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 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
  47
  48 int brw_inject_errors = 0;
  49 CFS_MODULE_PARM(brw_inject_errors, "i", int, 0644,
  50                 "# data errors to inject randomly, zero by default");
  51
  52 static int session_timeout = 100;
  53 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  54                 "test session timeout in seconds (100 by default, 0 == never)");
  55
  56 #define SFW_TEST_CONCURRENCY     128
  57 #define SFW_TEST_RPC_TIMEOUT     64
  58 #define SFW_CLIENT_RPC_TIMEOUT   64  /* in seconds */
  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         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 137                                        sfw_test_case_t, tsc_list) {
 138                 if (tsc->tsc_srv_service->sv_id == id)
 139                         return tsc;
 140         }
 141
 142         return NULL;
 143 }
 144
 145 static int
 146 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 147 {
 148         sfw_test_case_t *tsc;
 149
 150         if (sfw_find_test_case(service->sv_id) != NULL) {
 151                 CERROR ("Failed to register test %s (%d)\n",
 152                         service->sv_name, service->sv_id);
 153                 return -EEXIST;
 154         }
 155
 156         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 157         if (tsc == NULL)
 158                 return -ENOMEM;
 159
 160         memset(tsc, 0, sizeof(sfw_test_case_t));
 161         tsc->tsc_cli_ops     = cliops;
 162         tsc->tsc_srv_service = service;
 163
 164         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 165         return 0;
 166 }
 167
 168 void
 169 sfw_add_session_timer (void)
 170 {
 171         sfw_session_t *sn = sfw_data.fw_session;
 172         stt_timer_t   *timer = &sn->sn_timer;
 173
 174         LASSERT (!sfw_data.fw_shuttingdown);
 175
 176         if (sn == NULL || sn->sn_timeout == 0)
 177                 return;
 178
 179         LASSERT (!sn->sn_timer_active);
 180
 181         sn->sn_timer_active = 1;
 182         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 183                                           cfs_time_current_sec());
 184         stt_add_timer(timer);
 185         return;
 186 }
 187
 188 int
 189 sfw_del_session_timer (void)
 190 {
 191         sfw_session_t *sn = sfw_data.fw_session;
 192
 193         if (sn == NULL || !sn->sn_timer_active)
 194                 return 0;
 195
 196         LASSERT (sn->sn_timeout != 0);
 197
 198         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 199                 sn->sn_timer_active = 0;
 200                 return 0;
 201         }
 202
 203 #ifndef __KERNEL__
 204         /* Racing is impossible in single-threaded userland selftest */
 205         LBUG();
 206 #endif
 207         return EBUSY; /* racing with sfw_session_expired() */
 208 }
 209
 210 /* called with sfw_data.fw_lock held */
 211 static void
 212 sfw_deactivate_session (void)
 213 {
 214         sfw_session_t *sn = sfw_data.fw_session;
 215         int            nactive = 0;
 216         sfw_batch_t   *tsb;
 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         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 227                                        sfw_batch_t, 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         return;
 244 }
 245
 246 #ifndef __KERNEL__
 247
 248 int
 249 sfw_session_removed (void)
 250 {
 251         return (sfw_data.fw_session == NULL) ? 1 : 0;
 252 }
 253
 254 #endif
 255
 256 void
 257 sfw_session_expired (void *data)
 258 {
 259         sfw_session_t *sn = data;
 260
 261         spin_lock(&sfw_data.fw_lock);
 262
 263         LASSERT (sn->sn_timer_active);
 264         LASSERT (sn == sfw_data.fw_session);
 265
 266         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 267                libcfs_nid2str(sn->sn_id.ses_nid),
 268                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 269
 270         sn->sn_timer_active = 0;
 271         sfw_deactivate_session();
 272
 273         spin_unlock(&sfw_data.fw_lock);
 274         return;
 275 }
 276
 277 static inline void
 278 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
 279 {
 280         stt_timer_t *timer = &sn->sn_timer;
 281
 282         memset(sn, 0, sizeof(sfw_session_t));
 283         CFS_INIT_LIST_HEAD(&sn->sn_list);
 284         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 285         atomic_set(&sn->sn_brw_errors, 0);
 286         atomic_set(&sn->sn_ping_errors, 0);
 287         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 288
 289         sn->sn_timer_active = 0;
 290         sn->sn_id           = sid;
 291         sn->sn_timeout      = session_timeout;
 292
 293         timer->stt_data = sn;
 294         timer->stt_func = sfw_session_expired;
 295         CFS_INIT_LIST_HEAD(&timer->stt_list);
 296 }
 297
 298 /* completion handler for incoming framework RPCs */
 299 void
 300 sfw_server_rpc_done (srpc_server_rpc_t *rpc)
 301 {
 302         srpc_service_t *sv = rpc->srpc_service;
 303         int             status = rpc->srpc_status;
 304
 305         CDEBUG (D_NET,
 306                 "Incoming framework RPC done: "
 307                 "service %s, peer %s, status %s:%d\n",
 308                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 309                 swi_state2str(rpc->srpc_wi.wi_state),
 310                 status);
 311
 312         if (rpc->srpc_bulk != NULL)
 313                 sfw_free_pages(rpc);
 314         return;
 315 }
 316
 317 void
 318 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 319 {
 320         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 321         LASSERT (list_empty(&rpc->crpc_list));
 322         LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
 323 #ifndef __KERNEL__
 324         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 325 #endif
 326
 327         CDEBUG (D_NET,
 328                 "Outgoing framework RPC done: "
 329                 "service %d, peer %s, status %s:%d:%d\n",
 330                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 331                 swi_state2str(rpc->crpc_wi.wi_state),
 332                 rpc->crpc_aborted, rpc->crpc_status);
 333
 334         spin_lock(&sfw_data.fw_lock);
 335
 336         /* my callers must finish all RPCs before shutting me down */
 337         LASSERT (!sfw_data.fw_shuttingdown);
 338         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 339
 340         spin_unlock(&sfw_data.fw_lock);
 341         return;
 342 }
 343
 344 sfw_batch_t *
 345 sfw_find_batch (lst_bid_t bid)
 346 {
 347         sfw_session_t *sn = sfw_data.fw_session;
 348         sfw_batch_t   *bat;
 349
 350         LASSERT (sn != NULL);
 351
 352         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 353                                        sfw_batch_t, bat_list) {
 354                 if (bat->bat_id.bat_id == bid.bat_id)
 355                         return bat;
 356         }
 357
 358         return NULL;
 359 }
 360
 361 sfw_batch_t *
 362 sfw_bid2batch (lst_bid_t bid)
 363 {
 364         sfw_session_t *sn = sfw_data.fw_session;
 365         sfw_batch_t   *bat;
 366
 367         LASSERT (sn != NULL);
 368
 369         bat = sfw_find_batch(bid);
 370         if (bat != NULL)
 371                 return bat;
 372
 373         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 374         if (bat == NULL)
 375                 return NULL;
 376
 377         bat->bat_error    = 0;
 378         bat->bat_session  = sn;
 379         bat->bat_id       = bid;
 380         atomic_set(&bat->bat_nactive, 0);
 381         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 382
 383         list_add_tail(&bat->bat_list, &sn->sn_batches);
 384         return bat;
 385 }
 386
 387 int
 388 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 389 {
 390         sfw_session_t  *sn = sfw_data.fw_session;
 391         sfw_counters_t *cnt = &reply->str_fw;
 392         sfw_batch_t    *bat;
 393
 394         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 395
 396         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 397                 reply->str_status = EINVAL;
 398                 return 0;
 399         }
 400
 401         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 402                 reply->str_status = ESRCH;
 403                 return 0;
 404         }
 405
 406         LNET_LOCK();
 407         reply->str_lnet = the_lnet.ln_counters;
 408         LNET_UNLOCK();
 409
 410         srpc_get_counters(&reply->str_rpc);
 411
 412         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 413         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 414         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 415
 416         cnt->active_tests = cnt->active_batches = 0;
 417         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 418                                        sfw_batch_t, bat_list) {
 419                 int n = atomic_read(&bat->bat_nactive);
 420
 421                 if (n > 0) {
 422                         cnt->active_batches++;
 423                         cnt->active_tests += n;
 424                 }
 425         }
 426
 427         reply->str_status = 0;
 428         return 0;
 429 }
 430
 431 int
 432 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 433 {
 434         sfw_session_t *sn = sfw_data.fw_session;
 435
 436         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 437                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 438                 reply->mksn_status = EINVAL;
 439                 return 0;
 440         }
 441
 442         if (sn != NULL && !request->mksn_force) {
 443                 reply->mksn_sid    = sn->sn_id;
 444                 reply->mksn_status = EBUSY;
 445                 strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 446                 return 0;
 447         }
 448
 449         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 450         if (sn == NULL) {
 451                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 452                 return -ENOMEM;
 453         }
 454
 455         sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
 456
 457         spin_lock(&sfw_data.fw_lock);
 458
 459         sfw_deactivate_session();
 460         LASSERT (sfw_data.fw_session == NULL);
 461         sfw_data.fw_session = sn;
 462
 463         spin_unlock(&sfw_data.fw_lock);
 464
 465         reply->mksn_status  = 0;
 466         reply->mksn_sid     = sn->sn_id;
 467         reply->mksn_timeout = sn->sn_timeout;
 468         return 0;
 469 }
 470
 471 int
 472 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 473 {
 474         sfw_session_t *sn = sfw_data.fw_session;
 475
 476         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 477
 478         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 479                 reply->rmsn_status = EINVAL;
 480                 return 0;
 481         }
 482
 483         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 484                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 485                 return 0;
 486         }
 487
 488         spin_lock(&sfw_data.fw_lock);
 489         sfw_deactivate_session();
 490         spin_unlock(&sfw_data.fw_lock);
 491
 492         reply->rmsn_status = 0;
 493         reply->rmsn_sid    = LST_INVALID_SID;
 494         LASSERT (sfw_data.fw_session == NULL);
 495         return 0;
 496 }
 497
 498 int
 499 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 500 {
 501         sfw_session_t *sn = sfw_data.fw_session;
 502
 503         if (sn == NULL) {
 504                 reply->dbg_status = ESRCH;
 505                 reply->dbg_sid    = LST_INVALID_SID;
 506                 return 0;
 507         }
 508
 509         reply->dbg_status  = 0;
 510         reply->dbg_sid     = sn->sn_id;
 511         reply->dbg_timeout = sn->sn_timeout;
 512         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 513
 514         return 0;
 515 }
 516
 517 void
 518 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 519 {
 520         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 521         sfw_test_instance_t *tsi = tsu->tsu_instance;
 522
 523         /* Called with hold of tsi->tsi_lock */
 524         LASSERT (list_empty(&rpc->crpc_list));
 525         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 526 }
 527
 528 int
 529 sfw_load_test (sfw_test_instance_t *tsi)
 530 {
 531         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 532         int              nrequired = sfw_test_buffers(tsi);
 533         int              nposted;
 534
 535         LASSERT (tsc != NULL);
 536
 537         if (tsi->tsi_is_client) {
 538                 tsi->tsi_ops = tsc->tsc_cli_ops;
 539                 return 0;
 540         }
 541
 542         nposted = srpc_service_add_buffers(tsc->tsc_srv_service, nrequired);
 543         if (nposted != nrequired) {
 544                 CWARN ("Failed to reserve enough buffers: "
 545                        "service %s, %d needed, %d reserved\n",
 546                        tsc->tsc_srv_service->sv_name, nrequired, nposted);
 547                 srpc_service_remove_buffers(tsc->tsc_srv_service, nposted);
 548                 return -ENOMEM;
 549         }
 550
 551         CDEBUG (D_NET, "Reserved %d buffers for test %s\n",
 552                 nposted, tsc->tsc_srv_service->sv_name);
 553         return 0;
 554 }
 555
 556 void
 557 sfw_unload_test (sfw_test_instance_t *tsi)
 558 {
 559         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 560
 561         LASSERT (tsc != NULL);
 562
 563         if (!tsi->tsi_is_client)
 564                 srpc_service_remove_buffers(tsc->tsc_srv_service,
 565                                             sfw_test_buffers(tsi));
 566         return;
 567 }
 568
 569 void
 570 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 571 {
 572         srpc_client_rpc_t *rpc;
 573         sfw_test_unit_t   *tsu;
 574
 575         if (!tsi->tsi_is_client) goto clean;
 576
 577         tsi->tsi_ops->tso_fini(tsi);
 578
 579         LASSERT (!tsi->tsi_stopping);
 580         LASSERT (list_empty(&tsi->tsi_active_rpcs));
 581         LASSERT (!sfw_test_active(tsi));
 582
 583         while (!list_empty(&tsi->tsi_units)) {
 584                 tsu = list_entry(tsi->tsi_units.next,
 585                                  sfw_test_unit_t, tsu_list);
 586                 list_del(&tsu->tsu_list);
 587                 LIBCFS_FREE(tsu, sizeof(*tsu));
 588         }
 589
 590         while (!list_empty(&tsi->tsi_free_rpcs)) {
 591                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 592                                  srpc_client_rpc_t, crpc_list);
 593                 list_del(&rpc->crpc_list);
 594                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 595         }
 596
 597 clean:
 598         sfw_unload_test(tsi);
 599         LIBCFS_FREE(tsi, sizeof(*tsi));
 600         return;
 601 }
 602
 603 void
 604 sfw_destroy_batch (sfw_batch_t *tsb)
 605 {
 606         sfw_test_instance_t *tsi;
 607
 608         LASSERT (!sfw_batch_active(tsb));
 609         LASSERT (list_empty(&tsb->bat_list));
 610
 611         while (!list_empty(&tsb->bat_tests)) {
 612                 tsi = list_entry(tsb->bat_tests.next,
 613                                  sfw_test_instance_t, tsi_list);
 614                 list_del_init(&tsi->tsi_list);
 615                 sfw_destroy_test_instance(tsi);
 616         }
 617
 618         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 619         return;
 620 }
 621
 622 void
 623 sfw_destroy_session (sfw_session_t *sn)
 624 {
 625         sfw_batch_t *batch;
 626
 627         LASSERT (list_empty(&sn->sn_list));
 628         LASSERT (sn != sfw_data.fw_session);
 629
 630         while (!list_empty(&sn->sn_batches)) {
 631                 batch = list_entry(sn->sn_batches.next,
 632                                    sfw_batch_t, bat_list);
 633                 list_del_init(&batch->bat_list);
 634                 sfw_destroy_batch(batch);
 635         }
 636
 637         LIBCFS_FREE(sn, sizeof(*sn));
 638         atomic_dec(&sfw_data.fw_nzombies);
 639         return;
 640 }
 641
 642 void
 643 sfw_unpack_test_req (srpc_msg_t *msg)
 644 {
 645         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 646
 647         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 648         LASSERT (req->tsr_is_client);
 649
 650         if (msg->msg_magic == SRPC_MSG_MAGIC)
 651                 return; /* no flipping needed */
 652
 653         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 654
 655         if (req->tsr_service == SRPC_SERVICE_BRW) {
 656                 test_bulk_req_t *bulk = &req->tsr_u.bulk;
 657
 658                 __swab32s(&bulk->blk_opc);
 659                 __swab32s(&bulk->blk_npg);
 660                 __swab32s(&bulk->blk_flags);
 661                 return;
 662         }
 663
 664         if (req->tsr_service == SRPC_SERVICE_PING) {
 665                 test_ping_req_t *ping = &req->tsr_u.ping;
 666
 667                 __swab32s(&ping->png_size);
 668                 __swab32s(&ping->png_flags);
 669                 return;
 670         }
 671
 672         LBUG ();
 673         return;
 674 }
 675
 676 int
 677 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 678 {
 679         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 680         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 681         srpc_bulk_t         *bk = rpc->srpc_bulk;
 682         int                  ndest = req->tsr_ndest;
 683         sfw_test_unit_t     *tsu;
 684         sfw_test_instance_t *tsi;
 685         int                  i;
 686         int                  rc;
 687
 688         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 689         if (tsi == NULL) {
 690                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 691                         tsb->bat_id.bat_id);
 692                 return -ENOMEM;
 693         }
 694
 695         memset(tsi, 0, sizeof(*tsi));
 696         spin_lock_init(&tsi->tsi_lock);
 697         atomic_set(&tsi->tsi_nactive, 0);
 698         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 699         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 700         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 701
 702         tsi->tsi_stopping      = 0;
 703         tsi->tsi_batch         = tsb;
 704         tsi->tsi_loop          = req->tsr_loop;
 705         tsi->tsi_concur        = req->tsr_concur;
 706         tsi->tsi_service       = req->tsr_service;
 707         tsi->tsi_is_client     = !!(req->tsr_is_client);
 708         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 709
 710         rc = sfw_load_test(tsi);
 711         if (rc != 0) {
 712                 LIBCFS_FREE(tsi, sizeof(*tsi));
 713                 return rc;
 714         }
 715
 716         LASSERT (!sfw_batch_active(tsb));
 717
 718         if (!tsi->tsi_is_client) {
 719                 /* it's test server, just add it to tsb */
 720                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 721                 return 0;
 722         }
 723
 724         LASSERT (bk != NULL);
 725 #ifndef __KERNEL__
 726         LASSERT (bk->bk_pages != NULL);
 727 #endif
 728         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 729         LASSERT ((unsigned int)bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
 730
 731         sfw_unpack_test_req(msg);
 732         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 733
 734         for (i = 0; i < ndest; i++) {
 735                 lnet_process_id_t *dests;
 736                 lnet_process_id_t  id;
 737                 int                j;
 738
 739 #ifdef __KERNEL__
 740                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 741                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 742 #else
 743                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 744 #endif
 745                 id = dests[i % SFW_ID_PER_PAGE];
 746                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 747                         sfw_unpack_id(id);
 748
 749                 for (j = 0; j < tsi->tsi_concur; j++) {
 750                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 751                         if (tsu == NULL) {
 752                                 rc = -ENOMEM;
 753                                 CERROR ("Can't allocate tsu for %d\n",
 754                                         tsi->tsi_service);
 755                                 goto error;
 756                         }
 757
 758                         tsu->tsu_dest     = id;
 759                         tsu->tsu_instance = tsi;
 760                         tsu->tsu_private  = NULL;
 761                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 762                 }
 763         }
 764
 765         rc = tsi->tsi_ops->tso_init(tsi);
 766         if (rc == 0) {
 767                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 768                 return 0;
 769         }
 770
 771 error:
 772         LASSERT (rc != 0);
 773         sfw_destroy_test_instance(tsi);
 774         return rc;
 775 }
 776
 777 static void
 778 sfw_test_unit_done (sfw_test_unit_t *tsu)
 779 {
 780         sfw_test_instance_t *tsi = tsu->tsu_instance;
 781         sfw_batch_t         *tsb = tsi->tsi_batch;
 782         sfw_session_t       *sn = tsb->bat_session;
 783
 784         LASSERT (sfw_test_active(tsi));
 785
 786         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 787                 return;
 788
 789         /* the test instance is done */
 790         spin_lock(&tsi->tsi_lock);
 791
 792         tsi->tsi_stopping = 0;
 793
 794         spin_unlock(&tsi->tsi_lock);
 795
 796         spin_lock(&sfw_data.fw_lock);
 797
 798         if (!atomic_dec_and_test(&tsb->bat_nactive) || /* tsb still active */
 799             sn == sfw_data.fw_session) {               /* sn also active */
 800                 spin_unlock(&sfw_data.fw_lock);
 801                 return;
 802         }
 803
 804         LASSERT (!list_empty(&sn->sn_list)); /* I'm a zombie! */
 805
 806         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 807                                        sfw_batch_t, bat_list) {
 808                 if (sfw_batch_active(tsb)) {
 809                         spin_unlock(&sfw_data.fw_lock);
 810                         return;
 811                 }
 812         }
 813
 814         list_del_init(&sn->sn_list);
 815         spin_unlock(&sfw_data.fw_lock);
 816
 817         sfw_destroy_session(sn);
 818         return;
 819 }
 820
 821 void
 822 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 823 {
 824         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 825         sfw_test_instance_t *tsi = tsu->tsu_instance;
 826         int                  done = 0;
 827
 828         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 829
 830         spin_lock(&tsi->tsi_lock);
 831
 832         LASSERT (sfw_test_active(tsi));
 833         LASSERT (!list_empty(&rpc->crpc_list));
 834
 835         list_del_init(&rpc->crpc_list);
 836
 837         /* batch is stopping or loop is done or get error */
 838         if (tsi->tsi_stopping ||
 839             tsu->tsu_loop == 0 ||
 840             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 841                 done = 1;
 842
 843         /* dec ref for poster */
 844         srpc_client_rpc_decref(rpc);
 845
 846         spin_unlock(&tsi->tsi_lock);
 847
 848         if (!done) {
 849                 swi_schedule_workitem(&tsu->tsu_worker);
 850                 return;
 851         }
 852
 853         sfw_test_unit_done(tsu);
 854         return;
 855 }
 856
 857 int
 858 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
 859                      int nblk, int blklen, srpc_client_rpc_t **rpcpp)
 860 {
 861         srpc_client_rpc_t   *rpc = NULL;
 862         sfw_test_instance_t *tsi = tsu->tsu_instance;
 863
 864         spin_lock(&tsi->tsi_lock);
 865
 866         LASSERT (sfw_test_active(tsi));
 867
 868         if (!list_empty(&tsi->tsi_free_rpcs)) {
 869                 /* pick request from buffer */
 870                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 871                                  srpc_client_rpc_t, crpc_list);
 872                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 873                 list_del_init(&rpc->crpc_list);
 874
 875                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 876                                      blklen, sfw_test_rpc_done,
 877                                      sfw_test_rpc_fini, tsu);
 878         }
 879
 880         spin_unlock(&tsi->tsi_lock);
 881
 882         if (rpc == NULL)
 883                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 884                                              blklen, sfw_test_rpc_done,
 885                                              sfw_test_rpc_fini, tsu);
 886         if (rpc == NULL) {
 887                 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
 888                 return -ENOMEM;
 889         }
 890
 891         *rpcpp = rpc;
 892         return 0;
 893 }
 894
 895 int
 896 sfw_run_test (swi_workitem_t *wi)
 897 {
 898         sfw_test_unit_t     *tsu = wi->wi_data;
 899         sfw_test_instance_t *tsi = tsu->tsu_instance;
 900         srpc_client_rpc_t   *rpc = NULL;
 901
 902         LASSERT (wi == &tsu->tsu_worker);
 903
 904         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 905                 LASSERT (rpc == NULL);
 906                 goto test_done;
 907         }
 908
 909         LASSERT (rpc != NULL);
 910
 911         spin_lock(&tsi->tsi_lock);
 912
 913         if (tsi->tsi_stopping) {
 914                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 915                 spin_unlock(&tsi->tsi_lock);
 916                 goto test_done;
 917         }
 918
 919         if (tsu->tsu_loop > 0)
 920                 tsu->tsu_loop--;
 921
 922         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 923         spin_unlock(&tsi->tsi_lock);
 924
 925         rpc->crpc_timeout = SFW_TEST_RPC_TIMEOUT;
 926
 927         spin_lock(&rpc->crpc_lock);
 928         srpc_post_rpc(rpc);
 929         spin_unlock(&rpc->crpc_lock);
 930         return 0;
 931
 932 test_done:
 933         /*
 934          * No one can schedule me now since:
 935          * - previous RPC, if any, has done and
 936          * - no new RPC is initiated.
 937          * - my batch is still active; no one can run it again now.
 938          * Cancel pending schedules and prevent future schedule attempts:
 939          */
 940         swi_kill_workitem(wi);
 941         sfw_test_unit_done(tsu);
 942         return 1;
 943 }
 944
 945 int
 946 sfw_run_batch (sfw_batch_t *tsb)
 947 {
 948         swi_workitem_t      *wi;
 949         sfw_test_unit_t     *tsu;
 950         sfw_test_instance_t *tsi;
 951
 952         if (sfw_batch_active(tsb)) {
 953                 CDEBUG (D_NET, "Can't start active batch: "LPU64" (%d)\n",
 954                         tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
 955                 return -EPERM;
 956         }
 957
 958         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
 959                                        sfw_test_instance_t, tsi_list) {
 960                 if (!tsi->tsi_is_client) /* skip server instances */
 961                         continue;
 962
 963                 LASSERT (!tsi->tsi_stopping);
 964                 LASSERT (!sfw_test_active(tsi));
 965
 966                 atomic_inc(&tsb->bat_nactive);
 967
 968                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
 969                                                sfw_test_unit_t, tsu_list) {
 970                         atomic_inc(&tsi->tsi_nactive);
 971                         tsu->tsu_loop = tsi->tsi_loop;
 972                         wi = &tsu->tsu_worker;
 973                         swi_init_workitem(wi, tsu, sfw_run_test);
 974                         swi_schedule_workitem(wi);
 975                 }
 976         }
 977
 978         return 0;
 979 }
 980
 981 int
 982 sfw_stop_batch (sfw_batch_t *tsb, int force)
 983 {
 984         sfw_test_instance_t *tsi;
 985         srpc_client_rpc_t   *rpc;
 986
 987         if (!sfw_batch_active(tsb))
 988                 return -EPERM;
 989
 990         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
 991                                        sfw_test_instance_t, tsi_list) {
 992                 spin_lock(&tsi->tsi_lock);
 993
 994                 if (!tsi->tsi_is_client ||
 995                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
 996                         spin_unlock(&tsi->tsi_lock);
 997                         continue;
 998                 }
 999
1000                 tsi->tsi_stopping = 1;
1001
1002                 if (!force) {
1003                         spin_unlock(&tsi->tsi_lock);
1004                         continue;
1005                 }
1006
1007                 /* abort launched rpcs in the test */
1008                 cfs_list_for_each_entry_typed (rpc, &tsi->tsi_active_rpcs,
1009                                                srpc_client_rpc_t, crpc_list) {
1010                         spin_lock(&rpc->crpc_lock);
1011
1012                         srpc_abort_rpc(rpc, -EINTR);
1013
1014                         spin_unlock(&rpc->crpc_lock);
1015                 }
1016
1017                 spin_unlock(&tsi->tsi_lock);
1018         }
1019
1020         return 0;
1021 }
1022
1023 int
1024 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1025 {
1026         sfw_test_instance_t *tsi;
1027
1028         if (testidx < 0)
1029                 return -EINVAL;
1030
1031         if (testidx == 0) {
1032                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1033                 return 0;
1034         }
1035
1036         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1037                                        sfw_test_instance_t, tsi_list) {
1038                 if (testidx-- > 1)
1039                         continue;
1040
1041                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1042                 return 0;
1043         }
1044
1045         return -ENOENT;
1046 }
1047
1048 void
1049 sfw_free_pages (srpc_server_rpc_t *rpc)
1050 {
1051         srpc_free_bulk(rpc->srpc_bulk);
1052         rpc->srpc_bulk = NULL;
1053 }
1054
1055 int
1056 sfw_alloc_pages (srpc_server_rpc_t *rpc, int npages, int sink)
1057 {
1058         LASSERT (rpc->srpc_bulk == NULL);
1059         LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
1060
1061         rpc->srpc_bulk = srpc_alloc_bulk(npages, sink);
1062         if (rpc->srpc_bulk == NULL) return -ENOMEM;
1063
1064         return 0;
1065 }
1066
1067 int
1068 sfw_add_test (srpc_server_rpc_t *rpc)
1069 {
1070         sfw_session_t     *sn = sfw_data.fw_session;
1071         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1072         srpc_test_reqst_t *request;
1073         int                rc;
1074         sfw_batch_t       *bat;
1075
1076         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1077         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1078
1079         if (request->tsr_loop == 0 ||
1080             request->tsr_concur == 0 ||
1081             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1082             request->tsr_ndest > SFW_MAX_NDESTS ||
1083             (request->tsr_is_client && request->tsr_ndest == 0) ||
1084             request->tsr_concur > SFW_MAX_CONCUR ||
1085             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1086             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1087                 reply->tsr_status = EINVAL;
1088                 return 0;
1089         }
1090
1091         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1092             sfw_find_test_case(request->tsr_service) == NULL) {
1093                 reply->tsr_status = ENOENT;
1094                 return 0;
1095         }
1096
1097         bat = sfw_bid2batch(request->tsr_bid);
1098         if (bat == NULL) {
1099                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1100                         rpc->srpc_service->sv_name,
1101                         libcfs_id2str(rpc->srpc_peer));
1102                 return -ENOMEM;
1103         }
1104
1105         if (sfw_batch_active(bat)) {
1106                 reply->tsr_status = EBUSY;
1107                 return 0;
1108         }
1109
1110         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1111                 /* rpc will be resumed later in sfw_bulk_ready */
1112                 return sfw_alloc_pages(rpc,
1113                                        sfw_id_pages(request->tsr_ndest), 1);
1114         }
1115
1116         rc = sfw_add_test_instance(bat, rpc);
1117         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1118                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1119                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1120                 request->tsr_is_client ? "client" : "server",
1121                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1122
1123         reply->tsr_status = (rc < 0) ? -rc : rc;
1124         return 0;
1125 }
1126
1127 int
1128 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1129 {
1130         sfw_session_t *sn = sfw_data.fw_session;
1131         int            rc = 0;
1132         sfw_batch_t   *bat;
1133
1134         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1135
1136         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1137                 reply->bar_status = ESRCH;
1138                 return 0;
1139         }
1140
1141         bat = sfw_find_batch(request->bar_bid);
1142         if (bat == NULL) {
1143                 reply->bar_status = ENOENT;
1144                 return 0;
1145         }
1146
1147         switch (request->bar_opc) {
1148         case SRPC_BATCH_OPC_RUN:
1149                 rc = sfw_run_batch(bat);
1150                 break;
1151
1152         case SRPC_BATCH_OPC_STOP:
1153                 rc = sfw_stop_batch(bat, request->bar_arg);
1154                 break;
1155
1156         case SRPC_BATCH_OPC_QUERY:
1157                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1158                 break;
1159
1160         default:
1161                 return -EINVAL; /* drop it */
1162         }
1163
1164         reply->bar_status = (rc < 0) ? -rc : rc;
1165         return 0;
1166 }
1167
1168 int
1169 sfw_handle_server_rpc (srpc_server_rpc_t *rpc)
1170 {
1171         srpc_service_t *sv = rpc->srpc_service;
1172         srpc_msg_t     *reply = &rpc->srpc_replymsg;
1173         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1174         int             rc = 0;
1175
1176         LASSERT (sfw_data.fw_active_srpc == NULL);
1177         LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1178
1179         spin_lock(&sfw_data.fw_lock);
1180
1181         if (sfw_data.fw_shuttingdown) {
1182                 spin_unlock(&sfw_data.fw_lock);
1183                 return -ESHUTDOWN;
1184         }
1185
1186         /* Remove timer to avoid racing with it or expiring active session */
1187         if (sfw_del_session_timer() != 0) {
1188                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1189                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1190                 spin_unlock(&sfw_data.fw_lock);
1191                 return -EAGAIN;
1192         }
1193
1194         sfw_data.fw_active_srpc = rpc;
1195         spin_unlock(&sfw_data.fw_lock);
1196
1197         sfw_unpack_message(request);
1198         LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1199
1200         switch(sv->sv_id) {
1201         default:
1202                 LBUG ();
1203         case SRPC_SERVICE_TEST:
1204                 rc = sfw_add_test(rpc);
1205                 break;
1206
1207         case SRPC_SERVICE_BATCH:
1208                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1209                                        &reply->msg_body.bat_reply);
1210                 break;
1211
1212         case SRPC_SERVICE_QUERY_STAT:
1213                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1214                                    &reply->msg_body.stat_reply);
1215                 break;
1216
1217         case SRPC_SERVICE_DEBUG:
1218                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1219                                        &reply->msg_body.dbg_reply);
1220                 break;
1221
1222         case SRPC_SERVICE_MAKE_SESSION:
1223                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1224                                       &reply->msg_body.mksn_reply);
1225                 break;
1226
1227         case SRPC_SERVICE_REMOVE_SESSION:
1228                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1229                                         &reply->msg_body.rmsn_reply);
1230                 break;
1231         }
1232
1233         rpc->srpc_done = sfw_server_rpc_done;
1234         spin_lock(&sfw_data.fw_lock);
1235
1236 #ifdef __KERNEL__
1237         if (!sfw_data.fw_shuttingdown)
1238                 sfw_add_session_timer();
1239 #else
1240         LASSERT (!sfw_data.fw_shuttingdown);
1241         sfw_add_session_timer();
1242 #endif
1243
1244         sfw_data.fw_active_srpc = NULL;
1245         spin_unlock(&sfw_data.fw_lock);
1246         return rc;
1247 }
1248
1249 int
1250 sfw_bulk_ready (srpc_server_rpc_t *rpc, int status)
1251 {
1252         srpc_service_t *sv = rpc->srpc_service;
1253         int             rc;
1254
1255         LASSERT (rpc->srpc_bulk != NULL);
1256         LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1257         LASSERT (sfw_data.fw_active_srpc == NULL);
1258         LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1259
1260         spin_lock(&sfw_data.fw_lock);
1261
1262         if (status != 0) {
1263                 CERROR ("Bulk transfer failed for RPC: "
1264                         "service %s, peer %s, status %d\n",
1265                         sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1266                 spin_unlock(&sfw_data.fw_lock);
1267                 return -EIO;
1268         }
1269
1270         if (sfw_data.fw_shuttingdown) {
1271                 spin_unlock(&sfw_data.fw_lock);
1272                 return -ESHUTDOWN;
1273         }
1274
1275         if (sfw_del_session_timer() != 0) {
1276                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1277                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1278                 spin_unlock(&sfw_data.fw_lock);
1279                 return -EAGAIN;
1280         }
1281
1282         sfw_data.fw_active_srpc = rpc;
1283         spin_unlock(&sfw_data.fw_lock);
1284
1285         rc = sfw_add_test(rpc);
1286
1287         spin_lock(&sfw_data.fw_lock);
1288
1289 #ifdef __KERNEL__
1290         if (!sfw_data.fw_shuttingdown)
1291                 sfw_add_session_timer();
1292 #else
1293         LASSERT (!sfw_data.fw_shuttingdown);
1294         sfw_add_session_timer();
1295 #endif
1296
1297         sfw_data.fw_active_srpc = NULL;
1298         spin_unlock(&sfw_data.fw_lock);
1299         return rc;
1300 }
1301
1302 srpc_client_rpc_t *
1303 sfw_create_rpc (lnet_process_id_t peer, int service,
1304                 int nbulkiov, int bulklen,
1305                 void (*done) (srpc_client_rpc_t *), void *priv)
1306 {
1307         srpc_client_rpc_t *rpc;
1308
1309         spin_lock(&sfw_data.fw_lock);
1310
1311         LASSERT (!sfw_data.fw_shuttingdown);
1312         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1313
1314         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1315                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1316                                  srpc_client_rpc_t, crpc_list);
1317                 list_del(&rpc->crpc_list);
1318                 spin_unlock(&sfw_data.fw_lock);
1319
1320                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1321                                      done, sfw_client_rpc_fini, priv);
1322                 return rpc;
1323         }
1324
1325         spin_unlock(&sfw_data.fw_lock);
1326
1327         rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1328                                      nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1329                                      priv);
1330         return rpc;
1331 }
1332
1333 void
1334 sfw_unpack_message (srpc_msg_t *msg)
1335 {
1336         if (msg->msg_magic == SRPC_MSG_MAGIC)
1337                 return; /* no flipping needed */
1338
1339         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1340
1341         __swab32s(&msg->msg_type);
1342
1343         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1344                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1345
1346                 __swab32s(&req->str_type);
1347                 __swab64s(&req->str_rpyid);
1348                 sfw_unpack_sid(req->str_sid);
1349                 return;
1350         }
1351
1352         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1353                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1354
1355                 __swab32s(&rep->str_status);
1356                 sfw_unpack_sid(rep->str_sid);
1357                 sfw_unpack_fw_counters(rep->str_fw);
1358                 sfw_unpack_rpc_counters(rep->str_rpc);
1359                 sfw_unpack_lnet_counters(rep->str_lnet);
1360                 return;
1361         }
1362
1363         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1364                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1365
1366                 __swab64s(&req->mksn_rpyid);
1367                 __swab32s(&req->mksn_force);
1368                 sfw_unpack_sid(req->mksn_sid);
1369                 return;
1370         }
1371
1372         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1373                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1374
1375                 __swab32s(&rep->mksn_status);
1376                 __swab32s(&rep->mksn_timeout);
1377                 sfw_unpack_sid(rep->mksn_sid);
1378                 return;
1379         }
1380
1381         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1382                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1383
1384                 __swab64s(&req->rmsn_rpyid);
1385                 sfw_unpack_sid(req->rmsn_sid);
1386                 return;
1387         }
1388
1389         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1390                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1391
1392                 __swab32s(&rep->rmsn_status);
1393                 sfw_unpack_sid(rep->rmsn_sid);
1394                 return;
1395         }
1396
1397         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1398                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1399
1400                 __swab64s(&req->dbg_rpyid);
1401                 __swab32s(&req->dbg_flags);
1402                 sfw_unpack_sid(req->dbg_sid);
1403                 return;
1404         }
1405
1406         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1407                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1408
1409                 __swab32s(&rep->dbg_nbatch);
1410                 __swab32s(&rep->dbg_timeout);
1411                 sfw_unpack_sid(rep->dbg_sid);
1412                 return;
1413         }
1414
1415         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1416                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1417
1418                 __swab32s(&req->bar_opc);
1419                 __swab64s(&req->bar_rpyid);
1420                 __swab32s(&req->bar_testidx);
1421                 __swab32s(&req->bar_arg);
1422                 sfw_unpack_sid(req->bar_sid);
1423                 __swab64s(&req->bar_bid.bat_id);
1424                 return;
1425         }
1426
1427         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1428                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1429
1430                 __swab32s(&rep->bar_status);
1431                 sfw_unpack_sid(rep->bar_sid);
1432                 return;
1433         }
1434
1435         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1436                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1437
1438                 __swab64s(&req->tsr_rpyid);
1439                 __swab64s(&req->tsr_bulkid);
1440                 __swab32s(&req->tsr_loop);
1441                 __swab32s(&req->tsr_ndest);
1442                 __swab32s(&req->tsr_concur);
1443                 __swab32s(&req->tsr_service);
1444                 sfw_unpack_sid(req->tsr_sid);
1445                 __swab64s(&req->tsr_bid.bat_id);
1446                 return;
1447         }
1448
1449         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1450                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1451
1452                 __swab32s(&rep->tsr_status);
1453                 sfw_unpack_sid(rep->tsr_sid);
1454                 return;
1455         }
1456
1457         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1458                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1459
1460                 __swab64s(&req->join_rpyid);
1461                 sfw_unpack_sid(req->join_sid);
1462                 return;
1463         }
1464
1465         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1466                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1467
1468                 __swab32s(&rep->join_status);
1469                 __swab32s(&rep->join_timeout);
1470                 sfw_unpack_sid(rep->join_sid);
1471                 return;
1472         }
1473
1474         LBUG ();
1475         return;
1476 }
1477
1478 void
1479 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1480 {
1481         LASSERT (atomic_read(&rpc->crpc_refcount) > 0);
1482         LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1483
1484         spin_lock(&rpc->crpc_lock);
1485         srpc_abort_rpc(rpc, -EINTR);
1486         spin_unlock(&rpc->crpc_lock);
1487         return;
1488 }
1489
1490 void
1491 sfw_post_rpc (srpc_client_rpc_t *rpc)
1492 {
1493         spin_lock(&rpc->crpc_lock);
1494
1495         LASSERT (!rpc->crpc_closed);
1496         LASSERT (!rpc->crpc_aborted);
1497         LASSERT (list_empty(&rpc->crpc_list));
1498         LASSERT (!sfw_data.fw_shuttingdown);
1499
1500         rpc->crpc_timeout = SFW_CLIENT_RPC_TIMEOUT;
1501         srpc_post_rpc(rpc);
1502
1503         spin_unlock(&rpc->crpc_lock);
1504         return;
1505 }
1506
1507 static srpc_service_t sfw_services[] =
1508 {
1509         {
1510                 /* sv_id */    SRPC_SERVICE_DEBUG,
1511                 /* sv_name */  "debug",
1512                 0
1513         },
1514         {
1515                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1516                 /* sv_name */  "query stats",
1517                 0
1518         },
1519         {
1520                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1521                 /* sv_name */  "make session",
1522                 0
1523         },
1524         {
1525                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1526                 /* sv_name */  "remove session",
1527                 0
1528         },
1529         {
1530                 /* sv_id */    SRPC_SERVICE_BATCH,
1531                 /* sv_name */  "batch service",
1532                 0
1533         },
1534         {
1535                 /* sv_id */    SRPC_SERVICE_TEST,
1536                 /* sv_name */  "test service",
1537                 0
1538         },
1539         {
1540                 /* sv_id */    0,
1541                 /* sv_name */  NULL,
1542                 0
1543         }
1544 };
1545
1546 extern sfw_test_client_ops_t ping_test_client;
1547 extern srpc_service_t        ping_test_service;
1548 extern void ping_init_test_client(void);
1549 extern void ping_init_test_service(void);
1550
1551 extern sfw_test_client_ops_t brw_test_client;
1552 extern srpc_service_t        brw_test_service;
1553 extern void brw_init_test_client(void);
1554 extern void brw_init_test_service(void);
1555
1556
1557 int
1558 sfw_startup (void)
1559 {
1560         int              i;
1561         int              rc;
1562         int              error;
1563         srpc_service_t  *sv;
1564         sfw_test_case_t *tsc;
1565
1566 #ifndef __KERNEL__
1567         char *s;
1568
1569         s = getenv("SESSION_TIMEOUT");
1570         session_timeout = s != NULL ? atoi(s) : session_timeout;
1571
1572         s = getenv("BRW_INJECT_ERRORS");
1573         brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
1574 #endif
1575
1576         if (session_timeout < 0) {
1577                 CERROR ("Session timeout must be non-negative: %d\n",
1578                         session_timeout);
1579                 return -EINVAL;
1580         }
1581
1582         if (session_timeout == 0)
1583                 CWARN ("Zero session_timeout specified "
1584                        "- test sessions never expire.\n");
1585
1586         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1587
1588         sfw_data.fw_session     = NULL;
1589         sfw_data.fw_active_srpc = NULL;
1590         spin_lock_init(&sfw_data.fw_lock);
1591         atomic_set(&sfw_data.fw_nzombies, 0);
1592         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1593         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1594         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1595
1596         brw_init_test_client();
1597         brw_init_test_service();
1598         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1599         LASSERT (rc == 0);
1600
1601         ping_init_test_client();
1602         ping_init_test_service();
1603         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1604         LASSERT (rc == 0);
1605
1606         error = 0;
1607         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1608                                        sfw_test_case_t, tsc_list) {
1609                 sv = tsc->tsc_srv_service;
1610                 sv->sv_concur = SFW_TEST_CONCURRENCY;
1611
1612                 rc = srpc_add_service(sv);
1613                 LASSERT (rc != -EBUSY);
1614                 if (rc != 0) {
1615                         CWARN ("Failed to add %s service: %d\n",
1616                                sv->sv_name, rc);
1617                         error = rc;
1618                 }
1619         }
1620
1621         for (i = 0; ; i++) {
1622                 sv = &sfw_services[i];
1623                 if (sv->sv_name == NULL) break;
1624
1625                 sv->sv_bulk_ready = NULL;
1626                 sv->sv_handler    = sfw_handle_server_rpc;
1627                 sv->sv_concur     = SFW_SERVICE_CONCURRENCY;
1628                 if (sv->sv_id == SRPC_SERVICE_TEST)
1629                         sv->sv_bulk_ready = sfw_bulk_ready;
1630
1631                 rc = srpc_add_service(sv);
1632                 LASSERT (rc != -EBUSY);
1633                 if (rc != 0) {
1634                         CWARN ("Failed to add %s service: %d\n",
1635                                sv->sv_name, rc);
1636                         error = rc;
1637                 }
1638
1639                 /* about to sfw_shutdown, no need to add buffer */
1640                 if (error) continue;
1641
1642                 rc = srpc_service_add_buffers(sv, SFW_POST_BUFFERS);
1643                 if (rc != SFW_POST_BUFFERS) {
1644                         CWARN ("Failed to reserve enough buffers: "
1645                                "service %s, %d needed, %d reserved\n",
1646                                sv->sv_name, SFW_POST_BUFFERS, rc);
1647                         error = -ENOMEM;
1648                 }
1649         }
1650
1651         if (error != 0)
1652                 sfw_shutdown();
1653         return error;
1654 }
1655
1656 void
1657 sfw_shutdown (void)
1658 {
1659         srpc_service_t  *sv;
1660         sfw_test_case_t *tsc;
1661         int              i;
1662
1663         spin_lock(&sfw_data.fw_lock);
1664
1665         sfw_data.fw_shuttingdown = 1;
1666 #ifdef __KERNEL__
1667         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1668                        "waiting for active RPC to finish.\n");
1669 #else
1670         LASSERT (sfw_data.fw_active_srpc == NULL);
1671 #endif
1672
1673         if (sfw_del_session_timer() != 0)
1674                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1675                                "waiting for session timer to explode.\n");
1676
1677         sfw_deactivate_session();
1678         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1679                        sfw_data.fw_lock,
1680                        "waiting for %d zombie sessions to die.\n",
1681                        atomic_read(&sfw_data.fw_nzombies));
1682
1683         spin_unlock(&sfw_data.fw_lock);
1684
1685         for (i = 0; ; i++) {
1686                 sv = &sfw_services[i];
1687                 if (sv->sv_name == NULL)
1688                         break;
1689
1690                 srpc_shutdown_service(sv);
1691                 srpc_remove_service(sv);
1692         }
1693
1694         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1695                                        sfw_test_case_t, tsc_list) {
1696                 sv = tsc->tsc_srv_service;
1697                 srpc_shutdown_service(sv);
1698                 srpc_remove_service(sv);
1699         }
1700
1701         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1702                 srpc_client_rpc_t *rpc;
1703
1704                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1705                                  srpc_client_rpc_t, crpc_list);
1706                 list_del(&rpc->crpc_list);
1707
1708                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1709         }
1710
1711         for (i = 0; ; i++) {
1712                 sv = &sfw_services[i];
1713                 if (sv->sv_name == NULL)
1714                         break;
1715
1716                 srpc_wait_service_shutdown(sv);
1717         }
1718
1719         while (!list_empty(&sfw_data.fw_tests)) {
1720                 tsc = list_entry(sfw_data.fw_tests.next,
1721                                  sfw_test_case_t, tsc_list);
1722
1723                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1724
1725                 list_del(&tsc->tsc_list);
1726                 LIBCFS_FREE(tsc, sizeof(*tsc));
1727         }
1728
1729         return;
1730 }