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, 2014, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lnet/selftest/framework.c
  33  *
  34  * Author: Isaac Huang <isaac@clusterfs.com>
  35  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LNET
  39
  40 #include "selftest.h"
  41
  42 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
  43
  44 static int session_timeout = 100;
  45 module_param(session_timeout, int, 0444);
  46 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
  47
  48 static int rpc_timeout = 64;
  49 module_param(rpc_timeout, int, 0644);
  50 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
  51
  52 #define sfw_unpack_id(id)               \
  53 do {                                    \
  54         __swab64s(&(id).nid);           \
  55         __swab32s(&(id).pid);           \
  56 } while (0)
  57
  58 #define sfw_unpack_sid(sid)             \
  59 do {                                    \
  60         __swab64s(&(sid).ses_nid);      \
  61         __swab64s(&(sid).ses_stamp);    \
  62 } while (0)
  63
  64 #define sfw_unpack_fw_counters(fc)        \
  65 do {                                      \
  66         __swab32s(&(fc).running_ms);      \
  67         __swab32s(&(fc).active_batches);  \
  68         __swab32s(&(fc).zombie_sessions); \
  69         __swab32s(&(fc).brw_errors);      \
  70         __swab32s(&(fc).ping_errors);     \
  71 } while (0)
  72
  73 #define sfw_unpack_rpc_counters(rc)     \
  74 do {                                    \
  75         __swab32s(&(rc).errors);        \
  76         __swab32s(&(rc).rpcs_sent);     \
  77         __swab32s(&(rc).rpcs_rcvd);     \
  78         __swab32s(&(rc).rpcs_dropped);  \
  79         __swab32s(&(rc).rpcs_expired);  \
  80         __swab64s(&(rc).bulk_get);      \
  81         __swab64s(&(rc).bulk_put);      \
  82 } while (0)
  83
  84 #define sfw_unpack_lnet_counters(lc)    \
  85 do {                                    \
  86         __swab32s(&(lc).errors);        \
  87         __swab32s(&(lc).msgs_max);      \
  88         __swab32s(&(lc).msgs_alloc);    \
  89         __swab32s(&(lc).send_count);    \
  90         __swab32s(&(lc).recv_count);    \
  91         __swab32s(&(lc).drop_count);    \
  92         __swab32s(&(lc).route_count);   \
  93         __swab64s(&(lc).send_length);   \
  94         __swab64s(&(lc).recv_length);   \
  95         __swab64s(&(lc).drop_length);   \
  96         __swab64s(&(lc).route_length);  \
  97 } while (0)
  98
  99 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 100 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 101
 102 static struct smoketest_framework {
 103         /* RPCs to be recycled */
 104         struct list_head        fw_zombie_rpcs;
 105         /* stopping sessions */
 106         struct list_head        fw_zombie_sessions;
 107         /* registered test cases */
 108         struct list_head        fw_tests;
 109         /* # zombie sessions */
 110         atomic_t                fw_nzombies;
 111         /* serialise */
 112         spinlock_t              fw_lock;
 113         /* _the_ session */
 114         sfw_session_t           *fw_session;
 115         /* shutdown in progress */
 116         int                     fw_shuttingdown;
 117         /* running RPC */
 118         srpc_server_rpc_t       *fw_active_srpc;
 119 } sfw_data;
 120
 121 /* forward ref's */
 122 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 123 void sfw_destroy_session (sfw_session_t *sn);
 124
 125 static inline sfw_test_case_t *
 126 sfw_find_test_case(int id)
 127 {
 128         sfw_test_case_t *tsc;
 129
 130         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 131         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 132
 133         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 134                 if (tsc->tsc_srv_service->sv_id == id)
 135                         return tsc;
 136         }
 137
 138         return NULL;
 139 }
 140
 141 static int
 142 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 143 {
 144         sfw_test_case_t *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(sfw_test_case_t));
 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         sfw_session_t *sn = sfw_data.fw_session;
 167         stt_timer_t   *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 = cfs_time_add(sn->sn_timeout,
 178                                           cfs_time_current_sec());
 179         stt_add_timer(timer);
 180         return;
 181 }
 182
 183 static int
 184 sfw_del_session_timer (void)
 185 {
 186         sfw_session_t *sn = sfw_data.fw_session;
 187
 188         if (sn == NULL || !sn->sn_timer_active)
 189                 return 0;
 190
 191         LASSERT (sn->sn_timeout != 0);
 192
 193         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 194                 sn->sn_timer_active = 0;
 195                 return 0;
 196         }
 197
 198         return EBUSY; /* racing with sfw_session_expired() */
 199 }
 200
 201 /* called with sfw_data.fw_lock held */
 202 static void
 203 sfw_deactivate_session (void)
 204 __must_hold(&sfw_data.fw_lock)
 205 {
 206         sfw_session_t *sn = sfw_data.fw_session;
 207         int            nactive = 0;
 208         sfw_batch_t   *tsb;
 209         sfw_test_case_t *tsc;
 210
 211         if (sn == NULL) return;
 212
 213         LASSERT(!sn->sn_timer_active);
 214
 215         sfw_data.fw_session = NULL;
 216         atomic_inc(&sfw_data.fw_nzombies);
 217         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 218
 219         spin_unlock(&sfw_data.fw_lock);
 220
 221         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 222                 srpc_abort_service(tsc->tsc_srv_service);
 223         }
 224
 225         spin_lock(&sfw_data.fw_lock);
 226
 227         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 228                 if (sfw_batch_active(tsb)) {
 229                         nactive++;
 230                         sfw_stop_batch(tsb, 1);
 231                 }
 232         }
 233
 234         if (nactive != 0)
 235                 return; /* wait for active batches to stop */
 236
 237         list_del_init(&sn->sn_list);
 238         spin_unlock(&sfw_data.fw_lock);
 239
 240         sfw_destroy_session(sn);
 241
 242         spin_lock(&sfw_data.fw_lock);
 243 }
 244
 245
 246 static void
 247 sfw_session_expired (void *data)
 248 {
 249         sfw_session_t *sn = data;
 250
 251         spin_lock(&sfw_data.fw_lock);
 252
 253         LASSERT (sn->sn_timer_active);
 254         LASSERT (sn == sfw_data.fw_session);
 255
 256         CWARN ("Session expired! sid: %s-%llu, name: %s\n",
 257                libcfs_nid2str(sn->sn_id.ses_nid),
 258                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 259
 260         sn->sn_timer_active = 0;
 261         sfw_deactivate_session();
 262
 263         spin_unlock(&sfw_data.fw_lock);
 264 }
 265
 266 static inline void
 267 sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
 268                  unsigned features, const char *name)
 269 {
 270         stt_timer_t *timer = &sn->sn_timer;
 271
 272         memset(sn, 0, sizeof(sfw_session_t));
 273         INIT_LIST_HEAD(&sn->sn_list);
 274         INIT_LIST_HEAD(&sn->sn_batches);
 275         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 276         atomic_set(&sn->sn_brw_errors, 0);
 277         atomic_set(&sn->sn_ping_errors, 0);
 278         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 279
 280         sn->sn_timer_active = 0;
 281         sn->sn_id           = sid;
 282         sn->sn_features     = features;
 283         sn->sn_timeout      = session_timeout;
 284         sn->sn_started      = cfs_time_current();
 285
 286         timer->stt_data = sn;
 287         timer->stt_func = sfw_session_expired;
 288         INIT_LIST_HEAD(&timer->stt_list);
 289 }
 290
 291 /* completion handler for incoming framework RPCs */
 292 static void
 293 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 294 {
 295         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 296         int                     status  = rpc->srpc_status;
 297
 298         CDEBUG (D_NET,
 299                 "Incoming framework RPC done: "
 300                 "service %s, peer %s, status %s:%d\n",
 301                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 302                 swi_state2str(rpc->srpc_wi.swi_state),
 303                 status);
 304
 305         if (rpc->srpc_bulk != NULL)
 306                 sfw_free_pages(rpc);
 307         return;
 308 }
 309
 310 static void
 311 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 312 {
 313         LASSERT(rpc->crpc_bulk.bk_niov == 0);
 314         LASSERT(list_empty(&rpc->crpc_list));
 315         LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
 316
 317         CDEBUG(D_NET, "Outgoing framework RPC done: "
 318                "service %d, peer %s, status %s:%d:%d\n",
 319                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 320                swi_state2str(rpc->crpc_wi.swi_state),
 321                rpc->crpc_aborted, rpc->crpc_status);
 322
 323         spin_lock(&sfw_data.fw_lock);
 324
 325         /* my callers must finish all RPCs before shutting me down */
 326         LASSERT(!sfw_data.fw_shuttingdown);
 327         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 328
 329         spin_unlock(&sfw_data.fw_lock);
 330 }
 331
 332 static sfw_batch_t *
 333 sfw_find_batch (lst_bid_t bid)
 334 {
 335         sfw_session_t *sn = sfw_data.fw_session;
 336         sfw_batch_t   *bat;
 337
 338         LASSERT(sn != NULL);
 339
 340         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 341                 if (bat->bat_id.bat_id == bid.bat_id)
 342                         return bat;
 343         }
 344
 345         return NULL;
 346 }
 347
 348 static sfw_batch_t *
 349 sfw_bid2batch (lst_bid_t bid)
 350 {
 351         sfw_session_t *sn = sfw_data.fw_session;
 352         sfw_batch_t   *bat;
 353
 354         LASSERT (sn != NULL);
 355
 356         bat = sfw_find_batch(bid);
 357         if (bat != NULL)
 358                 return bat;
 359
 360         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 361         if (bat == NULL)
 362                 return NULL;
 363
 364         bat->bat_error   = 0;
 365         bat->bat_session = sn;
 366         bat->bat_id      = bid;
 367         atomic_set(&bat->bat_nactive, 0);
 368         INIT_LIST_HEAD(&bat->bat_tests);
 369
 370         list_add_tail(&bat->bat_list, &sn->sn_batches);
 371         return bat;
 372 }
 373
 374 static int
 375 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 376 {
 377         sfw_session_t  *sn = sfw_data.fw_session;
 378         sfw_counters_t *cnt = &reply->str_fw;
 379         sfw_batch_t    *bat;
 380         struct timeval  tv;
 381
 382         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 383
 384         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 385                 reply->str_status = EINVAL;
 386                 return 0;
 387         }
 388
 389         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 390                 reply->str_status = ESRCH;
 391                 return 0;
 392         }
 393
 394         lnet_counters_get(&reply->str_lnet);
 395         srpc_get_counters(&reply->str_rpc);
 396
 397         /* send over the msecs since the session was started
 398          - with 32 bits to send, this is ~49 days */
 399         cfs_duration_usec(cfs_time_sub(cfs_time_current(),
 400                                        sn->sn_started), &tv);
 401
 402         cnt->running_ms      = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
 403         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 404         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 405         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 406
 407         cnt->active_batches = 0;
 408         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 409                 if (atomic_read(&bat->bat_nactive) > 0)
 410                         cnt->active_batches++;
 411         }
 412
 413         reply->str_status = 0;
 414         return 0;
 415 }
 416
 417 int
 418 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 419 {
 420         sfw_session_t *sn = sfw_data.fw_session;
 421         srpc_msg_t    *msg = container_of(request, srpc_msg_t,
 422                                           msg_body.mksn_reqst);
 423         int            cplen = 0;
 424
 425         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 426                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 427                 reply->mksn_status = EINVAL;
 428                 return 0;
 429         }
 430
 431         if (sn != NULL) {
 432                 reply->mksn_status  = 0;
 433                 reply->mksn_sid     = sn->sn_id;
 434                 reply->mksn_timeout = sn->sn_timeout;
 435
 436                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 437                         atomic_inc(&sn->sn_refcount);
 438                         return 0;
 439                 }
 440
 441                 if (!request->mksn_force) {
 442                         reply->mksn_status = EBUSY;
 443                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 444                                         sizeof(reply->mksn_name));
 445                         if (cplen >= sizeof(reply->mksn_name))
 446                                 return -E2BIG;
 447                         return 0;
 448                 }
 449         }
 450
 451         /* reject the request if it requires unknown features
 452          * NB: old version will always accept all features because it's not
 453          * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
 454          * harmless because it will return zero feature to console, and it's
 455          * console's responsibility to make sure all nodes in a session have
 456          * same feature mask. */
 457         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 458                 reply->mksn_status = EPROTO;
 459                 return 0;
 460         }
 461
 462         /* brand new or create by force */
 463         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 464         if (sn == NULL) {
 465                 CERROR("dropping RPC mksn under memory pressure\n");
 466                 return -ENOMEM;
 467         }
 468
 469         sfw_init_session(sn, request->mksn_sid,
 470                          msg->msg_ses_feats, &request->mksn_name[0]);
 471
 472         spin_lock(&sfw_data.fw_lock);
 473
 474         sfw_deactivate_session();
 475         LASSERT(sfw_data.fw_session == NULL);
 476         sfw_data.fw_session = sn;
 477
 478         spin_unlock(&sfw_data.fw_lock);
 479
 480         reply->mksn_status  = 0;
 481         reply->mksn_sid     = sn->sn_id;
 482         reply->mksn_timeout = sn->sn_timeout;
 483         return 0;
 484 }
 485
 486 static int
 487 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 488 {
 489         sfw_session_t *sn = sfw_data.fw_session;
 490
 491         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 492
 493         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 494                 reply->rmsn_status = EINVAL;
 495                 return 0;
 496         }
 497
 498         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 499                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 500                 return 0;
 501         }
 502
 503         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 504                 reply->rmsn_status = 0;
 505                 return 0;
 506         }
 507
 508         spin_lock(&sfw_data.fw_lock);
 509         sfw_deactivate_session();
 510         spin_unlock(&sfw_data.fw_lock);
 511
 512         reply->rmsn_status = 0;
 513         reply->rmsn_sid    = LST_INVALID_SID;
 514         LASSERT(sfw_data.fw_session == NULL);
 515         return 0;
 516 }
 517
 518 static int
 519 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 520 {
 521         sfw_session_t *sn = sfw_data.fw_session;
 522
 523         if (sn == NULL) {
 524                 reply->dbg_status = ESRCH;
 525                 reply->dbg_sid    = LST_INVALID_SID;
 526                 return 0;
 527         }
 528
 529         reply->dbg_status  = 0;
 530         reply->dbg_sid     = sn->sn_id;
 531         reply->dbg_timeout = sn->sn_timeout;
 532         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 533             >= sizeof(reply->dbg_name))
 534                 return -E2BIG;
 535
 536         return 0;
 537 }
 538
 539 static void
 540 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 541 {
 542         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 543         sfw_test_instance_t *tsi = tsu->tsu_instance;
 544
 545         /* Called with hold of tsi->tsi_lock */
 546         LASSERT(list_empty(&rpc->crpc_list));
 547         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 548 }
 549
 550 static inline int
 551 sfw_test_buffers(sfw_test_instance_t *tsi)
 552 {
 553         struct sfw_test_case    *tsc;
 554         struct srpc_service     *svc;
 555         int                     nbuf;
 556
 557         LASSERT(tsi != NULL);
 558         tsc = sfw_find_test_case(tsi->tsi_service);
 559         LASSERT(tsc != NULL);
 560         svc = tsc->tsc_srv_service;
 561         LASSERT(svc != NULL);
 562
 563         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 564         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 565 }
 566
 567 static int
 568 sfw_load_test(struct sfw_test_instance *tsi)
 569 {
 570         struct sfw_test_case    *tsc;
 571         struct srpc_service     *svc;
 572         int                     nbuf;
 573         int                     rc;
 574
 575         LASSERT(tsi != NULL);
 576         tsc = sfw_find_test_case(tsi->tsi_service);
 577         nbuf = sfw_test_buffers(tsi);
 578         LASSERT(tsc != NULL);
 579         svc = tsc->tsc_srv_service;
 580
 581         if (tsi->tsi_is_client) {
 582                 tsi->tsi_ops = tsc->tsc_cli_ops;
 583                 return 0;
 584         }
 585
 586         rc = srpc_service_add_buffers(svc, nbuf);
 587         if (rc != 0) {
 588                 CWARN("Failed to reserve enough buffers: "
 589                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 590                 /* NB: this error handler is not strictly correct, because
 591                  * it may release more buffers than already allocated,
 592                  * but it doesn't matter because request portal should
 593                  * be lazy portal and will grow buffers if necessary. */
 594                 srpc_service_remove_buffers(svc, nbuf);
 595                 return -ENOMEM;
 596         }
 597
 598         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 599                nbuf * (srpc_serv_is_framework(svc) ?
 600                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 601         return 0;
 602 }
 603
 604 static void
 605 sfw_unload_test(struct sfw_test_instance *tsi)
 606 {
 607         struct sfw_test_case *tsc;
 608
 609         LASSERT(tsi != NULL);
 610         tsc = sfw_find_test_case(tsi->tsi_service);
 611         LASSERT(tsc != NULL);
 612
 613         if (tsi->tsi_is_client)
 614                 return;
 615
 616         /* shrink buffers, because request portal is lazy portal
 617          * which can grow buffers at runtime so we may leave
 618          * some buffers behind, but never mind... */
 619         srpc_service_remove_buffers(tsc->tsc_srv_service,
 620                                     sfw_test_buffers(tsi));
 621         return;
 622 }
 623
 624 static void
 625 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 626 {
 627         srpc_client_rpc_t *rpc;
 628         sfw_test_unit_t   *tsu;
 629
 630         if (!tsi->tsi_is_client) goto clean;
 631
 632         tsi->tsi_ops->tso_fini(tsi);
 633
 634         LASSERT(!tsi->tsi_stopping);
 635         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 636         LASSERT(!sfw_test_active(tsi));
 637
 638         while (!list_empty(&tsi->tsi_units)) {
 639                 tsu = list_entry(tsi->tsi_units.next,
 640                                  sfw_test_unit_t, tsu_list);
 641                 list_del(&tsu->tsu_list);
 642                 LIBCFS_FREE(tsu, sizeof(*tsu));
 643         }
 644
 645         while (!list_empty(&tsi->tsi_free_rpcs)) {
 646                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 647                                  srpc_client_rpc_t, crpc_list);
 648                 list_del(&rpc->crpc_list);
 649                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 650         }
 651
 652 clean:
 653         sfw_unload_test(tsi);
 654         LIBCFS_FREE(tsi, sizeof(*tsi));
 655         return;
 656 }
 657
 658 static void
 659 sfw_destroy_batch (sfw_batch_t *tsb)
 660 {
 661         sfw_test_instance_t *tsi;
 662
 663         LASSERT(!sfw_batch_active(tsb));
 664         LASSERT(list_empty(&tsb->bat_list));
 665
 666         while (!list_empty(&tsb->bat_tests)) {
 667                 tsi = list_entry(tsb->bat_tests.next,
 668                                  sfw_test_instance_t, tsi_list);
 669                 list_del_init(&tsi->tsi_list);
 670                 sfw_destroy_test_instance(tsi);
 671         }
 672
 673         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 674         return;
 675 }
 676
 677 void
 678 sfw_destroy_session (sfw_session_t *sn)
 679 {
 680         sfw_batch_t *batch;
 681
 682         LASSERT(list_empty(&sn->sn_list));
 683         LASSERT(sn != sfw_data.fw_session);
 684
 685         while (!list_empty(&sn->sn_batches)) {
 686                 batch = list_entry(sn->sn_batches.next,
 687                                    sfw_batch_t, bat_list);
 688                 list_del_init(&batch->bat_list);
 689                 sfw_destroy_batch(batch);
 690         }
 691
 692         LIBCFS_FREE(sn, sizeof(*sn));
 693         atomic_dec(&sfw_data.fw_nzombies);
 694         return;
 695 }
 696
 697 static void
 698 sfw_unpack_addtest_req(srpc_msg_t *msg)
 699 {
 700         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 701
 702         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 703         LASSERT (req->tsr_is_client);
 704
 705         if (msg->msg_magic == SRPC_MSG_MAGIC)
 706                 return; /* no flipping needed */
 707
 708         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 709
 710         if (req->tsr_service == SRPC_SERVICE_BRW) {
 711                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 712                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 713
 714                         __swab32s(&bulk->blk_opc);
 715                         __swab32s(&bulk->blk_npg);
 716                         __swab32s(&bulk->blk_flags);
 717
 718                 } else {
 719                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 720
 721                         __swab16s(&bulk->blk_opc);
 722                         __swab16s(&bulk->blk_flags);
 723                         __swab32s(&bulk->blk_offset);
 724                         __swab32s(&bulk->blk_len);
 725                 }
 726
 727                 return;
 728         }
 729
 730         if (req->tsr_service == SRPC_SERVICE_PING) {
 731                 test_ping_req_t *ping = &req->tsr_u.ping;
 732
 733                 __swab32s(&ping->png_size);
 734                 __swab32s(&ping->png_flags);
 735                 return;
 736         }
 737
 738         LBUG();
 739         return;
 740 }
 741
 742 static int
 743 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 744 {
 745         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 746         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 747         srpc_bulk_t         *bk = rpc->srpc_bulk;
 748         int                  ndest = req->tsr_ndest;
 749         sfw_test_unit_t     *tsu;
 750         sfw_test_instance_t *tsi;
 751         int                  i;
 752         int                  rc;
 753
 754         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 755         if (tsi == NULL) {
 756                 CERROR ("Can't allocate test instance for batch: %llu\n",
 757                         tsb->bat_id.bat_id);
 758                 return -ENOMEM;
 759         }
 760
 761         spin_lock_init(&tsi->tsi_lock);
 762         atomic_set(&tsi->tsi_nactive, 0);
 763         INIT_LIST_HEAD(&tsi->tsi_units);
 764         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 765         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 766
 767         tsi->tsi_stopping      = 0;
 768         tsi->tsi_batch         = tsb;
 769         tsi->tsi_loop          = req->tsr_loop;
 770         tsi->tsi_concur        = req->tsr_concur;
 771         tsi->tsi_service       = req->tsr_service;
 772         tsi->tsi_is_client     = !!(req->tsr_is_client);
 773         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 774
 775         rc = sfw_load_test(tsi);
 776         if (rc != 0) {
 777                 LIBCFS_FREE(tsi, sizeof(*tsi));
 778                 return rc;
 779         }
 780
 781         LASSERT (!sfw_batch_active(tsb));
 782
 783         if (!tsi->tsi_is_client) {
 784                 /* it's test server, just add it to tsb */
 785                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 786                 return 0;
 787         }
 788
 789         LASSERT (bk != NULL);
 790         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 791         LASSERT((unsigned int)bk->bk_len >=
 792                 sizeof(lnet_process_id_packed_t) * ndest);
 793
 794         sfw_unpack_addtest_req(msg);
 795         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 796
 797         for (i = 0; i < ndest; i++) {
 798                 lnet_process_id_packed_t *dests;
 799                 lnet_process_id_packed_t  id;
 800                 int                       j;
 801
 802                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 803                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 804                 id = dests[i % SFW_ID_PER_PAGE];
 805                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 806                         sfw_unpack_id(id);
 807
 808                 for (j = 0; j < tsi->tsi_concur; j++) {
 809                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 810                         if (tsu == NULL) {
 811                                 rc = -ENOMEM;
 812                                 CERROR ("Can't allocate tsu for %d\n",
 813                                         tsi->tsi_service);
 814                                 goto error;
 815                         }
 816
 817                         tsu->tsu_dest.nid = id.nid;
 818                         tsu->tsu_dest.pid = id.pid;
 819                         tsu->tsu_instance = tsi;
 820                         tsu->tsu_private  = NULL;
 821                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 822                 }
 823         }
 824
 825         rc = tsi->tsi_ops->tso_init(tsi);
 826         if (rc == 0) {
 827                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 828                 return 0;
 829         }
 830
 831 error:
 832         LASSERT(rc != 0);
 833         sfw_destroy_test_instance(tsi);
 834         return rc;
 835 }
 836
 837 static void
 838 sfw_test_unit_done (sfw_test_unit_t *tsu)
 839 {
 840         sfw_test_instance_t *tsi = tsu->tsu_instance;
 841         sfw_batch_t         *tsb = tsi->tsi_batch;
 842         sfw_session_t       *sn = tsb->bat_session;
 843
 844         LASSERT (sfw_test_active(tsi));
 845
 846         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 847                 return;
 848
 849         /* the test instance is done */
 850         spin_lock(&tsi->tsi_lock);
 851
 852         tsi->tsi_stopping = 0;
 853
 854         spin_unlock(&tsi->tsi_lock);
 855
 856         spin_lock(&sfw_data.fw_lock);
 857
 858         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 859             sn == sfw_data.fw_session) {                  /* sn also active */
 860                 spin_unlock(&sfw_data.fw_lock);
 861                 return;
 862         }
 863
 864         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 865
 866         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 867                 if (sfw_batch_active(tsb)) {
 868                         spin_unlock(&sfw_data.fw_lock);
 869                         return;
 870                 }
 871         }
 872
 873         list_del_init(&sn->sn_list);
 874         spin_unlock(&sfw_data.fw_lock);
 875
 876         sfw_destroy_session(sn);
 877         return;
 878 }
 879
 880 static void
 881 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 882 {
 883         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 884         sfw_test_instance_t *tsi = tsu->tsu_instance;
 885         int                  done = 0;
 886
 887         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 888
 889         spin_lock(&tsi->tsi_lock);
 890
 891         LASSERT(sfw_test_active(tsi));
 892         LASSERT(!list_empty(&rpc->crpc_list));
 893
 894         list_del_init(&rpc->crpc_list);
 895
 896         /* batch is stopping or loop is done or get error */
 897         if (tsi->tsi_stopping ||
 898             tsu->tsu_loop == 0 ||
 899             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 900                 done = 1;
 901
 902         /* dec ref for poster */
 903         srpc_client_rpc_decref(rpc);
 904
 905         spin_unlock(&tsi->tsi_lock);
 906
 907         if (!done) {
 908                 swi_schedule_workitem(&tsu->tsu_worker);
 909                 return;
 910         }
 911
 912         sfw_test_unit_done(tsu);
 913         return;
 914 }
 915
 916 int
 917 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
 918                     unsigned features, int nblk, int blklen,
 919                     srpc_client_rpc_t **rpcpp)
 920 {
 921         srpc_client_rpc_t   *rpc = NULL;
 922         sfw_test_instance_t *tsi = tsu->tsu_instance;
 923
 924         spin_lock(&tsi->tsi_lock);
 925
 926         LASSERT (sfw_test_active(tsi));
 927
 928         if (!list_empty(&tsi->tsi_free_rpcs)) {
 929                 /* pick request from buffer */
 930                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 931                                  srpc_client_rpc_t, crpc_list);
 932                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 933                 list_del_init(&rpc->crpc_list);
 934         }
 935
 936         spin_unlock(&tsi->tsi_lock);
 937
 938         if (rpc == NULL) {
 939                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 940                                              blklen, sfw_test_rpc_done,
 941                                              sfw_test_rpc_fini, tsu);
 942         } else {
 943                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 944                                      blklen, sfw_test_rpc_done,
 945                                      sfw_test_rpc_fini, tsu);
 946         }
 947
 948         if (rpc == NULL) {
 949                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 950                 return -ENOMEM;
 951         }
 952
 953         rpc->crpc_reqstmsg.msg_ses_feats = features;
 954         *rpcpp = rpc;
 955
 956         return 0;
 957 }
 958
 959 static int
 960 sfw_run_test (swi_workitem_t *wi)
 961 {
 962         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 963         sfw_test_instance_t *tsi = tsu->tsu_instance;
 964         srpc_client_rpc_t   *rpc = NULL;
 965
 966         LASSERT (wi == &tsu->tsu_worker);
 967
 968         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 969                 LASSERT (rpc == NULL);
 970                 goto test_done;
 971         }
 972
 973         LASSERT (rpc != NULL);
 974
 975         spin_lock(&tsi->tsi_lock);
 976
 977         if (tsi->tsi_stopping) {
 978                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 979                 spin_unlock(&tsi->tsi_lock);
 980                 goto test_done;
 981         }
 982
 983         if (tsu->tsu_loop > 0)
 984                 tsu->tsu_loop--;
 985
 986         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 987         spin_unlock(&tsi->tsi_lock);
 988
 989         spin_lock(&rpc->crpc_lock);
 990         rpc->crpc_timeout = rpc_timeout;
 991         srpc_post_rpc(rpc);
 992         spin_unlock(&rpc->crpc_lock);
 993         return 0;
 994
 995 test_done:
 996         /*
 997          * No one can schedule me now since:
 998          * - previous RPC, if any, has done and
 999          * - no new RPC is initiated.
1000          * - my batch is still active; no one can run it again now.
1001          * Cancel pending schedules and prevent future schedule attempts:
1002          */
1003         swi_exit_workitem(wi);
1004         sfw_test_unit_done(tsu);
1005         return 1;
1006 }
1007
1008 static int
1009 sfw_run_batch (sfw_batch_t *tsb)
1010 {
1011         swi_workitem_t      *wi;
1012         sfw_test_unit_t     *tsu;
1013         sfw_test_instance_t *tsi;
1014
1015         if (sfw_batch_active(tsb)) {
1016                 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1017                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1018                 return 0;
1019         }
1020
1021         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1022                 if (!tsi->tsi_is_client)        /* skip server instances */
1023                         continue;
1024
1025                 LASSERT(!tsi->tsi_stopping);
1026                 LASSERT(!sfw_test_active(tsi));
1027
1028                 atomic_inc(&tsb->bat_nactive);
1029
1030                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1031                         atomic_inc(&tsi->tsi_nactive);
1032                         tsu->tsu_loop = tsi->tsi_loop;
1033                         wi = &tsu->tsu_worker;
1034                         swi_init_workitem(wi, tsu, sfw_run_test,
1035                                           lst_sched_test[\
1036                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1037                         swi_schedule_workitem(wi);
1038                 }
1039         }
1040
1041         return 0;
1042 }
1043
1044 int
1045 sfw_stop_batch (sfw_batch_t *tsb, int force)
1046 {
1047         sfw_test_instance_t *tsi;
1048         srpc_client_rpc_t   *rpc;
1049
1050         if (!sfw_batch_active(tsb)) {
1051                 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1052                 return 0;
1053         }
1054
1055         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1056                 spin_lock(&tsi->tsi_lock);
1057
1058                 if (!tsi->tsi_is_client ||
1059                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1060                         spin_unlock(&tsi->tsi_lock);
1061                         continue;
1062                 }
1063
1064                 tsi->tsi_stopping = 1;
1065
1066                 if (!force) {
1067                         spin_unlock(&tsi->tsi_lock);
1068                         continue;
1069                 }
1070
1071                 /* abort launched rpcs in the test */
1072                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1073                         spin_lock(&rpc->crpc_lock);
1074
1075                         srpc_abort_rpc(rpc, -EINTR);
1076
1077                         spin_unlock(&rpc->crpc_lock);
1078                 }
1079
1080                 spin_unlock(&tsi->tsi_lock);
1081         }
1082
1083         return 0;
1084 }
1085
1086 static int
1087 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1088 {
1089         sfw_test_instance_t *tsi;
1090
1091         if (testidx < 0)
1092                 return -EINVAL;
1093
1094         if (testidx == 0) {
1095                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1096                 return 0;
1097         }
1098
1099         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1100                 if (testidx-- > 1)
1101                         continue;
1102
1103                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1104                 return 0;
1105         }
1106
1107         return -ENOENT;
1108 }
1109
1110 void
1111 sfw_free_pages (srpc_server_rpc_t *rpc)
1112 {
1113         srpc_free_bulk(rpc->srpc_bulk);
1114         rpc->srpc_bulk = NULL;
1115 }
1116
1117 int
1118 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1119                 int sink)
1120 {
1121         LASSERT(rpc->srpc_bulk == NULL);
1122         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1123
1124         rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1125         if (rpc->srpc_bulk == NULL)
1126                 return -ENOMEM;
1127
1128         return 0;
1129 }
1130
1131 static int
1132 sfw_add_test (srpc_server_rpc_t *rpc)
1133 {
1134         sfw_session_t     *sn = sfw_data.fw_session;
1135         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1136         srpc_test_reqst_t *request;
1137         int                rc;
1138         sfw_batch_t       *bat;
1139
1140         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1141         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1142
1143         if (request->tsr_loop == 0 ||
1144             request->tsr_concur == 0 ||
1145             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1146             request->tsr_ndest > SFW_MAX_NDESTS ||
1147             (request->tsr_is_client && request->tsr_ndest == 0) ||
1148             request->tsr_concur > SFW_MAX_CONCUR ||
1149             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1150             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1151                 reply->tsr_status = EINVAL;
1152                 return 0;
1153         }
1154
1155         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1156             sfw_find_test_case(request->tsr_service) == NULL) {
1157                 reply->tsr_status = ENOENT;
1158                 return 0;
1159         }
1160
1161         bat = sfw_bid2batch(request->tsr_bid);
1162         if (bat == NULL) {
1163                 CERROR("dropping RPC %s from %s under memory pressure\n",
1164                         rpc->srpc_scd->scd_svc->sv_name,
1165                         libcfs_id2str(rpc->srpc_peer));
1166                 return -ENOMEM;
1167         }
1168
1169         if (sfw_batch_active(bat)) {
1170                 reply->tsr_status = EBUSY;
1171                 return 0;
1172         }
1173
1174         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1175                 /* rpc will be resumed later in sfw_bulk_ready */
1176                 int     npg = sfw_id_pages(request->tsr_ndest);
1177                 int     len;
1178
1179                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1180                         len = npg * PAGE_SIZE;
1181
1182                 } else  {
1183                         len = sizeof(lnet_process_id_packed_t) *
1184                               request->tsr_ndest;
1185                 }
1186
1187                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1188         }
1189
1190         rc = sfw_add_test_instance(bat, rpc);
1191         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1192                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1193                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1194                 request->tsr_is_client ? "client" : "server",
1195                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1196
1197         reply->tsr_status = (rc < 0) ? -rc : rc;
1198         return 0;
1199 }
1200
1201 static int
1202 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1203 {
1204         sfw_session_t *sn = sfw_data.fw_session;
1205         int            rc = 0;
1206         sfw_batch_t   *bat;
1207
1208         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1209
1210         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1211                 reply->bar_status = ESRCH;
1212                 return 0;
1213         }
1214
1215         bat = sfw_find_batch(request->bar_bid);
1216         if (bat == NULL) {
1217                 reply->bar_status = ENOENT;
1218                 return 0;
1219         }
1220
1221         switch (request->bar_opc) {
1222         case SRPC_BATCH_OPC_RUN:
1223                 rc = sfw_run_batch(bat);
1224                 break;
1225
1226         case SRPC_BATCH_OPC_STOP:
1227                 rc = sfw_stop_batch(bat, request->bar_arg);
1228                 break;
1229
1230         case SRPC_BATCH_OPC_QUERY:
1231                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1232                 break;
1233
1234         default:
1235                 return -EINVAL; /* drop it */
1236         }
1237
1238         reply->bar_status = (rc < 0) ? -rc : rc;
1239         return 0;
1240 }
1241
1242 static int
1243 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1244 {
1245         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1246         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1247         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1248         unsigned        features = LST_FEATS_MASK;
1249         int             rc = 0;
1250
1251         LASSERT(sfw_data.fw_active_srpc == NULL);
1252         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1253
1254         spin_lock(&sfw_data.fw_lock);
1255
1256         if (sfw_data.fw_shuttingdown) {
1257                 spin_unlock(&sfw_data.fw_lock);
1258                 return -ESHUTDOWN;
1259         }
1260
1261         /* Remove timer to avoid racing with it or expiring active session */
1262         if (sfw_del_session_timer() != 0) {
1263                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1264                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1265                 spin_unlock(&sfw_data.fw_lock);
1266                 return -EAGAIN;
1267         }
1268
1269         sfw_data.fw_active_srpc = rpc;
1270         spin_unlock(&sfw_data.fw_lock);
1271
1272         sfw_unpack_message(request);
1273         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1274
1275         /* rpc module should have checked this */
1276         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1277
1278         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1279             sv->sv_id != SRPC_SERVICE_DEBUG) {
1280                 sfw_session_t *sn = sfw_data.fw_session;
1281
1282                 if (sn != NULL &&
1283                     sn->sn_features != request->msg_ses_feats) {
1284                         CNETERR("Features of framework RPC don't match "
1285                                 "features of current session: %x/%x\n",
1286                                 request->msg_ses_feats, sn->sn_features);
1287                         reply->msg_body.reply.status = EPROTO;
1288                         reply->msg_body.reply.sid    = sn->sn_id;
1289                         goto out;
1290                 }
1291
1292         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1293                 /* NB: at this point, old version will ignore features and
1294                  * create new session anyway, so console should be able
1295                  * to handle this */
1296                 reply->msg_body.reply.status = EPROTO;
1297                 goto out;
1298         }
1299
1300         switch(sv->sv_id) {
1301         default:
1302                 LBUG ();
1303         case SRPC_SERVICE_TEST:
1304                 rc = sfw_add_test(rpc);
1305                 break;
1306
1307         case SRPC_SERVICE_BATCH:
1308                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1309                                        &reply->msg_body.bat_reply);
1310                 break;
1311
1312         case SRPC_SERVICE_QUERY_STAT:
1313                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1314                                    &reply->msg_body.stat_reply);
1315                 break;
1316
1317         case SRPC_SERVICE_DEBUG:
1318                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1319                                        &reply->msg_body.dbg_reply);
1320                 break;
1321
1322         case SRPC_SERVICE_MAKE_SESSION:
1323                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1324                                       &reply->msg_body.mksn_reply);
1325                 break;
1326
1327         case SRPC_SERVICE_REMOVE_SESSION:
1328                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1329                                         &reply->msg_body.rmsn_reply);
1330                 break;
1331         }
1332
1333         if (sfw_data.fw_session != NULL)
1334                 features = sfw_data.fw_session->sn_features;
1335  out:
1336         reply->msg_ses_feats = features;
1337         rpc->srpc_done = sfw_server_rpc_done;
1338         spin_lock(&sfw_data.fw_lock);
1339
1340         if (!sfw_data.fw_shuttingdown)
1341                 sfw_add_session_timer();
1342
1343         sfw_data.fw_active_srpc = NULL;
1344         spin_unlock(&sfw_data.fw_lock);
1345         return rc;
1346 }
1347
1348 static int
1349 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1350 {
1351         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1352         int                     rc;
1353
1354         LASSERT(rpc->srpc_bulk != NULL);
1355         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1356         LASSERT(sfw_data.fw_active_srpc == NULL);
1357         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1358
1359         spin_lock(&sfw_data.fw_lock);
1360
1361         if (status != 0) {
1362                 CERROR("Bulk transfer failed for RPC: "
1363                        "service %s, peer %s, status %d\n",
1364                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1365                 spin_unlock(&sfw_data.fw_lock);
1366                 return -EIO;
1367         }
1368
1369         if (sfw_data.fw_shuttingdown) {
1370                 spin_unlock(&sfw_data.fw_lock);
1371                 return -ESHUTDOWN;
1372         }
1373
1374         if (sfw_del_session_timer() != 0) {
1375                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1376                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1377                 spin_unlock(&sfw_data.fw_lock);
1378                 return -EAGAIN;
1379         }
1380
1381         sfw_data.fw_active_srpc = rpc;
1382         spin_unlock(&sfw_data.fw_lock);
1383
1384         rc = sfw_add_test(rpc);
1385
1386         spin_lock(&sfw_data.fw_lock);
1387
1388         if (!sfw_data.fw_shuttingdown)
1389                 sfw_add_session_timer();
1390
1391         sfw_data.fw_active_srpc = NULL;
1392         spin_unlock(&sfw_data.fw_lock);
1393         return rc;
1394 }
1395
1396 srpc_client_rpc_t *
1397 sfw_create_rpc(lnet_process_id_t peer, int service,
1398                unsigned features, int nbulkiov, int bulklen,
1399                void (*done)(srpc_client_rpc_t *), void *priv)
1400 {
1401         srpc_client_rpc_t *rpc = NULL;
1402
1403         spin_lock(&sfw_data.fw_lock);
1404
1405         LASSERT (!sfw_data.fw_shuttingdown);
1406         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1407
1408         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1409                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1410                                      srpc_client_rpc_t, crpc_list);
1411                 list_del(&rpc->crpc_list);
1412
1413                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1414                                      done, sfw_client_rpc_fini, priv);
1415         }
1416
1417         spin_unlock(&sfw_data.fw_lock);
1418
1419         if (rpc == NULL) {
1420                 rpc = srpc_create_client_rpc(peer, service,
1421                                              nbulkiov, bulklen, done,
1422                                              nbulkiov != 0 ?  NULL :
1423                                              sfw_client_rpc_fini,
1424                                              priv);
1425         }
1426
1427         if (rpc != NULL) /* "session" is concept in framework */
1428                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1429
1430         return rpc;
1431 }
1432
1433 void
1434 sfw_unpack_message (srpc_msg_t *msg)
1435 {
1436         if (msg->msg_magic == SRPC_MSG_MAGIC)
1437                 return; /* no flipping needed */
1438
1439         /* srpc module should guarantee I wouldn't get crap */
1440         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1441
1442         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1443                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1444
1445                 __swab32s(&req->str_type);
1446                 __swab64s(&req->str_rpyid);
1447                 sfw_unpack_sid(req->str_sid);
1448                 return;
1449         }
1450
1451         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1452                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1453
1454                 __swab32s(&rep->str_status);
1455                 sfw_unpack_sid(rep->str_sid);
1456                 sfw_unpack_fw_counters(rep->str_fw);
1457                 sfw_unpack_rpc_counters(rep->str_rpc);
1458                 sfw_unpack_lnet_counters(rep->str_lnet);
1459                 return;
1460         }
1461
1462         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1463                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1464
1465                 __swab64s(&req->mksn_rpyid);
1466                 __swab32s(&req->mksn_force);
1467                 sfw_unpack_sid(req->mksn_sid);
1468                 return;
1469         }
1470
1471         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1472                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1473
1474                 __swab32s(&rep->mksn_status);
1475                 __swab32s(&rep->mksn_timeout);
1476                 sfw_unpack_sid(rep->mksn_sid);
1477                 return;
1478         }
1479
1480         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1481                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1482
1483                 __swab64s(&req->rmsn_rpyid);
1484                 sfw_unpack_sid(req->rmsn_sid);
1485                 return;
1486         }
1487
1488         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1489                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1490
1491                 __swab32s(&rep->rmsn_status);
1492                 sfw_unpack_sid(rep->rmsn_sid);
1493                 return;
1494         }
1495
1496         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1497                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1498
1499                 __swab64s(&req->dbg_rpyid);
1500                 __swab32s(&req->dbg_flags);
1501                 sfw_unpack_sid(req->dbg_sid);
1502                 return;
1503         }
1504
1505         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1506                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1507
1508                 __swab32s(&rep->dbg_nbatch);
1509                 __swab32s(&rep->dbg_timeout);
1510                 sfw_unpack_sid(rep->dbg_sid);
1511                 return;
1512         }
1513
1514         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1515                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1516
1517                 __swab32s(&req->bar_opc);
1518                 __swab64s(&req->bar_rpyid);
1519                 __swab32s(&req->bar_testidx);
1520                 __swab32s(&req->bar_arg);
1521                 sfw_unpack_sid(req->bar_sid);
1522                 __swab64s(&req->bar_bid.bat_id);
1523                 return;
1524         }
1525
1526         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1527                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1528
1529                 __swab32s(&rep->bar_status);
1530                 sfw_unpack_sid(rep->bar_sid);
1531                 return;
1532         }
1533
1534         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1535                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1536
1537                 __swab64s(&req->tsr_rpyid);
1538                 __swab64s(&req->tsr_bulkid);
1539                 __swab32s(&req->tsr_loop);
1540                 __swab32s(&req->tsr_ndest);
1541                 __swab32s(&req->tsr_concur);
1542                 __swab32s(&req->tsr_service);
1543                 sfw_unpack_sid(req->tsr_sid);
1544                 __swab64s(&req->tsr_bid.bat_id);
1545                 return;
1546         }
1547
1548         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1549                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1550
1551                 __swab32s(&rep->tsr_status);
1552                 sfw_unpack_sid(rep->tsr_sid);
1553                 return;
1554         }
1555
1556         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1557                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1558
1559                 __swab64s(&req->join_rpyid);
1560                 sfw_unpack_sid(req->join_sid);
1561                 return;
1562         }
1563
1564         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1565                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1566
1567                 __swab32s(&rep->join_status);
1568                 __swab32s(&rep->join_timeout);
1569                 sfw_unpack_sid(rep->join_sid);
1570                 return;
1571         }
1572
1573         LBUG ();
1574         return;
1575 }
1576
1577 void
1578 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1579 {
1580         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1581         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1582
1583         spin_lock(&rpc->crpc_lock);
1584         srpc_abort_rpc(rpc, -EINTR);
1585         spin_unlock(&rpc->crpc_lock);
1586         return;
1587 }
1588
1589 void
1590 sfw_post_rpc (srpc_client_rpc_t *rpc)
1591 {
1592         spin_lock(&rpc->crpc_lock);
1593
1594         LASSERT(!rpc->crpc_closed);
1595         LASSERT(!rpc->crpc_aborted);
1596         LASSERT(list_empty(&rpc->crpc_list));
1597         LASSERT(!sfw_data.fw_shuttingdown);
1598
1599         rpc->crpc_timeout = rpc_timeout;
1600         srpc_post_rpc(rpc);
1601
1602         spin_unlock(&rpc->crpc_lock);
1603         return;
1604 }
1605
1606 static srpc_service_t sfw_services[] =
1607 {
1608         {
1609                 /* sv_id */    SRPC_SERVICE_DEBUG,
1610                 /* sv_name */  "debug",
1611                 0
1612         },
1613         {
1614                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1615                 /* sv_name */  "query stats",
1616                 0
1617         },
1618         {
1619                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1620                 /* sv_name */  "make session",
1621                 0
1622         },
1623         {
1624                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1625                 /* sv_name */  "remove session",
1626                 0
1627         },
1628         {
1629                 /* sv_id */    SRPC_SERVICE_BATCH,
1630                 /* sv_name */  "batch service",
1631                 0
1632         },
1633         {
1634                 /* sv_id */    SRPC_SERVICE_TEST,
1635                 /* sv_name */  "test service",
1636                 0
1637         },
1638         {
1639                 /* sv_id */    0,
1640                 /* sv_name */  NULL,
1641                 0
1642         }
1643 };
1644
1645 int
1646 sfw_startup (void)
1647 {
1648         int              i;
1649         int              rc;
1650         int              error;
1651         srpc_service_t  *sv;
1652         sfw_test_case_t *tsc;
1653
1654
1655         if (session_timeout < 0) {
1656                 CERROR ("Session timeout must be non-negative: %d\n",
1657                         session_timeout);
1658                 return -EINVAL;
1659         }
1660
1661         if (rpc_timeout < 0) {
1662                 CERROR ("RPC timeout must be non-negative: %d\n",
1663                         rpc_timeout);
1664                 return -EINVAL;
1665         }
1666
1667         if (session_timeout == 0)
1668                 CWARN ("Zero session_timeout specified "
1669                        "- test sessions never expire.\n");
1670
1671         if (rpc_timeout == 0)
1672                 CWARN ("Zero rpc_timeout specified "
1673                        "- test RPC never expire.\n");
1674
1675         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1676
1677         sfw_data.fw_session     = NULL;
1678         sfw_data.fw_active_srpc = NULL;
1679         spin_lock_init(&sfw_data.fw_lock);
1680         atomic_set(&sfw_data.fw_nzombies, 0);
1681         INIT_LIST_HEAD(&sfw_data.fw_tests);
1682         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1683         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1684
1685         brw_init_test_client();
1686         brw_init_test_service();
1687         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1688         LASSERT (rc == 0);
1689
1690         ping_init_test_client();
1691         ping_init_test_service();
1692         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1693         LASSERT (rc == 0);
1694
1695         error = 0;
1696         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1697                 sv = tsc->tsc_srv_service;
1698
1699                 rc = srpc_add_service(sv);
1700                 LASSERT(rc != -EBUSY);
1701                 if (rc != 0) {
1702                         CWARN("Failed to add %s service: %d\n",
1703                               sv->sv_name, rc);
1704                         error = rc;
1705                 }
1706         }
1707
1708         for (i = 0; ; i++) {
1709                 sv = &sfw_services[i];
1710                 if (sv->sv_name == NULL) break;
1711
1712                 sv->sv_bulk_ready = NULL;
1713                 sv->sv_handler    = sfw_handle_server_rpc;
1714                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1715                 if (sv->sv_id == SRPC_SERVICE_TEST)
1716                         sv->sv_bulk_ready = sfw_bulk_ready;
1717
1718                 rc = srpc_add_service(sv);
1719                 LASSERT (rc != -EBUSY);
1720                 if (rc != 0) {
1721                         CWARN ("Failed to add %s service: %d\n",
1722                                sv->sv_name, rc);
1723                         error = rc;
1724                 }
1725
1726                 /* about to sfw_shutdown, no need to add buffer */
1727                 if (error) continue;
1728
1729                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1730                 if (rc != 0) {
1731                         CWARN("Failed to reserve enough buffers: "
1732                               "service %s, %d needed: %d\n",
1733                               sv->sv_name, sv->sv_wi_total, rc);
1734                         error = -ENOMEM;
1735                 }
1736         }
1737
1738         if (error != 0)
1739                 sfw_shutdown();
1740         return error;
1741 }
1742
1743 void
1744 sfw_shutdown (void)
1745 {
1746         srpc_service_t  *sv;
1747         sfw_test_case_t *tsc;
1748         int              i;
1749
1750         spin_lock(&sfw_data.fw_lock);
1751
1752         sfw_data.fw_shuttingdown = 1;
1753         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1754                        "waiting for active RPC to finish.\n");
1755
1756         if (sfw_del_session_timer() != 0)
1757                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1758                                "waiting for session timer to explode.\n");
1759
1760         sfw_deactivate_session();
1761         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1762                        sfw_data.fw_lock,
1763                        "waiting for %d zombie sessions to die.\n",
1764                        atomic_read(&sfw_data.fw_nzombies));
1765
1766         spin_unlock(&sfw_data.fw_lock);
1767
1768         for (i = 0; ; i++) {
1769                 sv = &sfw_services[i];
1770                 if (sv->sv_name == NULL)
1771                         break;
1772
1773                 srpc_shutdown_service(sv);
1774                 srpc_remove_service(sv);
1775         }
1776
1777         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1778                 sv = tsc->tsc_srv_service;
1779                 srpc_shutdown_service(sv);
1780                 srpc_remove_service(sv);
1781         }
1782
1783         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1784                 srpc_client_rpc_t *rpc;
1785
1786                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1787                                  srpc_client_rpc_t, crpc_list);
1788                 list_del(&rpc->crpc_list);
1789
1790                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1791         }
1792
1793         for (i = 0; ; i++) {
1794                 sv = &sfw_services[i];
1795                 if (sv->sv_name == NULL)
1796                         break;
1797
1798                 srpc_wait_service_shutdown(sv);
1799         }
1800
1801         while (!list_empty(&sfw_data.fw_tests)) {
1802                 tsc = list_entry(sfw_data.fw_tests.next,
1803                                  sfw_test_case_t, tsc_list);
1804
1805                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1806
1807                 list_del(&tsc->tsc_list);
1808                 LIBCFS_FREE(tsc, sizeof(*tsc));
1809         }
1810
1811         return;
1812 }