lnet/selftest/framework.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lnet/selftest/framework.c
  37  *
  38  * Author: Isaac Huang <isaac@clusterfs.com>
  39  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LNET
  43
  44 #include "selftest.h"
  45
  46 int brw_inject_errors = 0;
  47 CFS_MODULE_PARM(brw_inject_errors, "i", int, 0644,
  48                 "# data errors to inject randomly, zero by default");
  49
  50 static int session_timeout = 100;
  51 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  52                 "test session timeout in seconds (100 by default, 0 == never)");
  53
  54 static int rpc_timeout = 64;
  55 CFS_MODULE_PARM(rpc_timeout, "i", int, 0644,
  56                 "rpc timeout in seconds (64 by default, 0 == never)");
  57
  58 #define SFW_TEST_CONCURRENCY     1792
  59 #define SFW_EXTRA_TEST_BUFFERS   8 /* tolerate buggy peers with extra buffers */
  60
  61 #define sfw_test_buffers(tsi)    ((tsi)->tsi_loop + SFW_EXTRA_TEST_BUFFERS)
  62
  63 #define sfw_unpack_id(id)               \
  64 do {                                    \
  65         __swab64s(&(id).nid);           \
  66         __swab32s(&(id).pid);           \
  67 } while (0)
  68
  69 #define sfw_unpack_sid(sid)             \
  70 do {                                    \
  71         __swab64s(&(sid).ses_nid);      \
  72         __swab64s(&(sid).ses_stamp);    \
  73 } while (0)
  74
  75 #define sfw_unpack_fw_counters(fc)        \
  76 do {                                      \
  77         __swab32s(&(fc).brw_errors);      \
  78         __swab32s(&(fc).ping_errors);     \
  79         __swab32s(&(fc).active_tests);    \
  80         __swab32s(&(fc).active_batches);  \
  81         __swab32s(&(fc).zombie_sessions); \
  82 } while (0)
  83
  84 #define sfw_unpack_rpc_counters(rc)     \
  85 do {                                    \
  86         __swab32s(&(rc).errors);        \
  87         __swab32s(&(rc).rpcs_sent);     \
  88         __swab32s(&(rc).rpcs_rcvd);     \
  89         __swab32s(&(rc).rpcs_dropped);  \
  90         __swab32s(&(rc).rpcs_expired);  \
  91         __swab64s(&(rc).bulk_get);      \
  92         __swab64s(&(rc).bulk_put);      \
  93 } while (0)
  94
  95 #define sfw_unpack_lnet_counters(lc)    \
  96 do {                                    \
  97         __swab32s(&(lc).errors);        \
  98         __swab32s(&(lc).msgs_max);      \
  99         __swab32s(&(lc).msgs_alloc);    \
 100         __swab32s(&(lc).send_count);    \
 101         __swab32s(&(lc).recv_count);    \
 102         __swab32s(&(lc).drop_count);    \
 103         __swab32s(&(lc).route_count);   \
 104         __swab64s(&(lc).send_length);   \
 105         __swab64s(&(lc).recv_length);   \
 106         __swab64s(&(lc).drop_length);   \
 107         __swab64s(&(lc).route_length);  \
 108 } while (0)
 109
 110 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 111 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 112
 113 struct smoketest_framework {
 114         struct list_head   fw_zombie_rpcs;     /* RPCs to be recycled */
 115         struct list_head   fw_zombie_sessions; /* stopping sessions */
 116         struct list_head   fw_tests;           /* registered test cases */
 117         atomic_t           fw_nzombies;        /* # zombie sessions */
 118         spinlock_t         fw_lock;            /* serialise */
 119         sfw_session_t     *fw_session;         /* _the_ session */
 120         int                fw_shuttingdown;    /* shutdown in progress */
 121         srpc_server_rpc_t *fw_active_srpc;     /* running RPC */
 122 } sfw_data;
 123
 124 /* forward ref's */
 125 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 126 void sfw_destroy_session (sfw_session_t *sn);
 127
 128 static inline sfw_test_case_t *
 129 sfw_find_test_case(int id)
 130 {
 131         sfw_test_case_t *tsc;
 132
 133         LASSERT (id <= SRPC_SERVICE_MAX_ID);
 134         LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 135
 136         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
 137                 if (tsc->tsc_srv_service->sv_id == id)
 138                         return tsc;
 139         }
 140
 141         return NULL;
 142 }
 143
 144 static int
 145 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 146 {
 147         sfw_test_case_t *tsc;
 148
 149         if (sfw_find_test_case(service->sv_id) != NULL) {
 150                 CERROR ("Failed to register test %s (%d)\n",
 151                         service->sv_name, service->sv_id);
 152                 return -EEXIST;
 153         }
 154
 155         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 156         if (tsc == NULL)
 157                 return -ENOMEM;
 158
 159         memset(tsc, 0, sizeof(sfw_test_case_t));
 160         tsc->tsc_cli_ops     = cliops;
 161         tsc->tsc_srv_service = service;
 162
 163         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 164         return 0;
 165 }
 166
 167 void
 168 sfw_add_session_timer (void)
 169 {
 170         sfw_session_t *sn = sfw_data.fw_session;
 171         stt_timer_t   *timer = &sn->sn_timer;
 172
 173         LASSERT (!sfw_data.fw_shuttingdown);
 174
 175         if (sn == NULL || sn->sn_timeout == 0)
 176                 return;
 177
 178         LASSERT (!sn->sn_timer_active);
 179
 180         sn->sn_timer_active = 1;
 181         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 182                                           cfs_time_current_sec());
 183         stt_add_timer(timer);
 184         return;
 185 }
 186
 187 int
 188 sfw_del_session_timer (void)
 189 {
 190         sfw_session_t *sn = sfw_data.fw_session;
 191
 192         if (sn == NULL || !sn->sn_timer_active)
 193                 return 0;
 194
 195         LASSERT (sn->sn_timeout != 0);
 196
 197         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 198                 sn->sn_timer_active = 0;
 199                 return 0;
 200         }
 201
 202 #ifndef __KERNEL__
 203         /* Racing is impossible in single-threaded userland selftest */
 204         LBUG();
 205 #endif
 206         return EBUSY; /* racing with sfw_session_expired() */
 207 }
 208
 209 /* called with sfw_data.fw_lock held */
 210 static void
 211 sfw_deactivate_session (void)
 212 {
 213         sfw_session_t *sn = sfw_data.fw_session;
 214         int            nactive = 0;
 215         sfw_batch_t   *tsb;
 216         sfw_test_case_t *tsc;
 217
 218         if (sn == NULL) return;
 219
 220         LASSERT (!sn->sn_timer_active);
 221
 222         sfw_data.fw_session = NULL;
 223         atomic_inc(&sfw_data.fw_nzombies);
 224         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 225
 226         spin_unlock(&sfw_data.fw_lock);
 227
 228         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
 229                 srpc_abort_service(tsc->tsc_srv_service);
 230         }
 231
 232         spin_lock(&sfw_data.fw_lock);
 233
 234         list_for_each_entry (tsb, &sn->sn_batches, bat_list) {
 235                 if (sfw_batch_active(tsb)) {
 236                         nactive++;
 237                         sfw_stop_batch(tsb, 1);
 238                 }
 239         }
 240
 241         if (nactive != 0)
 242                 return;   /* wait for active batches to stop */
 243
 244         list_del_init(&sn->sn_list);
 245         spin_unlock(&sfw_data.fw_lock);
 246
 247         sfw_destroy_session(sn);
 248
 249         spin_lock(&sfw_data.fw_lock);
 250         return;
 251 }
 252
 253 #ifndef __KERNEL__
 254
 255 int
 256 sfw_session_removed (void)
 257 {
 258         return (sfw_data.fw_session == NULL) ? 1 : 0;
 259 }
 260
 261 #endif
 262
 263 void
 264 sfw_session_expired (void *data)
 265 {
 266         sfw_session_t *sn = data;
 267
 268         spin_lock(&sfw_data.fw_lock);
 269
 270         LASSERT (sn->sn_timer_active);
 271         LASSERT (sn == sfw_data.fw_session);
 272
 273         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 274                libcfs_nid2str(sn->sn_id.ses_nid),
 275                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 276
 277         sn->sn_timer_active = 0;
 278         sfw_deactivate_session();
 279
 280         spin_unlock(&sfw_data.fw_lock);
 281         return;
 282 }
 283
 284 static inline void
 285 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
 286 {
 287         stt_timer_t *timer = &sn->sn_timer;
 288
 289         memset(sn, 0, sizeof(sfw_session_t));
 290         CFS_INIT_LIST_HEAD(&sn->sn_list);
 291         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 292         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 293         atomic_set(&sn->sn_brw_errors, 0);
 294         atomic_set(&sn->sn_ping_errors, 0);
 295         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 296
 297         sn->sn_timer_active = 0;
 298         sn->sn_id           = sid;
 299         sn->sn_timeout      = session_timeout;
 300
 301         timer->stt_data = sn;
 302         timer->stt_func = sfw_session_expired;
 303         CFS_INIT_LIST_HEAD(&timer->stt_list);
 304 }
 305
 306 /* completion handler for incoming framework RPCs */
 307 void
 308 sfw_server_rpc_done (srpc_server_rpc_t *rpc)
 309 {
 310         srpc_service_t *sv = rpc->srpc_service;
 311         int             status = rpc->srpc_status;
 312
 313         CDEBUG (D_NET,
 314                 "Incoming framework RPC done: "
 315                 "service %s, peer %s, status %s:%d\n",
 316                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 317                 swi_state2str(rpc->srpc_wi.wi_state),
 318                 status);
 319
 320         if (rpc->srpc_bulk != NULL)
 321                 sfw_free_pages(rpc);
 322         return;
 323 }
 324
 325 void
 326 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 327 {
 328         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 329         LASSERT (list_empty(&rpc->crpc_list));
 330         LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
 331 #ifndef __KERNEL__
 332         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 333 #endif
 334
 335         CDEBUG (D_NET,
 336                 "Outgoing framework RPC done: "
 337                 "service %d, peer %s, status %s:%d:%d\n",
 338                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 339                 swi_state2str(rpc->crpc_wi.wi_state),
 340                 rpc->crpc_aborted, rpc->crpc_status);
 341
 342         spin_lock(&sfw_data.fw_lock);
 343
 344         /* my callers must finish all RPCs before shutting me down */
 345         LASSERT (!sfw_data.fw_shuttingdown);
 346         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 347
 348         spin_unlock(&sfw_data.fw_lock);
 349         return;
 350 }
 351
 352 sfw_batch_t *
 353 sfw_find_batch (lst_bid_t bid)
 354 {
 355         sfw_session_t *sn = sfw_data.fw_session;
 356         sfw_batch_t   *bat;
 357
 358         LASSERT (sn != NULL);
 359
 360         list_for_each_entry (bat, &sn->sn_batches, bat_list) {
 361                 if (bat->bat_id.bat_id == bid.bat_id)
 362                         return bat;
 363         }
 364
 365         return NULL;
 366 }
 367
 368 sfw_batch_t *
 369 sfw_bid2batch (lst_bid_t bid)
 370 {
 371         sfw_session_t *sn = sfw_data.fw_session;
 372         sfw_batch_t   *bat;
 373
 374         LASSERT (sn != NULL);
 375
 376         bat = sfw_find_batch(bid);
 377         if (bat != NULL)
 378                 return bat;
 379
 380         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 381         if (bat == NULL)
 382                 return NULL;
 383
 384         bat->bat_error    = 0;
 385         bat->bat_session  = sn;
 386         bat->bat_id       = bid;
 387         atomic_set(&bat->bat_nactive, 0);
 388         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 389
 390         list_add_tail(&bat->bat_list, &sn->sn_batches);
 391         return bat;
 392 }
 393
 394 int
 395 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 396 {
 397         sfw_session_t  *sn = sfw_data.fw_session;
 398         sfw_counters_t *cnt = &reply->str_fw;
 399         sfw_batch_t    *bat;
 400
 401         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 402
 403         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 404                 reply->str_status = EINVAL;
 405                 return 0;
 406         }
 407
 408         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 409                 reply->str_status = ESRCH;
 410                 return 0;
 411         }
 412
 413         LNET_LOCK();
 414         reply->str_lnet = the_lnet.ln_counters;
 415         LNET_UNLOCK();
 416
 417         srpc_get_counters(&reply->str_rpc);
 418
 419         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 420         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 421         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 422
 423         cnt->active_tests = cnt->active_batches = 0;
 424         list_for_each_entry (bat, &sn->sn_batches, bat_list) {
 425                 int n = atomic_read(&bat->bat_nactive);
 426
 427                 if (n > 0) {
 428                         cnt->active_batches++;
 429                         cnt->active_tests += n;
 430                 }
 431         }
 432
 433         reply->str_status = 0;
 434         return 0;
 435 }
 436
 437 int
 438 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 439 {
 440         sfw_session_t *sn = sfw_data.fw_session;
 441
 442         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 443                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 444                 reply->mksn_status = EINVAL;
 445                 return 0;
 446         }
 447
 448         if (sn != NULL) {
 449                 reply->mksn_status  = 0;
 450                 reply->mksn_sid     = sn->sn_id;
 451                 reply->mksn_timeout = sn->sn_timeout;
 452
 453                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 454                         atomic_inc(&sn->sn_refcount);
 455                         return 0;
 456                 }
 457
 458                 if (!request->mksn_force) {
 459                         reply->mksn_status = EBUSY;
 460                         strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 461                         return 0;
 462                 }
 463         }
 464
 465         /* brand new or create by force */
 466         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 467         if (sn == NULL) {
 468                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 469                 return -ENOMEM;
 470         }
 471
 472         sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
 473
 474         spin_lock(&sfw_data.fw_lock);
 475
 476         sfw_deactivate_session();
 477         LASSERT (sfw_data.fw_session == NULL);
 478         sfw_data.fw_session = sn;
 479
 480         spin_unlock(&sfw_data.fw_lock);
 481
 482         reply->mksn_status  = 0;
 483         reply->mksn_sid     = sn->sn_id;
 484         reply->mksn_timeout = sn->sn_timeout;
 485         return 0;
 486 }
 487
 488 int
 489 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 490 {
 491         sfw_session_t *sn = sfw_data.fw_session;
 492
 493         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 494
 495         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 496                 reply->rmsn_status = EINVAL;
 497                 return 0;
 498         }
 499
 500         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 501                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 502                 return 0;
 503         }
 504
 505         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 506                 reply->rmsn_status = 0;
 507                 return 0;
 508         }
 509
 510         spin_lock(&sfw_data.fw_lock);
 511         sfw_deactivate_session();
 512         spin_unlock(&sfw_data.fw_lock);
 513
 514         reply->rmsn_status = 0;
 515         reply->rmsn_sid    = LST_INVALID_SID;
 516         LASSERT (sfw_data.fw_session == NULL);
 517         return 0;
 518 }
 519
 520 int
 521 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 522 {
 523         sfw_session_t *sn = sfw_data.fw_session;
 524
 525         if (sn == NULL) {
 526                 reply->dbg_status = ESRCH;
 527                 reply->dbg_sid    = LST_INVALID_SID;
 528                 return 0;
 529         }
 530
 531         reply->dbg_status  = 0;
 532         reply->dbg_sid     = sn->sn_id;
 533         reply->dbg_timeout = sn->sn_timeout;
 534         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 535
 536         return 0;
 537 }
 538
 539 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 int
 551 sfw_load_test (sfw_test_instance_t *tsi)
 552 {
 553         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 554         int              nrequired = sfw_test_buffers(tsi);
 555         int              nposted;
 556
 557         LASSERT (tsc != NULL);
 558
 559         if (tsi->tsi_is_client) {
 560                 tsi->tsi_ops = tsc->tsc_cli_ops;
 561                 return 0;
 562         }
 563
 564         nposted = srpc_service_add_buffers(tsc->tsc_srv_service, nrequired);
 565         if (nposted != nrequired) {
 566                 CWARN ("Failed to reserve enough buffers: "
 567                        "service %s, %d needed, %d reserved\n",
 568                        tsc->tsc_srv_service->sv_name, nrequired, nposted);
 569                 srpc_service_remove_buffers(tsc->tsc_srv_service, nposted);
 570                 return -ENOMEM;
 571         }
 572
 573         CDEBUG (D_NET, "Reserved %d buffers for test %s\n",
 574                 nposted, tsc->tsc_srv_service->sv_name);
 575         return 0;
 576 }
 577
 578 void
 579 sfw_unload_test (sfw_test_instance_t *tsi)
 580 {
 581         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 582
 583         LASSERT (tsc != NULL);
 584
 585         if (!tsi->tsi_is_client)
 586                 srpc_service_remove_buffers(tsc->tsc_srv_service,
 587                                             sfw_test_buffers(tsi));
 588         return;
 589 }
 590
 591 void
 592 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 593 {
 594         srpc_client_rpc_t *rpc;
 595         sfw_test_unit_t   *tsu;
 596
 597         if (!tsi->tsi_is_client) goto clean;
 598
 599         tsi->tsi_ops->tso_fini(tsi);
 600
 601         LASSERT (!tsi->tsi_stopping);
 602         LASSERT (list_empty(&tsi->tsi_active_rpcs));
 603         LASSERT (!sfw_test_active(tsi));
 604
 605         while (!list_empty(&tsi->tsi_units)) {
 606                 tsu = list_entry(tsi->tsi_units.next,
 607                                  sfw_test_unit_t, tsu_list);
 608                 list_del(&tsu->tsu_list);
 609                 LIBCFS_FREE(tsu, sizeof(*tsu));
 610         }
 611
 612         while (!list_empty(&tsi->tsi_free_rpcs)) {
 613                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 614                                  srpc_client_rpc_t, crpc_list);
 615                 list_del(&rpc->crpc_list);
 616                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 617         }
 618
 619 clean:
 620         sfw_unload_test(tsi);
 621         LIBCFS_FREE(tsi, sizeof(*tsi));
 622         return;
 623 }
 624
 625 void
 626 sfw_destroy_batch (sfw_batch_t *tsb)
 627 {
 628         sfw_test_instance_t *tsi;
 629
 630         LASSERT (!sfw_batch_active(tsb));
 631         LASSERT (list_empty(&tsb->bat_list));
 632
 633         while (!list_empty(&tsb->bat_tests)) {
 634                 tsi = list_entry(tsb->bat_tests.next,
 635                                  sfw_test_instance_t, tsi_list);
 636                 list_del_init(&tsi->tsi_list);
 637                 sfw_destroy_test_instance(tsi);
 638         }
 639
 640         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 641         return;
 642 }
 643
 644 void
 645 sfw_destroy_session (sfw_session_t *sn)
 646 {
 647         sfw_batch_t *batch;
 648
 649         LASSERT (list_empty(&sn->sn_list));
 650         LASSERT (sn != sfw_data.fw_session);
 651
 652         while (!list_empty(&sn->sn_batches)) {
 653                 batch = list_entry(sn->sn_batches.next,
 654                                    sfw_batch_t, bat_list);
 655                 list_del_init(&batch->bat_list);
 656                 sfw_destroy_batch(batch);
 657         }
 658
 659         LIBCFS_FREE(sn, sizeof(*sn));
 660         atomic_dec(&sfw_data.fw_nzombies);
 661         return;
 662 }
 663
 664 void
 665 sfw_unpack_test_req (srpc_msg_t *msg)
 666 {
 667         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 668
 669         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 670         LASSERT (req->tsr_is_client);
 671
 672         if (msg->msg_magic == SRPC_MSG_MAGIC)
 673                 return; /* no flipping needed */
 674
 675         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 676
 677         if (req->tsr_service == SRPC_SERVICE_BRW) {
 678                 test_bulk_req_t *bulk = &req->tsr_u.bulk;
 679
 680                 __swab32s(&bulk->blk_opc);
 681                 __swab32s(&bulk->blk_npg);
 682                 __swab32s(&bulk->blk_flags);
 683                 return;
 684         }
 685
 686         if (req->tsr_service == SRPC_SERVICE_PING) {
 687                 test_ping_req_t *ping = &req->tsr_u.ping;
 688
 689                 __swab32s(&ping->png_size);
 690                 __swab32s(&ping->png_flags);
 691                 return;
 692         }
 693
 694         LBUG ();
 695         return;
 696 }
 697
 698 int
 699 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 700 {
 701         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 702         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 703         srpc_bulk_t         *bk = rpc->srpc_bulk;
 704         int                  ndest = req->tsr_ndest;
 705         sfw_test_unit_t     *tsu;
 706         sfw_test_instance_t *tsi;
 707         int                  i;
 708         int                  rc;
 709
 710         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 711         if (tsi == NULL) {
 712                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 713                         tsb->bat_id.bat_id);
 714                 return -ENOMEM;
 715         }
 716
 717         memset(tsi, 0, sizeof(*tsi));
 718         spin_lock_init(&tsi->tsi_lock);
 719         atomic_set(&tsi->tsi_nactive, 0);
 720         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 721         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 722         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 723
 724         tsi->tsi_stopping      = 0;
 725         tsi->tsi_batch         = tsb;
 726         tsi->tsi_loop          = req->tsr_loop;
 727         tsi->tsi_concur        = req->tsr_concur;
 728         tsi->tsi_service       = req->tsr_service;
 729         tsi->tsi_is_client     = !!(req->tsr_is_client);
 730         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 731
 732         rc = sfw_load_test(tsi);
 733         if (rc != 0) {
 734                 LIBCFS_FREE(tsi, sizeof(*tsi));
 735                 return rc;
 736         }
 737
 738         LASSERT (!sfw_batch_active(tsb));
 739
 740         if (!tsi->tsi_is_client) {
 741                 /* it's test server, just add it to tsb */
 742                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 743                 return 0;
 744         }
 745
 746         LASSERT (bk != NULL);
 747 #ifndef __KERNEL__
 748         LASSERT (bk->bk_pages != NULL);
 749 #endif
 750         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= ndest);
 751         LASSERT (bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
 752
 753         sfw_unpack_test_req(msg);
 754         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 755
 756         for (i = 0; i < ndest; i++) {
 757                 lnet_process_id_t *dests;
 758                 lnet_process_id_t  id;
 759                 int                j;
 760
 761 #ifdef __KERNEL__
 762                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 763                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 764 #else
 765                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 766 #endif
 767                 id = dests[i % SFW_ID_PER_PAGE];
 768                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 769                         sfw_unpack_id(id);
 770
 771                 for (j = 0; j < tsi->tsi_concur; j++) {
 772                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 773                         if (tsu == NULL) {
 774                                 rc = -ENOMEM;
 775                                 CERROR ("Can't allocate tsu for %d\n",
 776                                         tsi->tsi_service);
 777                                 goto error;
 778                         }
 779
 780                         tsu->tsu_dest     = id;
 781                         tsu->tsu_instance = tsi;
 782                         tsu->tsu_private  = NULL;
 783                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 784                 }
 785         }
 786
 787         rc = tsi->tsi_ops->tso_init(tsi);
 788         if (rc == 0) {
 789                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 790                 return 0;
 791         }
 792
 793 error:
 794         LASSERT (rc != 0);
 795         sfw_destroy_test_instance(tsi);
 796         return rc;
 797 }
 798
 799 static void
 800 sfw_test_unit_done (sfw_test_unit_t *tsu)
 801 {
 802         sfw_test_instance_t *tsi = tsu->tsu_instance;
 803         sfw_batch_t         *tsb = tsi->tsi_batch;
 804         sfw_session_t       *sn = tsb->bat_session;
 805
 806         LASSERT (sfw_test_active(tsi));
 807
 808         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 809                 return;
 810
 811         /* the test instance is done */
 812         spin_lock(&tsi->tsi_lock);
 813
 814         tsi->tsi_stopping = 0;
 815
 816         spin_unlock(&tsi->tsi_lock);
 817
 818         spin_lock(&sfw_data.fw_lock);
 819
 820         if (!atomic_dec_and_test(&tsb->bat_nactive) || /* tsb still active */
 821             sn == sfw_data.fw_session) {               /* sn also active */
 822                 spin_unlock(&sfw_data.fw_lock);
 823                 return;
 824         }
 825
 826         LASSERT (!list_empty(&sn->sn_list)); /* I'm a zombie! */
 827
 828         list_for_each_entry (tsb, &sn->sn_batches, bat_list) {
 829                 if (sfw_batch_active(tsb)) {
 830                         spin_unlock(&sfw_data.fw_lock);
 831                         return;
 832                 }
 833         }
 834
 835         list_del_init(&sn->sn_list);
 836         spin_unlock(&sfw_data.fw_lock);
 837
 838         sfw_destroy_session(sn);
 839         return;
 840 }
 841
 842 void
 843 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 844 {
 845         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 846         sfw_test_instance_t *tsi = tsu->tsu_instance;
 847         int                  done = 0;
 848
 849         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 850
 851         spin_lock(&tsi->tsi_lock);
 852
 853         LASSERT (sfw_test_active(tsi));
 854         LASSERT (!list_empty(&rpc->crpc_list));
 855
 856         list_del_init(&rpc->crpc_list);
 857
 858         /* batch is stopping or loop is done or get error */
 859         if (tsi->tsi_stopping ||
 860             tsu->tsu_loop == 0 ||
 861             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 862                 done = 1;
 863
 864         /* dec ref for poster */
 865         srpc_client_rpc_decref(rpc);
 866
 867         spin_unlock(&tsi->tsi_lock);
 868
 869         if (!done) {
 870                 swi_schedule_workitem(&tsu->tsu_worker);
 871                 return;
 872         }
 873
 874         sfw_test_unit_done(tsu);
 875         return;
 876 }
 877
 878 int
 879 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
 880                      int nblk, int blklen, srpc_client_rpc_t **rpcpp)
 881 {
 882         srpc_client_rpc_t   *rpc = NULL;
 883         sfw_test_instance_t *tsi = tsu->tsu_instance;
 884
 885         spin_lock(&tsi->tsi_lock);
 886
 887         LASSERT (sfw_test_active(tsi));
 888
 889         if (!list_empty(&tsi->tsi_free_rpcs)) {
 890                 /* pick request from buffer */
 891                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 892                                  srpc_client_rpc_t, crpc_list);
 893                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 894                 list_del_init(&rpc->crpc_list);
 895
 896                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 897                                      blklen, sfw_test_rpc_done,
 898                                      sfw_test_rpc_fini, tsu);
 899         }
 900
 901         spin_unlock(&tsi->tsi_lock);
 902
 903         if (rpc == NULL)
 904                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 905                                              blklen, sfw_test_rpc_done,
 906                                              sfw_test_rpc_fini, tsu);
 907         if (rpc == NULL) {
 908                 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
 909                 return -ENOMEM;
 910         }
 911
 912         *rpcpp = rpc;
 913         return 0;
 914 }
 915
 916 int
 917 sfw_run_test (swi_workitem_t *wi)
 918 {
 919         sfw_test_unit_t     *tsu = wi->wi_data;
 920         sfw_test_instance_t *tsi = tsu->tsu_instance;
 921         srpc_client_rpc_t   *rpc = NULL;
 922
 923         LASSERT (wi == &tsu->tsu_worker);
 924
 925         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 926                 LASSERT (rpc == NULL);
 927                 goto test_done;
 928         }
 929
 930         LASSERT (rpc != NULL);
 931
 932         spin_lock(&tsi->tsi_lock);
 933
 934         if (tsi->tsi_stopping) {
 935                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 936                 spin_unlock(&tsi->tsi_lock);
 937                 goto test_done;
 938         }
 939
 940         if (tsu->tsu_loop > 0)
 941                 tsu->tsu_loop--;
 942
 943         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 944         spin_unlock(&tsi->tsi_lock);
 945
 946         rpc->crpc_timeout = rpc_timeout;
 947
 948         spin_lock(&rpc->crpc_lock);
 949         srpc_post_rpc(rpc);
 950         spin_unlock(&rpc->crpc_lock);
 951         return 0;
 952
 953 test_done:
 954         /*
 955          * No one can schedule me now since:
 956          * - previous RPC, if any, has done and
 957          * - no new RPC is initiated.
 958          * - my batch is still active; no one can run it again now.
 959          * Cancel pending schedules and prevent future schedule attempts:
 960          */
 961         swi_kill_workitem(wi);
 962         sfw_test_unit_done(tsu);
 963         return 1;
 964 }
 965
 966 int
 967 sfw_run_batch (sfw_batch_t *tsb)
 968 {
 969         swi_workitem_t      *wi;
 970         sfw_test_unit_t     *tsu;
 971         sfw_test_instance_t *tsi;
 972
 973         if (sfw_batch_active(tsb)) {
 974                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
 975                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
 976                 return 0;
 977         }
 978
 979         list_for_each_entry (tsi, &tsb->bat_tests, tsi_list) {
 980                 if (!tsi->tsi_is_client) /* skip server instances */
 981                         continue;
 982
 983                 LASSERT (!tsi->tsi_stopping);
 984                 LASSERT (!sfw_test_active(tsi));
 985
 986                 atomic_inc(&tsb->bat_nactive);
 987
 988                 list_for_each_entry (tsu, &tsi->tsi_units, tsu_list) {
 989                         atomic_inc(&tsi->tsi_nactive);
 990                         tsu->tsu_loop = tsi->tsi_loop;
 991                         wi = &tsu->tsu_worker;
 992                         swi_init_workitem(wi, tsu, sfw_run_test);
 993                         swi_schedule_workitem(wi);
 994                 }
 995         }
 996
 997         return 0;
 998 }
 999
1000 int
1001 sfw_stop_batch (sfw_batch_t *tsb, int force)
1002 {
1003         sfw_test_instance_t *tsi;
1004         srpc_client_rpc_t   *rpc;
1005
1006         if (!sfw_batch_active(tsb)) {
1007                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1008                 return 0;
1009         }
1010
1011         list_for_each_entry (tsi, &tsb->bat_tests, tsi_list) {
1012                 spin_lock(&tsi->tsi_lock);
1013
1014                 if (!tsi->tsi_is_client ||
1015                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1016                         spin_unlock(&tsi->tsi_lock);
1017                         continue;
1018                 }
1019
1020                 tsi->tsi_stopping = 1;
1021
1022                 if (!force) {
1023                         spin_unlock(&tsi->tsi_lock);
1024                         continue;
1025                 }
1026
1027                 /* abort launched rpcs in the test */
1028                 list_for_each_entry (rpc, &tsi->tsi_active_rpcs, crpc_list) {
1029                         spin_lock(&rpc->crpc_lock);
1030
1031                         srpc_abort_rpc(rpc, -EINTR);
1032
1033                         spin_unlock(&rpc->crpc_lock);
1034                 }
1035
1036                 spin_unlock(&tsi->tsi_lock);
1037         }
1038
1039         return 0;
1040 }
1041
1042 int
1043 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1044 {
1045         sfw_test_instance_t *tsi;
1046
1047         if (testidx < 0)
1048                 return -EINVAL;
1049
1050         if (testidx == 0) {
1051                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1052                 return 0;
1053         }
1054
1055         list_for_each_entry (tsi, &tsb->bat_tests, tsi_list) {
1056                 if (testidx-- > 1)
1057                         continue;
1058
1059                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1060                 return 0;
1061         }
1062
1063         return -ENOENT;
1064 }
1065
1066 void
1067 sfw_free_pages (srpc_server_rpc_t *rpc)
1068 {
1069         srpc_free_bulk(rpc->srpc_bulk);
1070         rpc->srpc_bulk = NULL;
1071 }
1072
1073 int
1074 sfw_alloc_pages (srpc_server_rpc_t *rpc, int npages, int sink)
1075 {
1076         LASSERT (rpc->srpc_bulk == NULL);
1077         LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
1078
1079         rpc->srpc_bulk = srpc_alloc_bulk(npages, sink);
1080         if (rpc->srpc_bulk == NULL) return -ENOMEM;
1081
1082         return 0;
1083 }
1084
1085 int
1086 sfw_add_test (srpc_server_rpc_t *rpc)
1087 {
1088         sfw_session_t     *sn = sfw_data.fw_session;
1089         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1090         srpc_test_reqst_t *request;
1091         int                rc;
1092         sfw_batch_t       *bat;
1093
1094         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1095         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1096
1097         if (request->tsr_loop == 0 ||
1098             request->tsr_concur == 0 ||
1099             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1100             request->tsr_ndest > SFW_MAX_NDESTS ||
1101             (request->tsr_is_client && request->tsr_ndest == 0) ||
1102             request->tsr_concur > SFW_MAX_CONCUR ||
1103             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1104             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1105                 reply->tsr_status = EINVAL;
1106                 return 0;
1107         }
1108
1109         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1110             sfw_find_test_case(request->tsr_service) == NULL) {
1111                 reply->tsr_status = ENOENT;
1112                 return 0;
1113         }
1114
1115         bat = sfw_bid2batch(request->tsr_bid);
1116         if (bat == NULL) {
1117                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1118                         rpc->srpc_service->sv_name,
1119                         libcfs_id2str(rpc->srpc_peer));
1120                 return -ENOMEM;
1121         }
1122
1123         if (sfw_batch_active(bat)) {
1124                 reply->tsr_status = EBUSY;
1125                 return 0;
1126         }
1127
1128         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1129                 /* rpc will be resumed later in sfw_bulk_ready */
1130                 return sfw_alloc_pages(rpc,
1131                                        sfw_id_pages(request->tsr_ndest), 1);
1132         }
1133
1134         rc = sfw_add_test_instance(bat, rpc);
1135         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1136                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1137                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1138                 request->tsr_is_client ? "client" : "server",
1139                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1140
1141         reply->tsr_status = (rc < 0) ? -rc : rc;
1142         return 0;
1143 }
1144
1145 int
1146 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1147 {
1148         sfw_session_t *sn = sfw_data.fw_session;
1149         int            rc = 0;
1150         sfw_batch_t   *bat;
1151
1152         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1153
1154         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1155                 reply->bar_status = ESRCH;
1156                 return 0;
1157         }
1158
1159         bat = sfw_find_batch(request->bar_bid);
1160         if (bat == NULL) {
1161                 reply->bar_status = ENOENT;
1162                 return 0;
1163         }
1164
1165         switch (request->bar_opc) {
1166         case SRPC_BATCH_OPC_RUN:
1167                 rc = sfw_run_batch(bat);
1168                 break;
1169
1170         case SRPC_BATCH_OPC_STOP:
1171                 rc = sfw_stop_batch(bat, request->bar_arg);
1172                 break;
1173
1174         case SRPC_BATCH_OPC_QUERY:
1175                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1176                 break;
1177
1178         default:
1179                 return -EINVAL; /* drop it */
1180         }
1181
1182         reply->bar_status = (rc < 0) ? -rc : rc;
1183         return 0;
1184 }
1185
1186 int
1187 sfw_handle_server_rpc (srpc_server_rpc_t *rpc)
1188 {
1189         srpc_service_t *sv = rpc->srpc_service;
1190         srpc_msg_t     *reply = &rpc->srpc_replymsg;
1191         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1192         int             rc = 0;
1193
1194         LASSERT (sfw_data.fw_active_srpc == NULL);
1195         LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1196
1197         spin_lock(&sfw_data.fw_lock);
1198
1199         if (sfw_data.fw_shuttingdown) {
1200                 spin_unlock(&sfw_data.fw_lock);
1201                 return -ESHUTDOWN;
1202         }
1203
1204         /* Remove timer to avoid racing with it or expiring active session */
1205         if (sfw_del_session_timer() != 0) {
1206                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1207                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1208                 spin_unlock(&sfw_data.fw_lock);
1209                 return -EAGAIN;
1210         }
1211
1212         sfw_data.fw_active_srpc = rpc;
1213         spin_unlock(&sfw_data.fw_lock);
1214
1215         sfw_unpack_message(request);
1216         LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1217
1218         switch(sv->sv_id) {
1219         default:
1220                 LBUG ();
1221         case SRPC_SERVICE_TEST:
1222                 rc = sfw_add_test(rpc);
1223                 break;
1224
1225         case SRPC_SERVICE_BATCH:
1226                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1227                                        &reply->msg_body.bat_reply);
1228                 break;
1229
1230         case SRPC_SERVICE_QUERY_STAT:
1231                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1232                                    &reply->msg_body.stat_reply);
1233                 break;
1234
1235         case SRPC_SERVICE_DEBUG:
1236                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1237                                        &reply->msg_body.dbg_reply);
1238                 break;
1239
1240         case SRPC_SERVICE_MAKE_SESSION:
1241                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1242                                       &reply->msg_body.mksn_reply);
1243                 break;
1244
1245         case SRPC_SERVICE_REMOVE_SESSION:
1246                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1247                                         &reply->msg_body.rmsn_reply);
1248                 break;
1249         }
1250
1251         rpc->srpc_done = sfw_server_rpc_done;
1252         spin_lock(&sfw_data.fw_lock);
1253
1254 #ifdef __KERNEL__
1255         if (!sfw_data.fw_shuttingdown)
1256                 sfw_add_session_timer();
1257 #else
1258         LASSERT (!sfw_data.fw_shuttingdown);
1259         sfw_add_session_timer();
1260 #endif
1261
1262         sfw_data.fw_active_srpc = NULL;
1263         spin_unlock(&sfw_data.fw_lock);
1264         return rc;
1265 }
1266
1267 int
1268 sfw_bulk_ready (srpc_server_rpc_t *rpc, int status)
1269 {
1270         srpc_service_t *sv = rpc->srpc_service;
1271         int             rc;
1272
1273         LASSERT (rpc->srpc_bulk != NULL);
1274         LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1275         LASSERT (sfw_data.fw_active_srpc == NULL);
1276         LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1277
1278         spin_lock(&sfw_data.fw_lock);
1279
1280         if (status != 0) {
1281                 CERROR ("Bulk transfer failed for RPC: "
1282                         "service %s, peer %s, status %d\n",
1283                         sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1284                 spin_unlock(&sfw_data.fw_lock);
1285                 return -EIO;
1286         }
1287
1288         if (sfw_data.fw_shuttingdown) {
1289                 spin_unlock(&sfw_data.fw_lock);
1290                 return -ESHUTDOWN;
1291         }
1292
1293         if (sfw_del_session_timer() != 0) {
1294                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1295                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1296                 spin_unlock(&sfw_data.fw_lock);
1297                 return -EAGAIN;
1298         }
1299
1300         sfw_data.fw_active_srpc = rpc;
1301         spin_unlock(&sfw_data.fw_lock);
1302
1303         rc = sfw_add_test(rpc);
1304
1305         spin_lock(&sfw_data.fw_lock);
1306
1307 #ifdef __KERNEL__
1308         if (!sfw_data.fw_shuttingdown)
1309                 sfw_add_session_timer();
1310 #else
1311         LASSERT (!sfw_data.fw_shuttingdown);
1312         sfw_add_session_timer();
1313 #endif
1314
1315         sfw_data.fw_active_srpc = NULL;
1316         spin_unlock(&sfw_data.fw_lock);
1317         return rc;
1318 }
1319
1320 srpc_client_rpc_t *
1321 sfw_create_rpc (lnet_process_id_t peer, int service,
1322                 int nbulkiov, int bulklen,
1323                 void (*done) (srpc_client_rpc_t *), void *priv)
1324 {
1325         srpc_client_rpc_t *rpc;
1326
1327         spin_lock(&sfw_data.fw_lock);
1328
1329         LASSERT (!sfw_data.fw_shuttingdown);
1330         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1331
1332         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1333                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1334                                  srpc_client_rpc_t, crpc_list);
1335                 list_del(&rpc->crpc_list);
1336                 spin_unlock(&sfw_data.fw_lock);
1337
1338                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1339                                      done, sfw_client_rpc_fini, priv);
1340                 return rpc;
1341         }
1342
1343         spin_unlock(&sfw_data.fw_lock);
1344
1345         rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1346                                      nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1347                                      priv);
1348         return rpc;
1349 }
1350
1351 void
1352 sfw_unpack_message (srpc_msg_t *msg)
1353 {
1354         if (msg->msg_magic == SRPC_MSG_MAGIC)
1355                 return; /* no flipping needed */
1356
1357         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1358
1359         __swab32s(&msg->msg_type);
1360
1361         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1362                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1363
1364                 __swab32s(&req->str_type);
1365                 __swab64s(&req->str_rpyid);
1366                 sfw_unpack_sid(req->str_sid);
1367                 return;
1368         }
1369
1370         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1371                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1372
1373                 __swab32s(&rep->str_status);
1374                 sfw_unpack_sid(rep->str_sid);
1375                 sfw_unpack_fw_counters(rep->str_fw);
1376                 sfw_unpack_rpc_counters(rep->str_rpc);
1377                 sfw_unpack_lnet_counters(rep->str_lnet);
1378                 return;
1379         }
1380
1381         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1382                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1383
1384                 __swab64s(&req->mksn_rpyid);
1385                 __swab32s(&req->mksn_force);
1386                 sfw_unpack_sid(req->mksn_sid);
1387                 return;
1388         }
1389
1390         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1391                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1392
1393                 __swab32s(&rep->mksn_status);
1394                 __swab32s(&rep->mksn_timeout);
1395                 sfw_unpack_sid(rep->mksn_sid);
1396                 return;
1397         }
1398
1399         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1400                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1401
1402                 __swab64s(&req->rmsn_rpyid);
1403                 sfw_unpack_sid(req->rmsn_sid);
1404                 return;
1405         }
1406
1407         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1408                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1409
1410                 __swab32s(&rep->rmsn_status);
1411                 sfw_unpack_sid(rep->rmsn_sid);
1412                 return;
1413         }
1414
1415         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1416                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1417
1418                 __swab64s(&req->dbg_rpyid);
1419                 __swab32s(&req->dbg_flags);
1420                 sfw_unpack_sid(req->dbg_sid);
1421                 return;
1422         }
1423
1424         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1425                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1426
1427                 __swab32s(&rep->dbg_nbatch);
1428                 __swab32s(&rep->dbg_timeout);
1429                 sfw_unpack_sid(rep->dbg_sid);
1430                 return;
1431         }
1432
1433         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1434                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1435
1436                 __swab32s(&req->bar_opc);
1437                 __swab64s(&req->bar_rpyid);
1438                 __swab32s(&req->bar_testidx);
1439                 __swab32s(&req->bar_arg);
1440                 sfw_unpack_sid(req->bar_sid);
1441                 __swab64s(&req->bar_bid.bat_id);
1442                 return;
1443         }
1444
1445         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1446                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1447
1448                 __swab32s(&rep->bar_status);
1449                 sfw_unpack_sid(rep->bar_sid);
1450                 return;
1451         }
1452
1453         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1454                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1455
1456                 __swab64s(&req->tsr_rpyid);
1457                 __swab64s(&req->tsr_bulkid);
1458                 __swab32s(&req->tsr_loop);
1459                 __swab32s(&req->tsr_ndest);
1460                 __swab32s(&req->tsr_concur);
1461                 __swab32s(&req->tsr_service);
1462                 sfw_unpack_sid(req->tsr_sid);
1463                 __swab64s(&req->tsr_bid.bat_id);
1464                 return;
1465         }
1466
1467         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1468                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1469
1470                 __swab32s(&rep->tsr_status);
1471                 sfw_unpack_sid(rep->tsr_sid);
1472                 return;
1473         }
1474
1475         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1476                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1477
1478                 __swab64s(&req->join_rpyid);
1479                 sfw_unpack_sid(req->join_sid);
1480                 return;
1481         }
1482
1483         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1484                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1485
1486                 __swab32s(&rep->join_status);
1487                 __swab32s(&rep->join_timeout);
1488                 sfw_unpack_sid(rep->join_sid);
1489                 return;
1490         }
1491
1492         LBUG ();
1493         return;
1494 }
1495
1496 void
1497 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1498 {
1499         LASSERT (atomic_read(&rpc->crpc_refcount) > 0);
1500         LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1501
1502         spin_lock(&rpc->crpc_lock);
1503         srpc_abort_rpc(rpc, -EINTR);
1504         spin_unlock(&rpc->crpc_lock);
1505         return;
1506 }
1507
1508 void
1509 sfw_post_rpc (srpc_client_rpc_t *rpc)
1510 {
1511         spin_lock(&rpc->crpc_lock);
1512
1513         LASSERT (!rpc->crpc_closed);
1514         LASSERT (!rpc->crpc_aborted);
1515         LASSERT (list_empty(&rpc->crpc_list));
1516         LASSERT (!sfw_data.fw_shuttingdown);
1517
1518         rpc->crpc_timeout = rpc_timeout;
1519         srpc_post_rpc(rpc);
1520
1521         spin_unlock(&rpc->crpc_lock);
1522         return;
1523 }
1524
1525 static srpc_service_t sfw_services[] =
1526 {
1527         {
1528                 .sv_name = "debug",
1529                 .sv_id   = SRPC_SERVICE_DEBUG,
1530         },
1531         {
1532                 .sv_name = "query stats",
1533                 .sv_id   = SRPC_SERVICE_QUERY_STAT,
1534         },
1535         {
1536                 .sv_name = "make sessin",
1537                 .sv_id   = SRPC_SERVICE_MAKE_SESSION,
1538         },
1539         {
1540                 .sv_name = "remove session",
1541                 .sv_id   = SRPC_SERVICE_REMOVE_SESSION,
1542         },
1543         {
1544                 .sv_name = "batch service",
1545                 .sv_id   = SRPC_SERVICE_BATCH,
1546         },
1547         {
1548                 .sv_name = "test service",
1549                 .sv_id   = SRPC_SERVICE_TEST,
1550         },
1551         {       .sv_name = NULL, }
1552 };
1553
1554 extern sfw_test_client_ops_t ping_test_client;
1555 extern srpc_service_t        ping_test_service;
1556
1557 extern sfw_test_client_ops_t brw_test_client;
1558 extern srpc_service_t        brw_test_service;
1559
1560 int
1561 sfw_startup (void)
1562 {
1563         int              i;
1564         int              rc;
1565         int              error;
1566         srpc_service_t  *sv;
1567         sfw_test_case_t *tsc;
1568
1569 #ifndef __KERNEL__
1570         char *s;
1571
1572         s = getenv("SESSION_TIMEOUT");
1573         session_timeout = s != NULL ? atoi(s) : session_timeout;
1574
1575         s = getenv("BRW_INJECT_ERRORS");
1576         brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
1577
1578         s = getenv("RPC_TIMEOUT");
1579         rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1580 #endif
1581
1582         if (session_timeout < 0) {
1583                 CERROR ("Session timeout must be non-negative: %d\n",
1584                         session_timeout);
1585                 return -EINVAL;
1586         }
1587
1588         if (rpc_timeout < 0) {
1589                 CERROR ("RPC timeout must be non-negative: %d\n",
1590                         rpc_timeout);
1591                 return -EINVAL;
1592         }
1593
1594         if (session_timeout == 0)
1595                 CWARN ("Zero session_timeout specified "
1596                        "- test sessions never expire.\n");
1597
1598         if (rpc_timeout == 0)
1599                 CWARN ("Zero rpc_timeout specified "
1600                        "- test RPC never expire.\n");
1601
1602         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1603
1604         sfw_data.fw_session     = NULL;
1605         sfw_data.fw_active_srpc = NULL;
1606         spin_lock_init(&sfw_data.fw_lock);
1607         atomic_set(&sfw_data.fw_nzombies, 0);
1608         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1609         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1610         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1611
1612         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1613         LASSERT (rc == 0);
1614         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1615         LASSERT (rc == 0);
1616
1617         error = 0;
1618         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
1619                 sv = tsc->tsc_srv_service;
1620                 sv->sv_concur = SFW_TEST_CONCURRENCY;
1621
1622                 rc = srpc_add_service(sv);
1623                 LASSERT (rc != -EBUSY);
1624                 if (rc != 0) {
1625                         CWARN ("Failed to add %s service: %d\n",
1626                                sv->sv_name, rc);
1627                         error = rc;
1628                 }
1629         }
1630
1631         for (i = 0; ; i++) {
1632                 sv = &sfw_services[i];
1633                 if (sv->sv_name == NULL) break;
1634
1635                 sv->sv_bulk_ready = NULL;
1636                 sv->sv_handler    = sfw_handle_server_rpc;
1637                 sv->sv_concur     = SFW_SERVICE_CONCURRENCY;
1638                 if (sv->sv_id == SRPC_SERVICE_TEST)
1639                         sv->sv_bulk_ready = sfw_bulk_ready;
1640
1641                 rc = srpc_add_service(sv);
1642                 LASSERT (rc != -EBUSY);
1643                 if (rc != 0) {
1644                         CWARN ("Failed to add %s service: %d\n",
1645                                sv->sv_name, rc);
1646                         error = rc;
1647                 }
1648
1649                 /* about to sfw_shutdown, no need to add buffer */
1650                 if (error) continue;
1651
1652                 rc = srpc_service_add_buffers(sv, SFW_POST_BUFFERS);
1653                 if (rc != SFW_POST_BUFFERS) {
1654                         CWARN ("Failed to reserve enough buffers: "
1655                                "service %s, %d needed, %d reserved\n",
1656                                sv->sv_name, SFW_POST_BUFFERS, rc);
1657                         error = -ENOMEM;
1658                 }
1659         }
1660
1661         if (error != 0)
1662                 sfw_shutdown();
1663         return error;
1664 }
1665
1666 void
1667 sfw_shutdown (void)
1668 {
1669         srpc_service_t  *sv;
1670         sfw_test_case_t *tsc;
1671         int              i;
1672
1673         spin_lock(&sfw_data.fw_lock);
1674
1675         sfw_data.fw_shuttingdown = 1;
1676 #ifdef __KERNEL__
1677         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1678                        "waiting for active RPC to finish.\n");
1679 #else
1680         LASSERT (sfw_data.fw_active_srpc == NULL);
1681 #endif
1682
1683         if (sfw_del_session_timer() != 0)
1684                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1685                                "waiting for session timer to explode.\n");
1686
1687         sfw_deactivate_session();
1688         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1689                        sfw_data.fw_lock,
1690                        "waiting for %d zombie sessions to die.\n",
1691                        atomic_read(&sfw_data.fw_nzombies));
1692
1693         spin_unlock(&sfw_data.fw_lock);
1694
1695         for (i = 0; ; i++) {
1696                 sv = &sfw_services[i];
1697                 if (sv->sv_name == NULL)
1698                         break;
1699
1700                 srpc_shutdown_service(sv);
1701                 srpc_remove_service(sv);
1702         }
1703
1704         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
1705                 sv = tsc->tsc_srv_service;
1706                 srpc_shutdown_service(sv);
1707                 srpc_remove_service(sv);
1708         }
1709
1710         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1711                 srpc_client_rpc_t *rpc;
1712
1713                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1714                                  srpc_client_rpc_t, crpc_list);
1715                 list_del(&rpc->crpc_list);
1716
1717                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1718         }
1719
1720         for (i = 0; ; i++) {
1721                 sv = &sfw_services[i];
1722                 if (sv->sv_name == NULL)
1723                         break;
1724
1725                 srpc_wait_service_shutdown(sv);
1726         }
1727
1728         while (!list_empty(&sfw_data.fw_tests)) {
1729                 tsc = list_entry(sfw_data.fw_tests.next,
1730                                  sfw_test_case_t, tsc_list);
1731
1732                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1733
1734                 list_del(&tsc->tsc_list);
1735                 LIBCFS_FREE(tsc, sizeof(*tsc));
1736         }
1737
1738         return;
1739 }