1;
1592                 rpcev->ev_status = (ev->type == LNET_EVENT_UNLINK) ?
1593                                    -EINTR : ev->status;
1594                 swi_schedule_workitem(&srpc->srpc_wi);
1595
1596                 spin_unlock(&scd->scd_lock);
1597                 break;
1598         }
1599 }
1600
1601 int
1602 srpc_startup(void)
1603 {
1604         int rc;
1605
1606         memset(&srpc_data, 0, sizeof(struct smoketest_rpc));
1607         spin_lock_init(&srpc_data.rpc_glock);
1608
1609         /* 1 second pause to avoid timestamp reuse */
1610         set_current_state(TASK_UNINTERRUPTIBLE);
1611         schedule_timeout(cfs_time_seconds(1));
1612         srpc_data.rpc_matchbits = ((__u64)ktime_get_real_seconds()) << 48;
1613
1614         srpc_data.rpc_state = SRPC_STATE_NONE;
1615
1616         rc = LNetNIInit(LNET_PID_LUSTRE);
1617         if (rc < 0) {
1618                 CERROR("LNetNIInit() has failed: %d\n", rc);
1619                 return rc;
1620         }
1621
1622         srpc_data.rpc_state = SRPC_STATE_NI_INIT;
1623
1624         LNetInvalidateHandle(&srpc_data.rpc_lnet_eq);
1625         rc = LNetEQAlloc(0, srpc_lnet_ev_handler, &srpc_data.rpc_lnet_eq);
1626         if (rc) {
1627                 CERROR("LNetEQAlloc() has failed: %d\n", rc);
1628                 goto bail;
1629         }
1630
1631         rc = LNetSetLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1632         LASSERT(!rc);
1633         rc = LNetSetLazyPortal(SRPC_REQUEST_PORTAL);
1634         LASSERT(!rc);
1635
1636         srpc_data.rpc_state = SRPC_STATE_EQ_INIT;
1637
1638         rc = stt_startup();
1639
1640 bail:
1641         if (rc)
1642                 srpc_shutdown();
1643         else
1644                 srpc_data.rpc_state = SRPC_STATE_RUNNING;
1645
1646         return rc;
1647 }
1648
1649 void
1650 srpc_shutdown(void)
1651 {
1652         int i;
1653         int rc;
1654         int state;
1655
1656         state = srpc_data.rpc_state;
1657         srpc_data.rpc_state = SRPC_STATE_STOPPING;
1658
1659         switch (state) {
1660         default:
1661                 LBUG();
1662         case SRPC_STATE_RUNNING:
1663                 spin_lock(&srpc_data.rpc_glock);
1664
1665                 for (i = 0; i <= SRPC_SERVICE_MAX_ID; i++) {
1666                         srpc_service_t *sv = srpc_data.rpc_services[i];
1667
1668                         LASSERTF(!sv, "service not empty: id %d, name %s\n",
1669                                  i, sv->sv_name);
1670                 }
1671
1672                 spin_unlock(&srpc_data.rpc_glock);
1673
1674                 stt_shutdown();
1675
1676         case SRPC_STATE_EQ_INIT:
1677                 rc = LNetClearLazyPortal(SRPC_FRAMEWORK_REQUEST_PORTAL);
1678                 rc = LNetClearLazyPortal(SRPC_REQUEST_PORTAL);
1679                 LASSERT(!rc);
1680                 rc = LNetEQFree(srpc_data.rpc_lnet_eq);
1681                 LASSERT(!rc); /* the EQ should have no user by now */
1682
1683         case SRPC_STATE_NI_INIT:
1684                 LNetNIFini();
1685         }
1686 }