lnet/selftest/framework.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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lnet/selftest/framework.c
  32  *
  33  * Author: Isaac Huang <isaac@clusterfs.com>
  34  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LNET
  38
  39 #include "selftest.h"
  40
  41 struct lst_session_id LST_INVALID_SID = { .ses_nid = LNET_ANY_NID, .ses_stamp = -1};
  42
  43 static int session_timeout = 100;
  44 module_param(session_timeout, int, 0444);
  45 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
  46
  47 static int rpc_timeout = 64;
  48 module_param(rpc_timeout, int, 0644);
  49 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
  50
  51 #define sfw_unpack_id(id)               \
  52 do {                                    \
  53         __swab64s(&(id).nid);           \
  54         __swab32s(&(id).pid);           \
  55 } while (0)
  56
  57 #define sfw_unpack_sid(sid)             \
  58 do {                                    \
  59         __swab64s(&(sid).ses_nid);      \
  60         __swab64s(&(sid).ses_stamp);    \
  61 } while (0)
  62
  63 #define sfw_unpack_fw_counters(fc)        \
  64 do {                                      \
  65         __swab32s(&(fc).running_ms);      \
  66         __swab32s(&(fc).active_batches);  \
  67         __swab32s(&(fc).zombie_sessions); \
  68         __swab32s(&(fc).brw_errors);      \
  69         __swab32s(&(fc).ping_errors);     \
  70 } while (0)
  71
  72 #define sfw_unpack_rpc_counters(rc)     \
  73 do {                                    \
  74         __swab32s(&(rc).errors);        \
  75         __swab32s(&(rc).rpcs_sent);     \
  76         __swab32s(&(rc).rpcs_rcvd);     \
  77         __swab32s(&(rc).rpcs_dropped);  \
  78         __swab32s(&(rc).rpcs_expired);  \
  79         __swab64s(&(rc).bulk_get);      \
  80         __swab64s(&(rc).bulk_put);      \
  81 } while (0)
  82
  83 #define sfw_unpack_lnet_counters(lc)    \
  84 do {                                    \
  85         __swab32s(&(lc).lcc_errors);        \
  86         __swab32s(&(lc).lcc_msgs_max);      \
  87         __swab32s(&(lc).lcc_msgs_alloc);    \
  88         __swab32s(&(lc).lcc_send_count);    \
  89         __swab32s(&(lc).lcc_recv_count);    \
  90         __swab32s(&(lc).lcc_drop_count);    \
  91         __swab32s(&(lc).lcc_route_count);   \
  92         __swab64s(&(lc).lcc_send_length);   \
  93         __swab64s(&(lc).lcc_recv_length);   \
  94         __swab64s(&(lc).lcc_drop_length);   \
  95         __swab64s(&(lc).lcc_route_length);  \
  96 } while (0)
  97
  98 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
  99 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 100
 101 static struct smoketest_framework {
 102         /* RPCs to be recycled */
 103         struct list_head        fw_zombie_rpcs;
 104         /* stopping sessions */
 105         struct list_head        fw_zombie_sessions;
 106         /* registered test cases */
 107         struct list_head        fw_tests;
 108         /* # zombie sessions */
 109         atomic_t                fw_nzombies;
 110         /* serialise */
 111         spinlock_t              fw_lock;
 112         /* _the_ session */
 113         struct sfw_session      *fw_session;
 114         /* shutdown in progress */
 115         int                     fw_shuttingdown;
 116         /* running RPC */
 117         struct srpc_server_rpc  *fw_active_srpc;
 118 } sfw_data;
 119
 120 /* forward ref's */
 121 static int sfw_stop_batch(struct sfw_batch *tsb, int force);
 122 static void sfw_destroy_session(struct sfw_session *sn);
 123
 124 static inline struct sfw_test_case *
 125 sfw_find_test_case(int id)
 126 {
 127         struct sfw_test_case *tsc;
 128
 129         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 130         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 131
 132         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 133                 if (tsc->tsc_srv_service->sv_id == id)
 134                         return tsc;
 135         }
 136
 137         return NULL;
 138 }
 139
 140 static int
 141 sfw_register_test(struct srpc_service *service,
 142                   struct sfw_test_client_ops *cliops)
 143 {
 144         struct sfw_test_case *tsc;
 145
 146         if (sfw_find_test_case(service->sv_id) != NULL) {
 147                 CERROR ("Failed to register test %s (%d)\n",
 148                         service->sv_name, service->sv_id);
 149                 return -EEXIST;
 150         }
 151
 152         LIBCFS_ALLOC(tsc, sizeof(*tsc));
 153         if (tsc == NULL)
 154                 return -ENOMEM;
 155
 156         tsc->tsc_cli_ops     = cliops;
 157         tsc->tsc_srv_service = service;
 158
 159         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 160         return 0;
 161 }
 162
 163 static void
 164 sfw_add_session_timer (void)
 165 {
 166         struct sfw_session *sn = sfw_data.fw_session;
 167         struct stt_timer *timer = &sn->sn_timer;
 168
 169         LASSERT (!sfw_data.fw_shuttingdown);
 170
 171         if (sn == NULL || sn->sn_timeout == 0)
 172                 return;
 173
 174         LASSERT (!sn->sn_timer_active);
 175
 176         sn->sn_timer_active = 1;
 177         timer->stt_expires = ktime_get_real_seconds()+ sn->sn_timeout;
 178         stt_add_timer(timer);
 179 }
 180
 181 static int
 182 sfw_del_session_timer (void)
 183 {
 184         struct sfw_session *sn = sfw_data.fw_session;
 185
 186         if (sn == NULL || !sn->sn_timer_active)
 187                 return 0;
 188
 189         LASSERT (sn->sn_timeout != 0);
 190
 191         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 192                 sn->sn_timer_active = 0;
 193                 return 0;
 194         }
 195
 196         return EBUSY; /* racing with sfw_session_expired() */
 197 }
 198
 199 /* called with sfw_data.fw_lock held */
 200 static void
 201 sfw_deactivate_session (void)
 202 __must_hold(&sfw_data.fw_lock)
 203 {
 204         struct sfw_session *sn = sfw_data.fw_session;
 205         int            nactive = 0;
 206         struct sfw_batch *tsb;
 207         struct sfw_test_case *tsc;
 208
 209         if (sn == NULL) return;
 210
 211         LASSERT(!sn->sn_timer_active);
 212
 213         sfw_data.fw_session = NULL;
 214         atomic_inc(&sfw_data.fw_nzombies);
 215         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 216
 217         spin_unlock(&sfw_data.fw_lock);
 218
 219         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 220                 srpc_abort_service(tsc->tsc_srv_service);
 221         }
 222
 223         spin_lock(&sfw_data.fw_lock);
 224
 225         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 226                 if (sfw_batch_active(tsb)) {
 227                         nactive++;
 228                         sfw_stop_batch(tsb, 1);
 229                 }
 230         }
 231
 232         if (nactive != 0)
 233                 return; /* wait for active batches to stop */
 234
 235         list_del_init(&sn->sn_list);
 236         spin_unlock(&sfw_data.fw_lock);
 237
 238         sfw_destroy_session(sn);
 239
 240         spin_lock(&sfw_data.fw_lock);
 241 }
 242
 243
 244 static void
 245 sfw_session_expired (void *data)
 246 {
 247         struct sfw_session *sn = data;
 248
 249         spin_lock(&sfw_data.fw_lock);
 250
 251         LASSERT (sn->sn_timer_active);
 252         LASSERT (sn == sfw_data.fw_session);
 253
 254         CWARN ("Session expired! sid: %s-%llu, name: %s\n",
 255                libcfs_nidstr(&sn->sn_id.ses_nid),
 256                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 257
 258         sn->sn_timer_active = 0;
 259         sfw_deactivate_session();
 260
 261         spin_unlock(&sfw_data.fw_lock);
 262 }
 263
 264 static inline void
 265 sfw_init_session(struct sfw_session *sn, struct lst_sid sid,
 266                  unsigned features, const char *name)
 267 {
 268         struct stt_timer *timer = &sn->sn_timer;
 269
 270         memset(sn, 0, sizeof(struct sfw_session));
 271         INIT_LIST_HEAD(&sn->sn_list);
 272         INIT_LIST_HEAD(&sn->sn_batches);
 273         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 274         atomic_set(&sn->sn_brw_errors, 0);
 275         atomic_set(&sn->sn_ping_errors, 0);
 276         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 277
 278         sn->sn_timer_active = 0;
 279         sn->sn_id.ses_stamp = sid.ses_stamp;
 280         lnet_nid4_to_nid(sid.ses_nid, &sn->sn_id.ses_nid);
 281         sn->sn_features = features;
 282         sn->sn_timeout = session_timeout;
 283         sn->sn_started = ktime_get();
 284
 285         timer->stt_data = sn;
 286         timer->stt_func = sfw_session_expired;
 287         INIT_LIST_HEAD(&timer->stt_list);
 288 }
 289
 290 /* completion handler for incoming framework RPCs */
 291 static void
 292 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 293 {
 294         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 295         int                     status  = rpc->srpc_status;
 296
 297         CDEBUG (D_NET,
 298                 "Incoming framework RPC done: "
 299                 "service %s, peer %s, status %s:%d\n",
 300                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 301                 swi_state2str(rpc->srpc_wi.swi_state),
 302                 status);
 303
 304         if (rpc->srpc_bulk != NULL)
 305                 sfw_free_pages(rpc);
 306 }
 307
 308 static void
 309 sfw_client_rpc_fini(struct srpc_client_rpc *rpc)
 310 {
 311         LASSERT(rpc->crpc_bulk.bk_niov == 0);
 312         LASSERT(list_empty(&rpc->crpc_list));
 313         LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
 314
 315         CDEBUG(D_NET, "Outgoing framework RPC done: "
 316                "service %d, peer %s, status %s:%d:%d\n",
 317                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 318                swi_state2str(rpc->crpc_wi.swi_state),
 319                rpc->crpc_aborted, rpc->crpc_status);
 320
 321         spin_lock(&sfw_data.fw_lock);
 322
 323         /* my callers must finish all RPCs before shutting me down */
 324         LASSERT(!sfw_data.fw_shuttingdown);
 325         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 326
 327         spin_unlock(&sfw_data.fw_lock);
 328 }
 329
 330 static struct sfw_batch *
 331 sfw_find_batch(struct lst_bid bid)
 332 {
 333         struct sfw_session *sn = sfw_data.fw_session;
 334         struct sfw_batch *bat;
 335
 336         LASSERT(sn != NULL);
 337
 338         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 339                 if (bat->bat_id.bat_id == bid.bat_id)
 340                         return bat;
 341         }
 342
 343         return NULL;
 344 }
 345
 346 static struct sfw_batch *
 347 sfw_bid2batch(struct lst_bid bid)
 348 {
 349         struct sfw_session *sn = sfw_data.fw_session;
 350         struct sfw_batch *bat;
 351
 352         LASSERT (sn != NULL);
 353
 354         bat = sfw_find_batch(bid);
 355         if (bat != NULL)
 356                 return bat;
 357
 358         LIBCFS_ALLOC(bat, sizeof(*bat));
 359         if (bat == NULL)
 360                 return NULL;
 361
 362         bat->bat_error   = 0;
 363         bat->bat_session = sn;
 364         bat->bat_id      = bid;
 365         atomic_set(&bat->bat_nactive, 0);
 366         INIT_LIST_HEAD(&bat->bat_tests);
 367
 368         list_add_tail(&bat->bat_list, &sn->sn_batches);
 369         return bat;
 370 }
 371
 372 static struct lst_sid get_old_sid(struct sfw_session *sn)
 373 {
 374         struct lst_sid sid = { .ses_nid = LNET_NID_ANY, .ses_stamp = -1 };
 375
 376         if (sn) {
 377                 sid.ses_stamp = sn->sn_id.ses_stamp;
 378                 sid.ses_nid = lnet_nid_to_nid4(&sn->sn_id.ses_nid);
 379         }
 380
 381         return sid;
 382 }
 383
 384 static int
 385 sfw_get_stats(struct srpc_stat_reqst *request, struct srpc_stat_reply *reply)
 386 {
 387         struct sfw_session *sn = sfw_data.fw_session;
 388         struct sfw_counters *cnt = &reply->str_fw;
 389         struct sfw_batch *bat;
 390
 391         reply->str_sid = get_old_sid(sn);
 392
 393         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 394                 reply->str_status = EINVAL;
 395                 return 0;
 396         }
 397
 398         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 399                 reply->str_status = ESRCH;
 400                 return 0;
 401         }
 402
 403         lnet_counters_get_common(&reply->str_lnet);
 404         srpc_get_counters(&reply->str_rpc);
 405
 406         /* send over the msecs since the session was started
 407          - with 32 bits to send, this is ~49 days */
 408         cnt->running_ms = ktime_ms_delta(ktime_get(), sn->sn_started);
 409         cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
 410         cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
 411         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 412
 413         cnt->active_batches = 0;
 414         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 415                 if (atomic_read(&bat->bat_nactive) > 0)
 416                         cnt->active_batches++;
 417         }
 418
 419         reply->str_status = 0;
 420         return 0;
 421 }
 422
 423 int
 424 sfw_make_session(struct srpc_mksn_reqst *request, struct srpc_mksn_reply *reply)
 425 {
 426         struct sfw_session *sn = sfw_data.fw_session;
 427         struct srpc_msg *msg = container_of(request, struct srpc_msg,
 428                                             msg_body.mksn_reqst);
 429         int cplen = 0;
 430
 431         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 432                 reply->mksn_sid = get_old_sid(sn);
 433                 reply->mksn_status = EINVAL;
 434                 return 0;
 435         }
 436
 437         if (sn != NULL) {
 438                 reply->mksn_status  = 0;
 439                 reply->mksn_sid = get_old_sid(sn);
 440                 reply->mksn_timeout = sn->sn_timeout;
 441
 442                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 443                         atomic_inc(&sn->sn_refcount);
 444                         return 0;
 445                 }
 446
 447                 if (!request->mksn_force) {
 448                         reply->mksn_status = EBUSY;
 449                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 450                                         sizeof(reply->mksn_name));
 451                         if (cplen >= sizeof(reply->mksn_name))
 452                                 return -E2BIG;
 453                         return 0;
 454                 }
 455         }
 456
 457         /* reject the request if it requires unknown features
 458          * NB: old version will always accept all features because it's not
 459          * aware of struct srpc_msg::msg_ses_feats, it's a defect but it's also
 460          * harmless because it will return zero feature to console, and it's
 461          * console's responsibility to make sure all nodes in a session have
 462          * same feature mask. */
 463         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 464                 reply->mksn_status = EPROTO;
 465                 return 0;
 466         }
 467
 468         /* brand new or create by force */
 469         LIBCFS_ALLOC(sn, sizeof(*sn));
 470         if (sn == NULL) {
 471                 CERROR("dropping RPC mksn under memory pressure\n");
 472                 return -ENOMEM;
 473         }
 474
 475         sfw_init_session(sn, request->mksn_sid,
 476                          msg->msg_ses_feats, &request->mksn_name[0]);
 477
 478         spin_lock(&sfw_data.fw_lock);
 479
 480         sfw_deactivate_session();
 481         LASSERT(sfw_data.fw_session == NULL);
 482         sfw_data.fw_session = sn;
 483
 484         spin_unlock(&sfw_data.fw_lock);
 485
 486         reply->mksn_status  = 0;
 487         reply->mksn_sid = get_old_sid(sn);
 488         reply->mksn_timeout = sn->sn_timeout;
 489         return 0;
 490 }
 491
 492 static int
 493 sfw_remove_session(struct srpc_rmsn_reqst *request,
 494                    struct srpc_rmsn_reply *reply)
 495 {
 496         struct sfw_session *sn = sfw_data.fw_session;
 497
 498         reply->rmsn_sid = get_old_sid(sn);
 499
 500         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 501                 reply->rmsn_status = EINVAL;
 502                 return 0;
 503         }
 504
 505         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 506                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 507                 return 0;
 508         }
 509
 510         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 511                 reply->rmsn_status = 0;
 512                 return 0;
 513         }
 514
 515         spin_lock(&sfw_data.fw_lock);
 516         sfw_deactivate_session();
 517         spin_unlock(&sfw_data.fw_lock);
 518
 519         reply->rmsn_status = 0;
 520         reply->rmsn_sid = get_old_sid(NULL);
 521         LASSERT(sfw_data.fw_session == NULL);
 522         return 0;
 523 }
 524
 525 static int
 526 sfw_debug_session(struct srpc_debug_reqst *request,
 527                   struct srpc_debug_reply *reply)
 528 {
 529         struct sfw_session *sn = sfw_data.fw_session;
 530
 531         if (sn == NULL) {
 532                 reply->dbg_status = ESRCH;
 533                 reply->dbg_sid = get_old_sid(NULL);
 534                 return 0;
 535         }
 536
 537         reply->dbg_status  = 0;
 538         reply->dbg_sid = get_old_sid(sn);
 539         reply->dbg_timeout = sn->sn_timeout;
 540         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 541             >= sizeof(reply->dbg_name))
 542                 return -E2BIG;
 543
 544         return 0;
 545 }
 546
 547 static void
 548 sfw_test_rpc_fini(struct srpc_client_rpc *rpc)
 549 {
 550         struct sfw_test_unit *tsu = rpc->crpc_priv;
 551         struct sfw_test_instance *tsi = tsu->tsu_instance;
 552
 553         /* Called with hold of tsi->tsi_lock */
 554         LASSERT(list_empty(&rpc->crpc_list));
 555         rpc->crpc_wi.swi_state = SWI_STATE_DONE;
 556         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 557 }
 558
 559 static inline int
 560 sfw_test_buffers(struct sfw_test_instance *tsi)
 561 {
 562         struct sfw_test_case    *tsc;
 563         struct srpc_service     *svc;
 564         int                     nbuf;
 565
 566         LASSERT(tsi != NULL);
 567         tsc = sfw_find_test_case(tsi->tsi_service);
 568         LASSERT(tsc != NULL);
 569         svc = tsc->tsc_srv_service;
 570         LASSERT(svc != NULL);
 571
 572         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 573         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 574 }
 575
 576 static int
 577 sfw_load_test(struct sfw_test_instance *tsi)
 578 {
 579         struct sfw_test_case    *tsc;
 580         struct srpc_service     *svc;
 581         int                     nbuf;
 582         int                     rc;
 583
 584         LASSERT(tsi != NULL);
 585         tsc = sfw_find_test_case(tsi->tsi_service);
 586         nbuf = sfw_test_buffers(tsi);
 587         LASSERT(tsc != NULL);
 588         svc = tsc->tsc_srv_service;
 589
 590         if (tsi->tsi_is_client) {
 591                 tsi->tsi_ops = tsc->tsc_cli_ops;
 592                 return 0;
 593         }
 594
 595         rc = srpc_service_add_buffers(svc, nbuf);
 596         if (rc != 0) {
 597                 CWARN("Failed to reserve enough buffers: "
 598                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 599                 /* NB: this error handler is not strictly correct, because
 600                  * it may release more buffers than already allocated,
 601                  * but it doesn't matter because request portal should
 602                  * be lazy portal and will grow buffers if necessary. */
 603                 srpc_service_remove_buffers(svc, nbuf);
 604                 return -ENOMEM;
 605         }
 606
 607         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 608                nbuf * (srpc_serv_is_framework(svc) ?
 609                        1 : cfs_cpt_number(cfs_cpt_tab)), svc->sv_name);
 610         return 0;
 611 }
 612
 613 static void
 614 sfw_unload_test(struct sfw_test_instance *tsi)
 615 {
 616         struct sfw_test_case *tsc;
 617
 618         LASSERT(tsi != NULL);
 619         tsc = sfw_find_test_case(tsi->tsi_service);
 620         LASSERT(tsc != NULL);
 621
 622         if (tsi->tsi_is_client)
 623                 return;
 624
 625         /* shrink buffers, because request portal is lazy portal
 626          * which can grow buffers at runtime so we may leave
 627          * some buffers behind, but never mind... */
 628         srpc_service_remove_buffers(tsc->tsc_srv_service,
 629                                     sfw_test_buffers(tsi));
 630 }
 631
 632 static void
 633 sfw_destroy_test_instance(struct sfw_test_instance *tsi)
 634 {
 635         struct srpc_client_rpc *rpc;
 636         struct sfw_test_unit *tsu;
 637
 638         if (!tsi->tsi_is_client) goto clean;
 639
 640         tsi->tsi_ops->tso_fini(tsi);
 641
 642         LASSERT(!tsi->tsi_stopping);
 643         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 644         LASSERT(!sfw_test_active(tsi));
 645
 646         while (!list_empty(&tsi->tsi_units)) {
 647                 tsu = list_first_entry(&tsi->tsi_units,
 648                                        struct sfw_test_unit, tsu_list);
 649                 list_del(&tsu->tsu_list);
 650                 LIBCFS_FREE(tsu, sizeof(*tsu));
 651         }
 652
 653         while (!list_empty(&tsi->tsi_free_rpcs)) {
 654                 rpc = list_first_entry(&tsi->tsi_free_rpcs,
 655                                        struct srpc_client_rpc, crpc_list);
 656                 list_del(&rpc->crpc_list);
 657                 swi_cancel_workitem(&rpc->crpc_wi);
 658                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 659         }
 660
 661 clean:
 662         sfw_unload_test(tsi);
 663         LIBCFS_FREE(tsi, sizeof(*tsi));
 664 }
 665
 666 static void
 667 sfw_destroy_batch(struct sfw_batch *tsb)
 668 {
 669         struct sfw_test_instance *tsi;
 670
 671         LASSERT(!sfw_batch_active(tsb));
 672         LASSERT(list_empty(&tsb->bat_list));
 673
 674         while (!list_empty(&tsb->bat_tests)) {
 675                 tsi = list_first_entry(&tsb->bat_tests,
 676                                        struct sfw_test_instance, tsi_list);
 677                 list_del_init(&tsi->tsi_list);
 678                 sfw_destroy_test_instance(tsi);
 679         }
 680
 681         LIBCFS_FREE(tsb, sizeof(*tsb));
 682 }
 683
 684 static void
 685 sfw_destroy_session(struct sfw_session *sn)
 686 {
 687         struct sfw_batch *batch;
 688
 689         LASSERT(list_empty(&sn->sn_list));
 690         LASSERT(sn != sfw_data.fw_session);
 691
 692         while (!list_empty(&sn->sn_batches)) {
 693                 batch = list_first_entry(&sn->sn_batches,
 694                                          struct sfw_batch, bat_list);
 695                 list_del_init(&batch->bat_list);
 696                 sfw_destroy_batch(batch);
 697         }
 698
 699         LIBCFS_FREE(sn, sizeof(*sn));
 700         atomic_dec(&sfw_data.fw_nzombies);
 701 }
 702
 703 static void
 704 sfw_unpack_addtest_req(struct srpc_msg *msg)
 705 {
 706         struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
 707
 708         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 709         LASSERT (req->tsr_is_client);
 710
 711         if (msg->msg_magic == SRPC_MSG_MAGIC)
 712                 return; /* no flipping needed */
 713
 714         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 715
 716         if (req->tsr_service == SRPC_SERVICE_BRW) {
 717                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 718                         struct test_bulk_req *bulk = &req->tsr_u.bulk_v0;
 719
 720                         __swab32s(&bulk->blk_opc);
 721                         __swab32s(&bulk->blk_npg);
 722                         __swab32s(&bulk->blk_flags);
 723
 724                 } else {
 725                         struct test_bulk_req_v1 *bulk = &req->tsr_u.bulk_v1;
 726
 727                         __swab16s(&bulk->blk_opc);
 728                         __swab16s(&bulk->blk_flags);
 729                         __swab32s(&bulk->blk_offset);
 730                         __swab32s(&bulk->blk_len);
 731                 }
 732
 733                 return;
 734         }
 735
 736         if (req->tsr_service == SRPC_SERVICE_PING) {
 737                 struct test_ping_req *ping = &req->tsr_u.ping;
 738
 739                 __swab32s(&ping->png_size);
 740                 __swab32s(&ping->png_flags);
 741                 return;
 742         }
 743
 744         LBUG();
 745 }
 746
 747 static int
 748 sfw_add_test_instance(struct sfw_batch *tsb, struct srpc_server_rpc *rpc)
 749 {
 750         struct srpc_msg *msg = &rpc->srpc_reqstbuf->buf_msg;
 751         struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
 752         struct srpc_bulk *bk = rpc->srpc_bulk;
 753         int ndest = req->tsr_ndest;
 754         struct sfw_test_unit *tsu;
 755         struct sfw_test_instance *tsi;
 756         int i;
 757         int rc;
 758
 759         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 760         if (tsi == NULL) {
 761                 CERROR ("Can't allocate test instance for batch: %llu\n",
 762                         tsb->bat_id.bat_id);
 763                 return -ENOMEM;
 764         }
 765
 766         spin_lock_init(&tsi->tsi_lock);
 767         atomic_set(&tsi->tsi_nactive, 0);
 768         INIT_LIST_HEAD(&tsi->tsi_units);
 769         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 770         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 771
 772         tsi->tsi_stopping      = 0;
 773         tsi->tsi_batch         = tsb;
 774         tsi->tsi_loop          = req->tsr_loop;
 775         tsi->tsi_concur        = req->tsr_concur;
 776         tsi->tsi_service       = req->tsr_service;
 777         tsi->tsi_is_client     = !!(req->tsr_is_client);
 778         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 779
 780         rc = sfw_load_test(tsi);
 781         if (rc != 0) {
 782                 LIBCFS_FREE(tsi, sizeof(*tsi));
 783                 return rc;
 784         }
 785
 786         LASSERT (!sfw_batch_active(tsb));
 787
 788         if (!tsi->tsi_is_client) {
 789                 /* it's test server, just add it to tsb */
 790                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 791                 return 0;
 792         }
 793
 794         LASSERT (bk != NULL);
 795         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 796         LASSERT((unsigned int)bk->bk_len >=
 797                 sizeof(struct lnet_process_id_packed) * ndest);
 798
 799         sfw_unpack_addtest_req(msg);
 800         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 801
 802         for (i = 0; i < ndest; i++) {
 803                 struct lnet_process_id_packed *dests;
 804                 struct lnet_process_id_packed  id;
 805                 int                       j;
 806
 807                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].bv_page);
 808                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 809                 id = dests[i % SFW_ID_PER_PAGE];
 810                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 811                         sfw_unpack_id(id);
 812
 813                 for (j = 0; j < tsi->tsi_concur; j++) {
 814                         LIBCFS_ALLOC(tsu, sizeof(*tsu));
 815                         if (tsu == NULL) {
 816                                 rc = -ENOMEM;
 817                                 CERROR ("Can't allocate tsu for %d\n",
 818                                         tsi->tsi_service);
 819                                 goto error;
 820                         }
 821
 822                         tsu->tsu_dest.nid = id.nid;
 823                         tsu->tsu_dest.pid = id.pid;
 824                         tsu->tsu_instance = tsi;
 825                         tsu->tsu_private  = NULL;
 826                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 827                 }
 828         }
 829
 830         rc = tsi->tsi_ops->tso_init(tsi);
 831         if (rc == 0) {
 832                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 833                 return 0;
 834         }
 835
 836 error:
 837         LASSERT(rc != 0);
 838         sfw_destroy_test_instance(tsi);
 839         return rc;
 840 }
 841
 842 static void
 843 sfw_test_unit_done(struct sfw_test_unit *tsu)
 844 {
 845         struct sfw_test_instance *tsi = tsu->tsu_instance;
 846         struct sfw_batch *tsb = tsi->tsi_batch;
 847         struct sfw_session *sn = tsb->bat_session;
 848
 849         LASSERT (sfw_test_active(tsi));
 850
 851         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 852                 return;
 853
 854         /* the test instance is done */
 855         spin_lock(&tsi->tsi_lock);
 856
 857         tsi->tsi_stopping = 0;
 858
 859         spin_unlock(&tsi->tsi_lock);
 860
 861         spin_lock(&sfw_data.fw_lock);
 862
 863         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 864             sn == sfw_data.fw_session) {                  /* sn also active */
 865                 spin_unlock(&sfw_data.fw_lock);
 866                 return;
 867         }
 868
 869         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 870
 871         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 872                 if (sfw_batch_active(tsb)) {
 873                         spin_unlock(&sfw_data.fw_lock);
 874                         return;
 875                 }
 876         }
 877
 878         list_del_init(&sn->sn_list);
 879         spin_unlock(&sfw_data.fw_lock);
 880
 881         sfw_destroy_session(sn);
 882 }
 883
 884 static void
 885 sfw_test_rpc_done(struct srpc_client_rpc *rpc)
 886 {
 887         struct sfw_test_unit *tsu = rpc->crpc_priv;
 888         struct sfw_test_instance *tsi = tsu->tsu_instance;
 889         int                  done = 0;
 890
 891         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 892
 893         spin_lock(&tsi->tsi_lock);
 894
 895         LASSERT(sfw_test_active(tsi));
 896         LASSERT(!list_empty(&rpc->crpc_list));
 897
 898         list_del_init(&rpc->crpc_list);
 899
 900         /* batch is stopping or loop is done or get error */
 901         if (tsi->tsi_stopping ||
 902             tsu->tsu_loop == 0 ||
 903             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 904                 done = 1;
 905
 906         /* dec ref for poster */
 907         srpc_client_rpc_decref(rpc);
 908
 909         spin_unlock(&tsi->tsi_lock);
 910
 911         if (!done) {
 912                 swi_schedule_workitem(&tsu->tsu_worker);
 913                 return;
 914         }
 915
 916         sfw_test_unit_done(tsu);
 917 }
 918
 919 int
 920 sfw_create_test_rpc(struct sfw_test_unit *tsu, struct lnet_process_id peer,
 921                     unsigned features, int nblk, int blklen,
 922                     struct srpc_client_rpc **rpcpp)
 923 {
 924         struct srpc_client_rpc *rpc = NULL;
 925         struct sfw_test_instance *tsi = tsu->tsu_instance;
 926
 927         spin_lock(&tsi->tsi_lock);
 928
 929         LASSERT (sfw_test_active(tsi));
 930
 931         if (!list_empty(&tsi->tsi_free_rpcs)) {
 932                 /* pick request from buffer */
 933                 rpc = list_first_entry(&tsi->tsi_free_rpcs,
 934                                        struct srpc_client_rpc, crpc_list);
 935                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 936                 list_del_init(&rpc->crpc_list);
 937         }
 938
 939         spin_unlock(&tsi->tsi_lock);
 940
 941         if (rpc == NULL) {
 942                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 943                                              blklen, sfw_test_rpc_done,
 944                                              sfw_test_rpc_fini, tsu);
 945         } else {
 946                 swi_cancel_workitem(&rpc->crpc_wi);
 947                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 948                                      blklen, sfw_test_rpc_done,
 949                                      sfw_test_rpc_fini, tsu);
 950         }
 951
 952         if (rpc == NULL) {
 953                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 954                 return -ENOMEM;
 955         }
 956
 957         rpc->crpc_reqstmsg.msg_ses_feats = features;
 958         *rpcpp = rpc;
 959
 960         return 0;
 961 }
 962
 963 static void
 964 sfw_run_test(struct swi_workitem *wi)
 965 {
 966         struct sfw_test_unit *tsu = container_of(wi, struct sfw_test_unit, tsu_worker);
 967         struct sfw_test_instance *tsi = tsu->tsu_instance;
 968         struct srpc_client_rpc *rpc = NULL;
 969
 970         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 971                 LASSERT(rpc == NULL);
 972                 wi->swi_state = SWI_STATE_DONE;
 973                 goto test_done;
 974         }
 975
 976         LASSERT(rpc != NULL);
 977
 978         spin_lock(&tsi->tsi_lock);
 979         if (wi->swi_state == SWI_STATE_DONE) {
 980                 spin_unlock(&tsi->tsi_lock);
 981                 return;
 982         }
 983
 984         if (tsi->tsi_stopping) {
 985                 wi->swi_state = SWI_STATE_DONE;
 986                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 987                 spin_unlock(&tsi->tsi_lock);
 988                 goto test_done;
 989         }
 990
 991         if (tsu->tsu_loop > 0)
 992                 tsu->tsu_loop--;
 993
 994         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 995         wi->swi_state = SWI_STATE_RUNNING;
 996         spin_unlock(&tsi->tsi_lock);
 997
 998         spin_lock(&rpc->crpc_lock);
 999         rpc->crpc_timeout = rpc_timeout;
1000         srpc_post_rpc(rpc);
1001         spin_unlock(&rpc->crpc_lock);
1002         return;
1003
1004 test_done:
1005         /*
1006          * No one can schedule me now since:
1007          * - previous RPC, if any, has done and
1008          * - no new RPC is initiated.
1009          * - my batch is still active; no one can run it again now.
1010          * Cancel pending schedules and prevent future schedule attempts:
1011          */
1012         sfw_test_unit_done(tsu);
1013 }
1014
1015 static int
1016 sfw_run_batch(struct sfw_batch *tsb)
1017 {
1018         struct swi_workitem *wi;
1019         struct sfw_test_unit *tsu;
1020         struct sfw_test_instance *tsi;
1021
1022         if (sfw_batch_active(tsb)) {
1023                 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1024                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1025                 return 0;
1026         }
1027
1028         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1029                 if (!tsi->tsi_is_client)        /* skip server instances */
1030                         continue;
1031
1032                 LASSERT(!tsi->tsi_stopping);
1033                 LASSERT(!sfw_test_active(tsi));
1034
1035                 atomic_inc(&tsb->bat_nactive);
1036
1037                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1038                         atomic_inc(&tsi->tsi_nactive);
1039                         tsu->tsu_loop = tsi->tsi_loop;
1040                         wi = &tsu->tsu_worker;
1041                         swi_init_workitem(wi, sfw_run_test,
1042                                           lst_test_wq[lnet_cpt_of_nid(
1043                                                               tsu->tsu_dest.nid,
1044                                                               NULL)]);
1045                         swi_schedule_workitem(wi);
1046                 }
1047         }
1048
1049         return 0;
1050 }
1051
1052 static int
1053 sfw_stop_batch(struct sfw_batch *tsb, int force)
1054 {
1055         struct sfw_test_instance *tsi;
1056         struct srpc_client_rpc *rpc;
1057
1058         if (!sfw_batch_active(tsb)) {
1059                 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1060                 return 0;
1061         }
1062
1063         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1064                 spin_lock(&tsi->tsi_lock);
1065
1066                 if (!tsi->tsi_is_client ||
1067                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1068                         spin_unlock(&tsi->tsi_lock);
1069                         continue;
1070                 }
1071
1072                 tsi->tsi_stopping = 1;
1073
1074                 if (!force) {
1075                         spin_unlock(&tsi->tsi_lock);
1076                         continue;
1077                 }
1078
1079                 /* abort launched rpcs in the test */
1080                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1081                         spin_lock(&rpc->crpc_lock);
1082
1083                         srpc_abort_rpc(rpc, -EINTR);
1084
1085                         spin_unlock(&rpc->crpc_lock);
1086                 }
1087
1088                 spin_unlock(&tsi->tsi_lock);
1089         }
1090
1091         return 0;
1092 }
1093
1094 static int
1095 sfw_query_batch(struct sfw_batch *tsb, int testidx,
1096                 struct srpc_batch_reply *reply)
1097 {
1098         struct sfw_test_instance *tsi;
1099
1100         if (testidx < 0)
1101                 return -EINVAL;
1102
1103         if (testidx == 0) {
1104                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1105                 return 0;
1106         }
1107
1108         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1109                 if (testidx-- > 1)
1110                         continue;
1111
1112                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1113                 return 0;
1114         }
1115
1116         return -ENOENT;
1117 }
1118
1119 void
1120 sfw_free_pages(struct srpc_server_rpc *rpc)
1121 {
1122         srpc_free_bulk(rpc->srpc_bulk);
1123         rpc->srpc_bulk = NULL;
1124 }
1125
1126 int
1127 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1128                 int sink)
1129 {
1130         LASSERT(rpc->srpc_bulk == NULL);
1131         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1132
1133         rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1134         if (rpc->srpc_bulk == NULL)
1135                 return -ENOMEM;
1136
1137         return 0;
1138 }
1139
1140 static int
1141 sfw_add_test(struct srpc_server_rpc *rpc)
1142 {
1143         struct sfw_session *sn = sfw_data.fw_session;
1144         struct srpc_test_reply *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1145         struct srpc_test_reqst *request;
1146         int rc;
1147         struct sfw_batch *bat;
1148
1149         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1150         reply->tsr_sid = get_old_sid(sn);
1151
1152         if (request->tsr_loop == 0 ||
1153             request->tsr_concur == 0 ||
1154             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1155             request->tsr_ndest > SFW_MAX_NDESTS ||
1156             (request->tsr_is_client && request->tsr_ndest == 0) ||
1157             request->tsr_concur > SFW_MAX_CONCUR ||
1158             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1159             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1160                 reply->tsr_status = EINVAL;
1161                 return 0;
1162         }
1163
1164         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1165             sfw_find_test_case(request->tsr_service) == NULL) {
1166                 reply->tsr_status = ENOENT;
1167                 return 0;
1168         }
1169
1170         bat = sfw_bid2batch(request->tsr_bid);
1171         if (bat == NULL) {
1172                 CERROR("dropping RPC %s from %s under memory pressure\n",
1173                         rpc->srpc_scd->scd_svc->sv_name,
1174                         libcfs_id2str(rpc->srpc_peer));
1175                 return -ENOMEM;
1176         }
1177
1178         if (sfw_batch_active(bat)) {
1179                 reply->tsr_status = EBUSY;
1180                 return 0;
1181         }
1182
1183         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1184                 /* rpc will be resumed later in sfw_bulk_ready */
1185                 int     npg = sfw_id_pages(request->tsr_ndest);
1186                 int     len;
1187
1188                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1189                         len = npg * PAGE_SIZE;
1190
1191                 } else  {
1192                         len = sizeof(struct lnet_process_id_packed) *
1193                               request->tsr_ndest;
1194                 }
1195
1196                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1197         }
1198
1199         rc = sfw_add_test_instance(bat, rpc);
1200         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1201                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1202                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1203                 request->tsr_is_client ? "client" : "server",
1204                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1205
1206         reply->tsr_status = (rc < 0) ? -rc : rc;
1207         return 0;
1208 }
1209
1210 static int
1211 sfw_control_batch(struct srpc_batch_reqst *request,
1212                   struct srpc_batch_reply *reply)
1213 {
1214         struct sfw_session *sn = sfw_data.fw_session;
1215         int            rc = 0;
1216         struct sfw_batch *bat;
1217
1218         reply->bar_sid = get_old_sid(sn);
1219
1220         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1221                 reply->bar_status = ESRCH;
1222                 return 0;
1223         }
1224
1225         bat = sfw_find_batch(request->bar_bid);
1226         if (bat == NULL) {
1227                 reply->bar_status = ENOENT;
1228                 return 0;
1229         }
1230
1231         switch (request->bar_opc) {
1232         case SRPC_BATCH_OPC_RUN:
1233                 rc = sfw_run_batch(bat);
1234                 break;
1235
1236         case SRPC_BATCH_OPC_STOP:
1237                 rc = sfw_stop_batch(bat, request->bar_arg);
1238                 break;
1239
1240         case SRPC_BATCH_OPC_QUERY:
1241                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1242                 break;
1243
1244         default:
1245                 return -EINVAL; /* drop it */
1246         }
1247
1248         reply->bar_status = (rc < 0) ? -rc : rc;
1249         return 0;
1250 }
1251
1252 static int
1253 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1254 {
1255         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1256         struct srpc_msg     *reply      = &rpc->srpc_replymsg;
1257         struct srpc_msg     *request    = &rpc->srpc_reqstbuf->buf_msg;
1258         unsigned        features = LST_FEATS_MASK;
1259         int             rc = 0;
1260
1261         LASSERT(sfw_data.fw_active_srpc == NULL);
1262         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1263
1264         spin_lock(&sfw_data.fw_lock);
1265
1266         if (sfw_data.fw_shuttingdown) {
1267                 spin_unlock(&sfw_data.fw_lock);
1268                 return -ESHUTDOWN;
1269         }
1270
1271         /* Remove timer to avoid racing with it or expiring active session */
1272         if (sfw_del_session_timer() != 0) {
1273                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1274                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1275                 spin_unlock(&sfw_data.fw_lock);
1276                 return -EAGAIN;
1277         }
1278
1279         sfw_data.fw_active_srpc = rpc;
1280         spin_unlock(&sfw_data.fw_lock);
1281
1282         sfw_unpack_message(request);
1283         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1284
1285         /* rpc module should have checked this */
1286         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1287
1288         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1289             sv->sv_id != SRPC_SERVICE_DEBUG) {
1290                 struct sfw_session *sn = sfw_data.fw_session;
1291
1292                 if (sn != NULL &&
1293                     sn->sn_features != request->msg_ses_feats) {
1294                         CNETERR("Features of framework RPC don't match "
1295                                 "features of current session: %x/%x\n",
1296                                 request->msg_ses_feats, sn->sn_features);
1297                         reply->msg_body.reply.status = EPROTO;
1298                         reply->msg_body.reply.sid.ses_stamp = sn->sn_id.ses_stamp;
1299                         reply->msg_body.reply.sid.ses_nid =
1300                                 lnet_nid_to_nid4(&sn->sn_id.ses_nid);
1301                         goto out;
1302                 }
1303
1304         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1305                 /* NB: at this point, old version will ignore features and
1306                  * create new session anyway, so console should be able
1307                  * to handle this */
1308                 reply->msg_body.reply.status = EPROTO;
1309                 goto out;
1310         }
1311
1312         switch(sv->sv_id) {
1313         default:
1314                 LBUG ();
1315         case SRPC_SERVICE_TEST:
1316                 rc = sfw_add_test(rpc);
1317                 break;
1318
1319         case SRPC_SERVICE_BATCH:
1320                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1321                                        &reply->msg_body.bat_reply);
1322                 break;
1323
1324         case SRPC_SERVICE_QUERY_STAT:
1325                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1326                                    &reply->msg_body.stat_reply);
1327                 break;
1328
1329         case SRPC_SERVICE_DEBUG:
1330                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1331                                        &reply->msg_body.dbg_reply);
1332                 break;
1333
1334         case SRPC_SERVICE_MAKE_SESSION:
1335                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1336                                       &reply->msg_body.mksn_reply);
1337                 break;
1338
1339         case SRPC_SERVICE_REMOVE_SESSION:
1340                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1341                                         &reply->msg_body.rmsn_reply);
1342                 break;
1343         }
1344
1345         if (sfw_data.fw_session != NULL)
1346                 features = sfw_data.fw_session->sn_features;
1347  out:
1348         reply->msg_ses_feats = features;
1349         rpc->srpc_done = sfw_server_rpc_done;
1350         spin_lock(&sfw_data.fw_lock);
1351
1352         if (!sfw_data.fw_shuttingdown)
1353                 sfw_add_session_timer();
1354
1355         sfw_data.fw_active_srpc = NULL;
1356         spin_unlock(&sfw_data.fw_lock);
1357         return rc;
1358 }
1359
1360 static int
1361 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1362 {
1363         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1364         int                     rc;
1365
1366         LASSERT(rpc->srpc_bulk != NULL);
1367         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1368         LASSERT(sfw_data.fw_active_srpc == NULL);
1369         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1370
1371         spin_lock(&sfw_data.fw_lock);
1372
1373         if (status != 0) {
1374                 CERROR("Bulk transfer failed for RPC: "
1375                        "service %s, peer %s, status %d\n",
1376                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1377                 spin_unlock(&sfw_data.fw_lock);
1378                 return -EIO;
1379         }
1380
1381         if (sfw_data.fw_shuttingdown) {
1382                 spin_unlock(&sfw_data.fw_lock);
1383                 return -ESHUTDOWN;
1384         }
1385
1386         if (sfw_del_session_timer() != 0) {
1387                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1388                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1389                 spin_unlock(&sfw_data.fw_lock);
1390                 return -EAGAIN;
1391         }
1392
1393         sfw_data.fw_active_srpc = rpc;
1394         spin_unlock(&sfw_data.fw_lock);
1395
1396         rc = sfw_add_test(rpc);
1397
1398         spin_lock(&sfw_data.fw_lock);
1399
1400         if (!sfw_data.fw_shuttingdown)
1401                 sfw_add_session_timer();
1402
1403         sfw_data.fw_active_srpc = NULL;
1404         spin_unlock(&sfw_data.fw_lock);
1405         return rc;
1406 }
1407
1408 struct srpc_client_rpc *
1409 sfw_create_rpc(struct lnet_process_id peer, int service,
1410                unsigned features, int nbulkiov, int bulklen,
1411                void (*done)(struct srpc_client_rpc *), void *priv)
1412 {
1413         struct srpc_client_rpc *rpc = NULL;
1414
1415         spin_lock(&sfw_data.fw_lock);
1416
1417         LASSERT (!sfw_data.fw_shuttingdown);
1418         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1419
1420         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1421                 rpc = list_first_entry(&sfw_data.fw_zombie_rpcs,
1422                                        struct srpc_client_rpc, crpc_list);
1423                 list_del(&rpc->crpc_list);
1424         }
1425         spin_unlock(&sfw_data.fw_lock);
1426
1427         if (rpc) {
1428                 /* Ensure that rpc is done */
1429                 swi_cancel_workitem(&rpc->crpc_wi);
1430                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1431                                      done, sfw_client_rpc_fini, priv);
1432         } else {
1433                 rpc = srpc_create_client_rpc(peer, service,
1434                                              nbulkiov, bulklen, done,
1435                                              nbulkiov != 0 ?  NULL :
1436                                              sfw_client_rpc_fini,
1437                                              priv);
1438         }
1439
1440         if (rpc != NULL) /* "session" is concept in framework */
1441                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1442
1443         return rpc;
1444 }
1445
1446 void
1447 sfw_unpack_message(struct srpc_msg *msg)
1448 {
1449         if (msg->msg_magic == SRPC_MSG_MAGIC)
1450                 return; /* no flipping needed */
1451
1452         /* srpc module should guarantee I wouldn't get crap */
1453         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1454
1455         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1456                 struct srpc_stat_reqst *req = &msg->msg_body.stat_reqst;
1457
1458                 __swab32s(&req->str_type);
1459                 __swab64s(&req->str_rpyid);
1460                 sfw_unpack_sid(req->str_sid);
1461                 return;
1462         }
1463
1464         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1465                 struct srpc_stat_reply *rep = &msg->msg_body.stat_reply;
1466
1467                 __swab32s(&rep->str_status);
1468                 sfw_unpack_sid(rep->str_sid);
1469                 sfw_unpack_fw_counters(rep->str_fw);
1470                 sfw_unpack_rpc_counters(rep->str_rpc);
1471                 sfw_unpack_lnet_counters(rep->str_lnet);
1472                 return;
1473         }
1474
1475         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1476                 struct srpc_mksn_reqst *req = &msg->msg_body.mksn_reqst;
1477
1478                 __swab64s(&req->mksn_rpyid);
1479                 __swab32s(&req->mksn_force);
1480                 sfw_unpack_sid(req->mksn_sid);
1481                 return;
1482         }
1483
1484         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1485                 struct srpc_mksn_reply *rep = &msg->msg_body.mksn_reply;
1486
1487                 __swab32s(&rep->mksn_status);
1488                 __swab32s(&rep->mksn_timeout);
1489                 sfw_unpack_sid(rep->mksn_sid);
1490                 return;
1491         }
1492
1493         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1494                 struct srpc_rmsn_reqst *req = &msg->msg_body.rmsn_reqst;
1495
1496                 __swab64s(&req->rmsn_rpyid);
1497                 sfw_unpack_sid(req->rmsn_sid);
1498                 return;
1499         }
1500
1501         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1502                 struct srpc_rmsn_reply *rep = &msg->msg_body.rmsn_reply;
1503
1504                 __swab32s(&rep->rmsn_status);
1505                 sfw_unpack_sid(rep->rmsn_sid);
1506                 return;
1507         }
1508
1509         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1510                 struct srpc_debug_reqst *req = &msg->msg_body.dbg_reqst;
1511
1512                 __swab64s(&req->dbg_rpyid);
1513                 __swab32s(&req->dbg_flags);
1514                 sfw_unpack_sid(req->dbg_sid);
1515                 return;
1516         }
1517
1518         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1519                 struct srpc_debug_reply *rep = &msg->msg_body.dbg_reply;
1520
1521                 __swab32s(&rep->dbg_nbatch);
1522                 __swab32s(&rep->dbg_timeout);
1523                 sfw_unpack_sid(rep->dbg_sid);
1524                 return;
1525         }
1526
1527         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1528                 struct srpc_batch_reqst *req = &msg->msg_body.bat_reqst;
1529
1530                 __swab32s(&req->bar_opc);
1531                 __swab64s(&req->bar_rpyid);
1532                 __swab32s(&req->bar_testidx);
1533                 __swab32s(&req->bar_arg);
1534                 sfw_unpack_sid(req->bar_sid);
1535                 __swab64s(&req->bar_bid.bat_id);
1536                 return;
1537         }
1538
1539         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1540                 struct srpc_batch_reply *rep = &msg->msg_body.bat_reply;
1541
1542                 __swab32s(&rep->bar_status);
1543                 sfw_unpack_sid(rep->bar_sid);
1544                 return;
1545         }
1546
1547         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1548                 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
1549
1550                 __swab64s(&req->tsr_rpyid);
1551                 __swab64s(&req->tsr_bulkid);
1552                 __swab32s(&req->tsr_loop);
1553                 __swab32s(&req->tsr_ndest);
1554                 __swab32s(&req->tsr_concur);
1555                 __swab32s(&req->tsr_service);
1556                 sfw_unpack_sid(req->tsr_sid);
1557                 __swab64s(&req->tsr_bid.bat_id);
1558                 return;
1559         }
1560
1561         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1562                 struct srpc_test_reply *rep = &msg->msg_body.tes_reply;
1563
1564                 __swab32s(&rep->tsr_status);
1565                 sfw_unpack_sid(rep->tsr_sid);
1566                 return;
1567         }
1568
1569         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1570                 struct srpc_join_reqst *req = &msg->msg_body.join_reqst;
1571
1572                 __swab64s(&req->join_rpyid);
1573                 sfw_unpack_sid(req->join_sid);
1574                 return;
1575         }
1576
1577         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1578                 struct srpc_join_reply *rep = &msg->msg_body.join_reply;
1579
1580                 __swab32s(&rep->join_status);
1581                 __swab32s(&rep->join_timeout);
1582                 sfw_unpack_sid(rep->join_sid);
1583                 return;
1584         }
1585
1586         LBUG ();
1587 }
1588
1589 void
1590 sfw_abort_rpc(struct srpc_client_rpc *rpc)
1591 {
1592         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1593         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1594
1595         spin_lock(&rpc->crpc_lock);
1596         srpc_abort_rpc(rpc, -EINTR);
1597         spin_unlock(&rpc->crpc_lock);
1598 }
1599
1600 void
1601 sfw_post_rpc(struct srpc_client_rpc *rpc)
1602 {
1603         spin_lock(&rpc->crpc_lock);
1604
1605         LASSERT(!rpc->crpc_closed);
1606         LASSERT(!rpc->crpc_aborted);
1607         LASSERT(list_empty(&rpc->crpc_list));
1608         LASSERT(!sfw_data.fw_shuttingdown);
1609
1610         rpc->crpc_timeout = rpc_timeout;
1611         srpc_post_rpc(rpc);
1612
1613         spin_unlock(&rpc->crpc_lock);
1614 }
1615
1616 static struct srpc_service sfw_services[] = {
1617         { .sv_id = SRPC_SERVICE_DEBUG,          .sv_name = "debug", },
1618         { .sv_id = SRPC_SERVICE_QUERY_STAT,     .sv_name = "query stats", },
1619         { .sv_id = SRPC_SERVICE_MAKE_SESSION,   .sv_name = "make session", },
1620         { .sv_id = SRPC_SERVICE_REMOVE_SESSION, .sv_name = "remove session", },
1621         { .sv_id = SRPC_SERVICE_BATCH,          .sv_name = "batch service", },
1622         { .sv_id = SRPC_SERVICE_TEST,           .sv_name = "test service", },
1623         { .sv_id = 0, } };
1624
1625 int
1626 sfw_startup (void)
1627 {
1628         int              i;
1629         int              rc;
1630         int              error;
1631         struct srpc_service *sv;
1632         struct sfw_test_case *tsc;
1633
1634
1635         if (session_timeout < 0) {
1636                 CERROR ("Session timeout must be non-negative: %d\n",
1637                         session_timeout);
1638                 return -EINVAL;
1639         }
1640
1641         if (rpc_timeout < 0) {
1642                 CERROR ("RPC timeout must be non-negative: %d\n",
1643                         rpc_timeout);
1644                 return -EINVAL;
1645         }
1646
1647         if (session_timeout == 0)
1648                 CWARN ("Zero session_timeout specified "
1649                        "- test sessions never expire.\n");
1650
1651         if (rpc_timeout == 0)
1652                 CWARN ("Zero rpc_timeout specified "
1653                        "- test RPC never expire.\n");
1654
1655         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1656
1657         sfw_data.fw_session     = NULL;
1658         sfw_data.fw_active_srpc = NULL;
1659         spin_lock_init(&sfw_data.fw_lock);
1660         atomic_set(&sfw_data.fw_nzombies, 0);
1661         INIT_LIST_HEAD(&sfw_data.fw_tests);
1662         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1663         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1664
1665         brw_init_test_client();
1666         brw_init_test_service();
1667         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1668         LASSERT (rc == 0);
1669
1670         ping_init_test_client();
1671         ping_init_test_service();
1672         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1673         LASSERT (rc == 0);
1674
1675         error = 0;
1676         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1677                 sv = tsc->tsc_srv_service;
1678
1679                 rc = srpc_add_service(sv);
1680                 LASSERT(rc != -EBUSY);
1681                 if (rc != 0) {
1682                         CWARN("Failed to add %s service: %d\n",
1683                               sv->sv_name, rc);
1684                         error = rc;
1685                 }
1686         }
1687
1688         for (i = 0; ; i++) {
1689                 sv = &sfw_services[i];
1690                 if (sv->sv_name == NULL) break;
1691
1692                 sv->sv_bulk_ready = NULL;
1693                 sv->sv_handler    = sfw_handle_server_rpc;
1694                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1695                 if (sv->sv_id == SRPC_SERVICE_TEST)
1696                         sv->sv_bulk_ready = sfw_bulk_ready;
1697
1698                 rc = srpc_add_service(sv);
1699                 LASSERT (rc != -EBUSY);
1700                 if (rc != 0) {
1701                         CWARN ("Failed to add %s service: %d\n",
1702                                sv->sv_name, rc);
1703                         error = rc;
1704                 }
1705
1706                 /* about to sfw_shutdown, no need to add buffer */
1707                 if (error) continue;
1708
1709                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1710                 if (rc != 0) {
1711                         CWARN("Failed to reserve enough buffers: "
1712                               "service %s, %d needed: %d\n",
1713                               sv->sv_name, sv->sv_wi_total, rc);
1714                         error = -ENOMEM;
1715                 }
1716         }
1717
1718         if (error != 0)
1719                 sfw_shutdown();
1720         return error;
1721 }
1722
1723 void
1724 sfw_shutdown (void)
1725 {
1726         struct srpc_service *sv;
1727         struct sfw_test_case *tsc;
1728         int              i;
1729
1730         spin_lock(&sfw_data.fw_lock);
1731
1732         sfw_data.fw_shuttingdown = 1;
1733         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1734                        "waiting for active RPC to finish.\n");
1735
1736         if (sfw_del_session_timer() != 0)
1737                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1738                                "waiting for session timer to explode.\n");
1739
1740         sfw_deactivate_session();
1741         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1742                        sfw_data.fw_lock,
1743                        "waiting for %d zombie sessions to die.\n",
1744                        atomic_read(&sfw_data.fw_nzombies));
1745
1746         spin_unlock(&sfw_data.fw_lock);
1747
1748         for (i = 0; ; i++) {
1749                 sv = &sfw_services[i];
1750                 if (sv->sv_name == NULL)
1751                         break;
1752
1753                 srpc_shutdown_service(sv);
1754                 srpc_remove_service(sv);
1755         }
1756
1757         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1758                 sv = tsc->tsc_srv_service;
1759                 srpc_shutdown_service(sv);
1760                 srpc_remove_service(sv);
1761         }
1762
1763         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1764                 struct srpc_client_rpc *rpc;
1765
1766                 rpc = list_first_entry(&sfw_data.fw_zombie_rpcs,
1767                                        struct srpc_client_rpc, crpc_list);
1768                 list_del(&rpc->crpc_list);
1769
1770                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1771         }
1772
1773         for (i = 0; ; i++) {
1774                 sv = &sfw_services[i];
1775                 if (sv->sv_name == NULL)
1776                         break;
1777
1778                 srpc_wait_service_shutdown(sv);
1779         }
1780
1781         while (!list_empty(&sfw_data.fw_tests)) {
1782                 tsc = list_first_entry(&sfw_data.fw_tests,
1783                                        struct sfw_test_case, tsc_list);
1784
1785                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1786
1787                 list_del(&tsc->tsc_list);
1788                 LIBCFS_FREE(tsc, sizeof(*tsc));
1789         }
1790 }