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 [sun.com URL with a
  20  * copy of GPLv2].
  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 #define SFW_TEST_CONCURRENCY     128
  55 #define SFW_TEST_RPC_TIMEOUT     64
  56 #define SFW_CLIENT_RPC_TIMEOUT   64  /* in seconds */
  57 #define SFW_EXTRA_TEST_BUFFERS   8 /* tolerate buggy peers with extra buffers */
  58
  59 #define sfw_test_buffers(tsi)    ((tsi)->tsi_loop + SFW_EXTRA_TEST_BUFFERS)
  60
  61 #define sfw_unpack_id(id)               \
  62 do {                                    \
  63         __swab64s(&(id).nid);           \
  64         __swab32s(&(id).pid);           \
  65 } while (0)
  66
  67 #define sfw_unpack_sid(sid)             \
  68 do {                                    \
  69         __swab64s(&(sid).ses_nid);      \
  70         __swab64s(&(sid).ses_stamp);    \
  71 } while (0)
  72
  73 #define sfw_unpack_fw_counters(fc)        \
  74 do {                                      \
  75         __swab32s(&(fc).brw_errors);      \
  76         __swab32s(&(fc).ping_errors);     \
  77         __swab32s(&(fc).active_tests);    \
  78         __swab32s(&(fc).active_batches);  \
  79         __swab32s(&(fc).zombie_sessions); \
  80 } while (0)
  81
  82 #define sfw_unpack_rpc_counters(rc)     \
  83 do {                                    \
  84         __swab32s(&(rc).errors);        \
  85         __swab32s(&(rc).rpcs_sent);     \
  86         __swab32s(&(rc).rpcs_rcvd);     \
  87         __swab32s(&(rc).rpcs_dropped);  \
  88         __swab32s(&(rc).rpcs_expired);  \
  89         __swab64s(&(rc).bulk_get);      \
  90         __swab64s(&(rc).bulk_put);      \
  91 } while (0)
  92
  93 #define sfw_unpack_lnet_counters(lc)    \
  94 do {                                    \
  95         __swab32s(&(lc).errors);        \
  96         __swab32s(&(lc).msgs_max);      \
  97         __swab32s(&(lc).msgs_alloc);    \
  98         __swab32s(&(lc).send_count);    \
  99         __swab32s(&(lc).recv_count);    \
 100         __swab32s(&(lc).drop_count);    \
 101         __swab32s(&(lc).route_count);   \
 102         __swab64s(&(lc).send_length);   \
 103         __swab64s(&(lc).recv_length);   \
 104         __swab64s(&(lc).drop_length);   \
 105         __swab64s(&(lc).route_length);  \
 106 } while (0)
 107
 108 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 109 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 110
 111 struct smoketest_framework {
 112         struct list_head   fw_zombie_rpcs;     /* RPCs to be recycled */
 113         struct list_head   fw_zombie_sessions; /* stopping sessions */
 114         struct list_head   fw_tests;           /* registered test cases */
 115         atomic_t           fw_nzombies;        /* # zombie sessions */
 116         spinlock_t         fw_lock;            /* serialise */
 117         sfw_session_t     *fw_session;         /* _the_ session */
 118         int                fw_shuttingdown;    /* shutdown in progress */
 119         srpc_server_rpc_t *fw_active_srpc;     /* running RPC */
 120 } sfw_data;
 121
 122 /* forward ref's */
 123 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 124 void sfw_destroy_session (sfw_session_t *sn);
 125
 126 static inline sfw_test_case_t *
 127 sfw_find_test_case(int id)
 128 {
 129         sfw_test_case_t *tsc;
 130
 131         LASSERT (id <= SRPC_SERVICE_MAX_ID);
 132         LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 133
 134         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
 135                 if (tsc->tsc_srv_service->sv_id == id)
 136                         return tsc;
 137         }
 138
 139         return NULL;
 140 }
 141
 142 static int
 143 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 144 {
 145         sfw_test_case_t *tsc;
 146
 147         if (sfw_find_test_case(service->sv_id) != NULL) {
 148                 CERROR ("Failed to register test %s (%d)\n",
 149                         service->sv_name, service->sv_id);
 150                 return -EEXIST;
 151         }
 152
 153         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 154         if (tsc == NULL)
 155                 return -ENOMEM;
 156
 157         memset(tsc, 0, sizeof(sfw_test_case_t));
 158         tsc->tsc_cli_ops     = cliops;
 159         tsc->tsc_srv_service = service;
 160
 161         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 162         return 0;
 163 }
 164
 165 void
 166 sfw_add_session_timer (void)
 167 {
 168         sfw_session_t *sn = sfw_data.fw_session;
 169         stt_timer_t   *timer = &sn->sn_timer;
 170
 171         LASSERT (!sfw_data.fw_shuttingdown);
 172
 173         if (sn == NULL || sn->sn_timeout == 0)
 174                 return;
 175
 176         LASSERT (!sn->sn_timer_active);
 177
 178         sn->sn_timer_active = 1;
 179         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 180                                           cfs_time_current_sec());
 181         stt_add_timer(timer);
 182         return;
 183 }
 184
 185 int
 186 sfw_del_session_timer (void)
 187 {
 188         sfw_session_t *sn = sfw_data.fw_session;
 189
 190         if (sn == NULL || !sn->sn_timer_active)
 191                 return 0;
 192
 193         LASSERT (sn->sn_timeout != 0);
 194
 195         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 196                 sn->sn_timer_active = 0;
 197                 return 0;
 198         }
 199
 200 #ifndef __KERNEL__
 201         /* Racing is impossible in single-threaded userland selftest */
 202         LBUG();
 203 #endif
 204         return EBUSY; /* racing with sfw_session_expired() */
 205 }
 206
 207 /* called with sfw_data.fw_lock held */
 208 static void
 209 sfw_deactivate_session (void)
 210 {
 211         sfw_session_t *sn = sfw_data.fw_session;
 212         int            nactive = 0;
 213         sfw_batch_t   *tsb;
 214
 215         if (sn == NULL) return;
 216
 217         LASSERT (!sn->sn_timer_active);
 218
 219         sfw_data.fw_session = NULL;
 220         atomic_inc(&sfw_data.fw_nzombies);
 221         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 222
 223         list_for_each_entry (tsb, &sn->sn_batches, bat_list) {
 224                 if (sfw_batch_active(tsb)) {
 225                         nactive++;
 226                         sfw_stop_batch(tsb, 1);
 227                 }
 228         }
 229
 230         if (nactive != 0)
 231                 return;   /* wait for active batches to stop */
 232
 233         list_del_init(&sn->sn_list);
 234         spin_unlock(&sfw_data.fw_lock);
 235
 236         sfw_destroy_session(sn);
 237
 238         spin_lock(&sfw_data.fw_lock);
 239         return;
 240 }
 241
 242 #ifndef __KERNEL__
 243
 244 int
 245 sfw_session_removed (void)
 246 {
 247         return (sfw_data.fw_session == NULL) ? 1 : 0;
 248 }
 249
 250 #endif
 251
 252 void
 253 sfw_session_expired (void *data)
 254 {
 255         sfw_session_t *sn = data;
 256
 257         spin_lock(&sfw_data.fw_lock);
 258
 259         LASSERT (sn->sn_timer_active);
 260         LASSERT (sn == sfw_data.fw_session);
 261
 262         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 263                libcfs_nid2str(sn->sn_id.ses_nid),
 264                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 265
 266         sn->sn_timer_active = 0;
 267         sfw_deactivate_session();
 268
 269         spin_unlock(&sfw_data.fw_lock);
 270         return;
 271 }
 272
 273 static inline void
 274 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
 275 {
 276         stt_timer_t *timer = &sn->sn_timer;
 277
 278         memset(sn, 0, sizeof(sfw_session_t));
 279         CFS_INIT_LIST_HEAD(&sn->sn_list);
 280         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 281         atomic_set(&sn->sn_brw_errors, 0);
 282         atomic_set(&sn->sn_ping_errors, 0);
 283         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 284
 285         sn->sn_timer_active = 0;
 286         sn->sn_id           = sid;
 287         sn->sn_timeout      = session_timeout;
 288
 289         timer->stt_data = sn;
 290         timer->stt_func = sfw_session_expired;
 291         CFS_INIT_LIST_HEAD(&timer->stt_list);
 292 }
 293
 294 /* completion handler for incoming framework RPCs */
 295 void
 296 sfw_server_rpc_done (srpc_server_rpc_t *rpc)
 297 {
 298         srpc_service_t *sv = rpc->srpc_service;
 299         int             status = rpc->srpc_status;
 300
 301         CDEBUG (D_NET,
 302                 "Incoming framework RPC done: "
 303                 "service %s, peer %s, status %s:%d\n",
 304                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 305                 swi_state2str(rpc->srpc_wi.wi_state),
 306                 status);
 307
 308         if (rpc->srpc_bulk != NULL)
 309                 sfw_free_pages(rpc);
 310         return;
 311 }
 312
 313 void
 314 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 315 {
 316         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 317         LASSERT (list_empty(&rpc->crpc_list));
 318         LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
 319 #ifndef __KERNEL__
 320         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 321 #endif
 322
 323         CDEBUG (D_NET,
 324                 "Outgoing framework RPC done: "
 325                 "service %d, peer %s, status %s:%d:%d\n",
 326                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 327                 swi_state2str(rpc->crpc_wi.wi_state),
 328                 rpc->crpc_aborted, rpc->crpc_status);
 329
 330         spin_lock(&sfw_data.fw_lock);
 331
 332         /* my callers must finish all RPCs before shutting me down */
 333         LASSERT (!sfw_data.fw_shuttingdown);
 334         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 335
 336         spin_unlock(&sfw_data.fw_lock);
 337         return;
 338 }
 339
 340 sfw_batch_t *
 341 sfw_find_batch (lst_bid_t bid)
 342 {
 343         sfw_session_t *sn = sfw_data.fw_session;
 344         sfw_batch_t   *bat;
 345
 346         LASSERT (sn != NULL);
 347
 348         list_for_each_entry (bat, &sn->sn_batches, bat_list) {
 349                 if (bat->bat_id.bat_id == bid.bat_id)
 350                         return bat;
 351         }
 352
 353         return NULL;
 354 }
 355
 356 sfw_batch_t *
 357 sfw_bid2batch (lst_bid_t bid)
 358 {
 359         sfw_session_t *sn = sfw_data.fw_session;
 360         sfw_batch_t   *bat;
 361
 362         LASSERT (sn != NULL);
 363
 364         bat = sfw_find_batch(bid);
 365         if (bat != NULL)
 366                 return bat;
 367
 368         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 369         if (bat == NULL)
 370                 return NULL;
 371
 372         bat->bat_error    = 0;
 373         bat->bat_session  = sn;
 374         bat->bat_id       = bid;
 375         atomic_set(&bat->bat_nactive, 0);
 376         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 377
 378         list_add_tail(&bat->bat_list, &sn->sn_batches);
 379         return bat;
 380 }
 381
 382 int
 383 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 384 {
 385         sfw_session_t  *sn = sfw_data.fw_session;
 386         sfw_counters_t *cnt = &reply->str_fw;
 387         sfw_batch_t    *bat;
 388
 389         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 390
 391         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 392                 reply->str_status = EINVAL;
 393                 return 0;
 394         }
 395
 396         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 397                 reply->str_status = ESRCH;
 398                 return 0;
 399         }
 400
 401         LNET_LOCK();
 402         reply->str_lnet = the_lnet.ln_counters;
 403         LNET_UNLOCK();
 404
 405         srpc_get_counters(&reply->str_rpc);
 406
 407         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 408         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 409         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 410
 411         cnt->active_tests = cnt->active_batches = 0;
 412         list_for_each_entry (bat, &sn->sn_batches, bat_list) {
 413                 int n = atomic_read(&bat->bat_nactive);
 414
 415                 if (n > 0) {
 416                         cnt->active_batches++;
 417                         cnt->active_tests += n;
 418                 }
 419         }
 420
 421         reply->str_status = 0;
 422         return 0;
 423 }
 424
 425 int
 426 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 427 {
 428         sfw_session_t *sn = sfw_data.fw_session;
 429
 430         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 431                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 432                 reply->mksn_status = EINVAL;
 433                 return 0;
 434         }
 435
 436         if (sn != NULL && !request->mksn_force) {
 437                 reply->mksn_sid    = sn->sn_id;
 438                 reply->mksn_status = EBUSY;
 439                 strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 440                 return 0;
 441         }
 442
 443         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 444         if (sn == NULL) {
 445                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 446                 return -ENOMEM;
 447         }
 448
 449         sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
 450
 451         spin_lock(&sfw_data.fw_lock);
 452
 453         sfw_deactivate_session();
 454         LASSERT (sfw_data.fw_session == NULL);
 455         sfw_data.fw_session = sn;
 456
 457         spin_unlock(&sfw_data.fw_lock);
 458
 459         reply->mksn_status  = 0;
 460         reply->mksn_sid     = sn->sn_id;
 461         reply->mksn_timeout = sn->sn_timeout;
 462         return 0;
 463 }
 464
 465 int
 466 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 467 {
 468         sfw_session_t *sn = sfw_data.fw_session;
 469
 470         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 471
 472         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 473                 reply->rmsn_status = EINVAL;
 474                 return 0;
 475         }
 476
 477         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 478                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 479                 return 0;
 480         }
 481
 482         spin_lock(&sfw_data.fw_lock);
 483         sfw_deactivate_session();
 484         spin_unlock(&sfw_data.fw_lock);
 485
 486         reply->rmsn_status = 0;
 487         reply->rmsn_sid    = LST_INVALID_SID;
 488         LASSERT (sfw_data.fw_session == NULL);
 489         return 0;
 490 }
 491
 492 int
 493 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 494 {
 495         sfw_session_t *sn = sfw_data.fw_session;
 496
 497         if (sn == NULL) {
 498                 reply->dbg_status = ESRCH;
 499                 reply->dbg_sid    = LST_INVALID_SID;
 500                 return 0;
 501         }
 502
 503         reply->dbg_status  = 0;
 504         reply->dbg_sid     = sn->sn_id;
 505         reply->dbg_timeout = sn->sn_timeout;
 506         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 507
 508         return 0;
 509 }
 510
 511 void
 512 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 513 {
 514         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 515         sfw_test_instance_t *tsi = tsu->tsu_instance;
 516
 517         /* Called with hold of tsi->tsi_lock */
 518         LASSERT (list_empty(&rpc->crpc_list));
 519         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 520 }
 521
 522 int
 523 sfw_load_test (sfw_test_instance_t *tsi)
 524 {
 525         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 526         int              nrequired = sfw_test_buffers(tsi);
 527         int              nposted;
 528
 529         LASSERT (tsc != NULL);
 530
 531         if (tsi->tsi_is_client) {
 532                 tsi->tsi_ops = tsc->tsc_cli_ops;
 533                 return 0;
 534         }
 535
 536         nposted = srpc_service_add_buffers(tsc->tsc_srv_service, nrequired);
 537         if (nposted != nrequired) {
 538                 CWARN ("Failed to reserve enough buffers: "
 539                        "service %s, %d needed, %d reserved\n",
 540                        tsc->tsc_srv_service->sv_name, nrequired, nposted);
 541                 srpc_service_remove_buffers(tsc->tsc_srv_service, nposted);
 542                 return -ENOMEM;
 543         }
 544
 545         CDEBUG (D_NET, "Reserved %d buffers for test %s\n",
 546                 nposted, tsc->tsc_srv_service->sv_name);
 547         return 0;
 548 }
 549
 550 void
 551 sfw_unload_test (sfw_test_instance_t *tsi)
 552 {
 553         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 554
 555         LASSERT (tsc != NULL);
 556
 557         if (!tsi->tsi_is_client)
 558                 srpc_service_remove_buffers(tsc->tsc_srv_service,
 559                                             sfw_test_buffers(tsi));
 560         return;
 561 }
 562
 563 void
 564 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 565 {
 566         srpc_client_rpc_t *rpc;
 567         sfw_test_unit_t   *tsu;
 568
 569         if (!tsi->tsi_is_client) goto clean;
 570
 571         tsi->tsi_ops->tso_fini(tsi);
 572
 573         LASSERT (!tsi->tsi_stopping);
 574         LASSERT (list_empty(&tsi->tsi_active_rpcs));
 575         LASSERT (!sfw_test_active(tsi));
 576
 577         while (!list_empty(&tsi->tsi_units)) {
 578                 tsu = list_entry(tsi->tsi_units.next,
 579                                  sfw_test_unit_t, tsu_list);
 580                 list_del(&tsu->tsu_list);
 581                 LIBCFS_FREE(tsu, sizeof(*tsu));
 582         }
 583
 584         while (!list_empty(&tsi->tsi_free_rpcs)) {
 585                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 586                                  srpc_client_rpc_t, crpc_list);
 587                 list_del(&rpc->crpc_list);
 588                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 589         }
 590
 591 clean:
 592         sfw_unload_test(tsi);
 593         LIBCFS_FREE(tsi, sizeof(*tsi));
 594         return;
 595 }
 596
 597 void
 598 sfw_destroy_batch (sfw_batch_t *tsb)
 599 {
 600         sfw_test_instance_t *tsi;
 601
 602         LASSERT (!sfw_batch_active(tsb));
 603         LASSERT (list_empty(&tsb->bat_list));
 604
 605         while (!list_empty(&tsb->bat_tests)) {
 606                 tsi = list_entry(tsb->bat_tests.next,
 607                                  sfw_test_instance_t, tsi_list);
 608                 list_del_init(&tsi->tsi_list);
 609                 sfw_destroy_test_instance(tsi);
 610         }
 611
 612         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 613         return;
 614 }
 615
 616 void
 617 sfw_destroy_session (sfw_session_t *sn)
 618 {
 619         sfw_batch_t *batch;
 620
 621         LASSERT (list_empty(&sn->sn_list));
 622         LASSERT (sn != sfw_data.fw_session);
 623
 624         while (!list_empty(&sn->sn_batches)) {
 625                 batch = list_entry(sn->sn_batches.next,
 626                                    sfw_batch_t, bat_list);
 627                 list_del_init(&batch->bat_list);
 628                 sfw_destroy_batch(batch);
 629         }
 630
 631         LIBCFS_FREE(sn, sizeof(*sn));
 632         atomic_dec(&sfw_data.fw_nzombies);
 633         return;
 634 }
 635
 636 void
 637 sfw_unpack_test_req (srpc_msg_t *msg)
 638 {
 639         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 640
 641         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 642         LASSERT (req->tsr_is_client);
 643
 644         if (msg->msg_magic == SRPC_MSG_MAGIC)
 645                 return; /* no flipping needed */
 646
 647         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 648
 649         if (req->tsr_service == SRPC_SERVICE_BRW) {
 650                 test_bulk_req_t *bulk = &req->tsr_u.bulk;
 651
 652                 __swab32s(&bulk->blk_opc);
 653                 __swab32s(&bulk->blk_npg);
 654                 __swab32s(&bulk->blk_flags);
 655                 return;
 656         }
 657
 658         if (req->tsr_service == SRPC_SERVICE_PING) {
 659                 test_ping_req_t *ping = &req->tsr_u.ping;
 660
 661                 __swab32s(&ping->png_size);
 662                 __swab32s(&ping->png_flags);
 663                 return;
 664         }
 665
 666         LBUG ();
 667         return;
 668 }
 669
 670 int
 671 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 672 {
 673         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 674         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 675         srpc_bulk_t         *bk = rpc->srpc_bulk;
 676         int                  ndest = req->tsr_ndest;
 677         sfw_test_unit_t     *tsu;
 678         sfw_test_instance_t *tsi;
 679         int                  i;
 680         int                  rc;
 681
 682         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 683         if (tsi == NULL) {
 684                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 685                         tsb->bat_id.bat_id);
 686                 return -ENOMEM;
 687         }
 688
 689         memset(tsi, 0, sizeof(*tsi));
 690         spin_lock_init(&tsi->tsi_lock);
 691         atomic_set(&tsi->tsi_nactive, 0);
 692         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 693         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 694         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 695
 696         tsi->tsi_stopping      = 0;
 697         tsi->tsi_batch         = tsb;
 698         tsi->tsi_loop          = req->tsr_loop;
 699         tsi->tsi_concur        = req->tsr_concur;
 700         tsi->tsi_service       = req->tsr_service;
 701         tsi->tsi_is_client     = !!(req->tsr_is_client);
 702         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 703
 704         rc = sfw_load_test(tsi);
 705         if (rc != 0) {
 706                 LIBCFS_FREE(tsi, sizeof(*tsi));
 707                 return rc;
 708         }
 709
 710         LASSERT (!sfw_batch_active(tsb));
 711
 712         if (!tsi->tsi_is_client) {
 713                 /* it's test server, just add it to tsb */
 714                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 715                 return 0;
 716         }
 717
 718         LASSERT (bk != NULL);
 719 #ifndef __KERNEL__
 720         LASSERT (bk->bk_pages != NULL);
 721 #endif
 722         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= ndest);
 723         LASSERT (bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
 724
 725         sfw_unpack_test_req(msg);
 726         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 727
 728         for (i = 0; i < ndest; i++) {
 729                 lnet_process_id_t *dests;
 730                 lnet_process_id_t  id;
 731                 int                j;
 732
 733 #ifdef __KERNEL__
 734                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 735                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 736 #else
 737                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 738 #endif
 739                 id = dests[i % SFW_ID_PER_PAGE];
 740                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 741                         sfw_unpack_id(id);
 742
 743                 for (j = 0; j < tsi->tsi_concur; j++) {
 744                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 745                         if (tsu == NULL) {
 746                                 rc = -ENOMEM;
 747                                 CERROR ("Can't allocate tsu for %d\n",
 748                                         tsi->tsi_service);
 749                                 goto error;
 750                         }
 751
 752                         tsu->tsu_dest     = id;
 753                         tsu->tsu_instance = tsi;
 754                         tsu->tsu_private  = NULL;
 755                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 756                 }
 757         }
 758
 759         rc = tsi->tsi_ops->tso_init(tsi);
 760         if (rc == 0) {
 761                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 762                 return 0;
 763         }
 764
 765 error:
 766         LASSERT (rc != 0);
 767         sfw_destroy_test_instance(tsi);
 768         return rc;
 769 }
 770
 771 static void
 772 sfw_test_unit_done (sfw_test_unit_t *tsu)
 773 {
 774         sfw_test_instance_t *tsi = tsu->tsu_instance;
 775         sfw_batch_t         *tsb = tsi->tsi_batch;
 776         sfw_session_t       *sn = tsb->bat_session;
 777
 778         LASSERT (sfw_test_active(tsi));
 779
 780         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 781                 return;
 782
 783         /* the test instance is done */
 784         spin_lock(&tsi->tsi_lock);
 785
 786         tsi->tsi_stopping = 0;
 787
 788         spin_unlock(&tsi->tsi_lock);
 789
 790         spin_lock(&sfw_data.fw_lock);
 791
 792         if (!atomic_dec_and_test(&tsb->bat_nactive) || /* tsb still active */
 793             sn == sfw_data.fw_session) {               /* sn also active */
 794                 spin_unlock(&sfw_data.fw_lock);
 795                 return;
 796         }
 797
 798         LASSERT (!list_empty(&sn->sn_list)); /* I'm a zombie! */
 799
 800         list_for_each_entry (tsb, &sn->sn_batches, bat_list) {
 801                 if (sfw_batch_active(tsb)) {
 802                         spin_unlock(&sfw_data.fw_lock);
 803                         return;
 804                 }
 805         }
 806
 807         list_del_init(&sn->sn_list);
 808         spin_unlock(&sfw_data.fw_lock);
 809
 810         sfw_destroy_session(sn);
 811         return;
 812 }
 813
 814 void
 815 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 816 {
 817         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 818         sfw_test_instance_t *tsi = tsu->tsu_instance;
 819         int                  done = 0;
 820
 821         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 822
 823         spin_lock(&tsi->tsi_lock);
 824
 825         LASSERT (sfw_test_active(tsi));
 826         LASSERT (!list_empty(&rpc->crpc_list));
 827
 828         list_del_init(&rpc->crpc_list);
 829
 830         /* batch is stopping or loop is done or get error */
 831         if (tsi->tsi_stopping ||
 832             tsu->tsu_loop == 0 ||
 833             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 834                 done = 1;
 835
 836         /* dec ref for poster */
 837         srpc_client_rpc_decref(rpc);
 838
 839         spin_unlock(&tsi->tsi_lock);
 840
 841         if (!done) {
 842                 swi_schedule_workitem(&tsu->tsu_worker);
 843                 return;
 844         }
 845
 846         sfw_test_unit_done(tsu);
 847         return;
 848 }
 849
 850 int
 851 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
 852                      int nblk, int blklen, srpc_client_rpc_t **rpcpp)
 853 {
 854         srpc_client_rpc_t   *rpc = NULL;
 855         sfw_test_instance_t *tsi = tsu->tsu_instance;
 856
 857         spin_lock(&tsi->tsi_lock);
 858
 859         LASSERT (sfw_test_active(tsi));
 860
 861         if (!list_empty(&tsi->tsi_free_rpcs)) {
 862                 /* pick request from buffer */
 863                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 864                                  srpc_client_rpc_t, crpc_list);
 865                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 866                 list_del_init(&rpc->crpc_list);
 867
 868                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 869                                      blklen, sfw_test_rpc_done,
 870                                      sfw_test_rpc_fini, tsu);
 871         }
 872
 873         spin_unlock(&tsi->tsi_lock);
 874
 875         if (rpc == NULL)
 876                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 877                                              blklen, sfw_test_rpc_done,
 878                                              sfw_test_rpc_fini, tsu);
 879         if (rpc == NULL) {
 880                 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
 881                 return -ENOMEM;
 882         }
 883
 884         *rpcpp = rpc;
 885         return 0;
 886 }
 887
 888 int
 889 sfw_run_test (swi_workitem_t *wi)
 890 {
 891         sfw_test_unit_t     *tsu = wi->wi_data;
 892         sfw_test_instance_t *tsi = tsu->tsu_instance;
 893         srpc_client_rpc_t   *rpc = NULL;
 894
 895         LASSERT (wi == &tsu->tsu_worker);
 896
 897         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 898                 LASSERT (rpc == NULL);
 899                 goto test_done;
 900         }
 901
 902         LASSERT (rpc != NULL);
 903
 904         spin_lock(&tsi->tsi_lock);
 905
 906         if (tsi->tsi_stopping) {
 907                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 908                 spin_unlock(&tsi->tsi_lock);
 909                 goto test_done;
 910         }
 911
 912         if (tsu->tsu_loop > 0)
 913                 tsu->tsu_loop--;
 914
 915         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 916         spin_unlock(&tsi->tsi_lock);
 917
 918         rpc->crpc_timeout = SFW_TEST_RPC_TIMEOUT;
 919
 920         spin_lock(&rpc->crpc_lock);
 921         srpc_post_rpc(rpc);
 922         spin_unlock(&rpc->crpc_lock);
 923         return 0;
 924
 925 test_done:
 926         /*
 927          * No one can schedule me now since:
 928          * - previous RPC, if any, has done and
 929          * - no new RPC is initiated.
 930          * - my batch is still active; no one can run it again now.
 931          * Cancel pending schedules and prevent future schedule attempts:
 932          */
 933         swi_kill_workitem(wi);
 934         sfw_test_unit_done(tsu);
 935         return 1;
 936 }
 937
 938 int
 939 sfw_run_batch (sfw_batch_t *tsb)
 940 {
 941         swi_workitem_t      *wi;
 942         sfw_test_unit_t     *tsu;
 943         sfw_test_instance_t *tsi;
 944
 945         if (sfw_batch_active(tsb)) {
 946                 CDEBUG (D_NET, "Can't start active batch: "LPU64" (%d)\n",
 947                         tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
 948                 return -EPERM;
 949         }
 950
 951         list_for_each_entry (tsi, &tsb->bat_tests, tsi_list) {
 952                 if (!tsi->tsi_is_client) /* skip server instances */
 953                         continue;
 954
 955                 LASSERT (!tsi->tsi_stopping);
 956                 LASSERT (!sfw_test_active(tsi));
 957
 958                 atomic_inc(&tsb->bat_nactive);
 959
 960                 list_for_each_entry (tsu, &tsi->tsi_units, tsu_list) {
 961                         atomic_inc(&tsi->tsi_nactive);
 962                         tsu->tsu_loop = tsi->tsi_loop;
 963                         wi = &tsu->tsu_worker;
 964                         swi_init_workitem(wi, tsu, sfw_run_test);
 965                         swi_schedule_workitem(wi);
 966                 }
 967         }
 968
 969         return 0;
 970 }
 971
 972 int
 973 sfw_stop_batch (sfw_batch_t *tsb, int force)
 974 {
 975         sfw_test_instance_t *tsi;
 976         srpc_client_rpc_t   *rpc;
 977
 978         if (!sfw_batch_active(tsb))
 979                 return -EPERM;
 980
 981         list_for_each_entry (tsi, &tsb->bat_tests, tsi_list) {
 982                 spin_lock(&tsi->tsi_lock);
 983
 984                 if (!tsi->tsi_is_client ||
 985                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
 986                         spin_unlock(&tsi->tsi_lock);
 987                         continue;
 988                 }
 989
 990                 tsi->tsi_stopping = 1;
 991
 992                 if (!force) {
 993                         spin_unlock(&tsi->tsi_lock);
 994                         continue;
 995                 }
 996
 997                 /* abort launched rpcs in the test */
 998                 list_for_each_entry (rpc, &tsi->tsi_active_rpcs, crpc_list) {
 999                         spin_lock(&rpc->crpc_lock);
1000
1001                         srpc_abort_rpc(rpc, -EINTR);
1002
1003                         spin_unlock(&rpc->crpc_lock);
1004                 }
1005
1006                 spin_unlock(&tsi->tsi_lock);
1007         }
1008
1009         return 0;
1010 }
1011
1012 int
1013 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1014 {
1015         sfw_test_instance_t *tsi;
1016
1017         if (testidx < 0)
1018                 return -EINVAL;
1019
1020         if (testidx == 0) {
1021                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1022                 return 0;
1023         }
1024
1025         list_for_each_entry (tsi, &tsb->bat_tests, tsi_list) {
1026                 if (testidx-- > 1)
1027                         continue;
1028
1029                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1030                 return 0;
1031         }
1032
1033         return -ENOENT;
1034 }
1035
1036 void
1037 sfw_free_pages (srpc_server_rpc_t *rpc)
1038 {
1039         srpc_free_bulk(rpc->srpc_bulk);
1040         rpc->srpc_bulk = NULL;
1041 }
1042
1043 int
1044 sfw_alloc_pages (srpc_server_rpc_t *rpc, int npages, int sink)
1045 {
1046         LASSERT (rpc->srpc_bulk == NULL);
1047         LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
1048
1049         rpc->srpc_bulk = srpc_alloc_bulk(npages, sink);
1050         if (rpc->srpc_bulk == NULL) return -ENOMEM;
1051
1052         return 0;
1053 }
1054
1055 int
1056 sfw_add_test (srpc_server_rpc_t *rpc)
1057 {
1058         sfw_session_t     *sn = sfw_data.fw_session;
1059         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1060         srpc_test_reqst_t *request;
1061         int                rc;
1062         sfw_batch_t       *bat;
1063
1064         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1065         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1066
1067         if (request->tsr_loop == 0 ||
1068             request->tsr_concur == 0 ||
1069             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1070             request->tsr_ndest > SFW_MAX_NDESTS ||
1071             (request->tsr_is_client && request->tsr_ndest == 0) ||
1072             request->tsr_concur > SFW_MAX_CONCUR ||
1073             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1074             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1075                 reply->tsr_status = EINVAL;
1076                 return 0;
1077         }
1078
1079         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1080             sfw_find_test_case(request->tsr_service) == NULL) {
1081                 reply->tsr_status = ENOENT;
1082                 return 0;
1083         }
1084
1085         bat = sfw_bid2batch(request->tsr_bid);
1086         if (bat == NULL) {
1087                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1088                         rpc->srpc_service->sv_name,
1089                         libcfs_id2str(rpc->srpc_peer));
1090                 return -ENOMEM;
1091         }
1092
1093         if (sfw_batch_active(bat)) {
1094                 reply->tsr_status = EBUSY;
1095                 return 0;
1096         }
1097
1098         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1099                 /* rpc will be resumed later in sfw_bulk_ready */
1100                 return sfw_alloc_pages(rpc,
1101                                        sfw_id_pages(request->tsr_ndest), 1);
1102         }
1103
1104         rc = sfw_add_test_instance(bat, rpc);
1105         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1106                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1107                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1108                 request->tsr_is_client ? "client" : "server",
1109                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1110
1111         reply->tsr_status = (rc < 0) ? -rc : rc;
1112         return 0;
1113 }
1114
1115 int
1116 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1117 {
1118         sfw_session_t *sn = sfw_data.fw_session;
1119         int            rc = 0;
1120         sfw_batch_t   *bat;
1121
1122         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1123
1124         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1125                 reply->bar_status = ESRCH;
1126                 return 0;
1127         }
1128
1129         bat = sfw_find_batch(request->bar_bid);
1130         if (bat == NULL) {
1131                 reply->bar_status = ENOENT;
1132                 return 0;
1133         }
1134
1135         switch (request->bar_opc) {
1136         case SRPC_BATCH_OPC_RUN:
1137                 rc = sfw_run_batch(bat);
1138                 break;
1139
1140         case SRPC_BATCH_OPC_STOP:
1141                 rc = sfw_stop_batch(bat, request->bar_arg);
1142                 break;
1143
1144         case SRPC_BATCH_OPC_QUERY:
1145                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1146                 break;
1147
1148         default:
1149                 return -EINVAL; /* drop it */
1150         }
1151
1152         reply->bar_status = (rc < 0) ? -rc : rc;
1153         return 0;
1154 }
1155
1156 int
1157 sfw_handle_server_rpc (srpc_server_rpc_t *rpc)
1158 {
1159         srpc_service_t *sv = rpc->srpc_service;
1160         srpc_msg_t     *reply = &rpc->srpc_replymsg;
1161         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1162         int             rc = 0;
1163
1164         LASSERT (sfw_data.fw_active_srpc == NULL);
1165         LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1166
1167         spin_lock(&sfw_data.fw_lock);
1168
1169         if (sfw_data.fw_shuttingdown) {
1170                 spin_unlock(&sfw_data.fw_lock);
1171                 return -ESHUTDOWN;
1172         }
1173
1174         /* Remove timer to avoid racing with it or expiring active session */
1175         if (sfw_del_session_timer() != 0) {
1176                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1177                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1178                 spin_unlock(&sfw_data.fw_lock);
1179                 return -EAGAIN;
1180         }
1181
1182         sfw_data.fw_active_srpc = rpc;
1183         spin_unlock(&sfw_data.fw_lock);
1184
1185         sfw_unpack_message(request);
1186         LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1187
1188         switch(sv->sv_id) {
1189         default:
1190                 LBUG ();
1191         case SRPC_SERVICE_TEST:
1192                 rc = sfw_add_test(rpc);
1193                 break;
1194
1195         case SRPC_SERVICE_BATCH:
1196                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1197                                        &reply->msg_body.bat_reply);
1198                 break;
1199
1200         case SRPC_SERVICE_QUERY_STAT:
1201                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1202                                    &reply->msg_body.stat_reply);
1203                 break;
1204
1205         case SRPC_SERVICE_DEBUG:
1206                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1207                                        &reply->msg_body.dbg_reply);
1208                 break;
1209
1210         case SRPC_SERVICE_MAKE_SESSION:
1211                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1212                                       &reply->msg_body.mksn_reply);
1213                 break;
1214
1215         case SRPC_SERVICE_REMOVE_SESSION:
1216                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1217                                         &reply->msg_body.rmsn_reply);
1218                 break;
1219         }
1220
1221         rpc->srpc_done = sfw_server_rpc_done;
1222         spin_lock(&sfw_data.fw_lock);
1223
1224 #ifdef __KERNEL__
1225         if (!sfw_data.fw_shuttingdown)
1226                 sfw_add_session_timer();
1227 #else
1228         LASSERT (!sfw_data.fw_shuttingdown);
1229         sfw_add_session_timer();
1230 #endif
1231
1232         sfw_data.fw_active_srpc = NULL;
1233         spin_unlock(&sfw_data.fw_lock);
1234         return rc;
1235 }
1236
1237 int
1238 sfw_bulk_ready (srpc_server_rpc_t *rpc, int status)
1239 {
1240         srpc_service_t *sv = rpc->srpc_service;
1241         int             rc;
1242
1243         LASSERT (rpc->srpc_bulk != NULL);
1244         LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1245         LASSERT (sfw_data.fw_active_srpc == NULL);
1246         LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1247
1248         spin_lock(&sfw_data.fw_lock);
1249
1250         if (status != 0) {
1251                 CERROR ("Bulk transfer failed for RPC: "
1252                         "service %s, peer %s, status %d\n",
1253                         sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1254                 spin_unlock(&sfw_data.fw_lock);
1255                 return -EIO;
1256         }
1257
1258         if (sfw_data.fw_shuttingdown) {
1259                 spin_unlock(&sfw_data.fw_lock);
1260                 return -ESHUTDOWN;
1261         }
1262
1263         if (sfw_del_session_timer() != 0) {
1264                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1265                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1266                 spin_unlock(&sfw_data.fw_lock);
1267                 return -EAGAIN;
1268         }
1269
1270         sfw_data.fw_active_srpc = rpc;
1271         spin_unlock(&sfw_data.fw_lock);
1272
1273         rc = sfw_add_test(rpc);
1274
1275         spin_lock(&sfw_data.fw_lock);
1276
1277 #ifdef __KERNEL__
1278         if (!sfw_data.fw_shuttingdown)
1279                 sfw_add_session_timer();
1280 #else
1281         LASSERT (!sfw_data.fw_shuttingdown);
1282         sfw_add_session_timer();
1283 #endif
1284
1285         sfw_data.fw_active_srpc = NULL;
1286         spin_unlock(&sfw_data.fw_lock);
1287         return rc;
1288 }
1289
1290 srpc_client_rpc_t *
1291 sfw_create_rpc (lnet_process_id_t peer, int service,
1292                 int nbulkiov, int bulklen,
1293                 void (*done) (srpc_client_rpc_t *), void *priv)
1294 {
1295         srpc_client_rpc_t *rpc;
1296
1297         spin_lock(&sfw_data.fw_lock);
1298
1299         LASSERT (!sfw_data.fw_shuttingdown);
1300         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1301
1302         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1303                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1304                                  srpc_client_rpc_t, crpc_list);
1305                 list_del(&rpc->crpc_list);
1306                 spin_unlock(&sfw_data.fw_lock);
1307
1308                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1309                                      done, sfw_client_rpc_fini, priv);
1310                 return rpc;
1311         }
1312
1313         spin_unlock(&sfw_data.fw_lock);
1314
1315         rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1316                                      nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1317                                      priv);
1318         return rpc;
1319 }
1320
1321 void
1322 sfw_unpack_message (srpc_msg_t *msg)
1323 {
1324         if (msg->msg_magic == SRPC_MSG_MAGIC)
1325                 return; /* no flipping needed */
1326
1327         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1328
1329         __swab32s(&msg->msg_type);
1330
1331         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1332                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1333
1334                 __swab32s(&req->str_type);
1335                 __swab64s(&req->str_rpyid);
1336                 sfw_unpack_sid(req->str_sid);
1337                 return;
1338         }
1339
1340         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1341                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1342
1343                 __swab32s(&rep->str_status);
1344                 sfw_unpack_sid(rep->str_sid);
1345                 sfw_unpack_fw_counters(rep->str_fw);
1346                 sfw_unpack_rpc_counters(rep->str_rpc);
1347                 sfw_unpack_lnet_counters(rep->str_lnet);
1348                 return;
1349         }
1350
1351         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1352                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1353
1354                 __swab64s(&req->mksn_rpyid);
1355                 __swab32s(&req->mksn_force);
1356                 sfw_unpack_sid(req->mksn_sid);
1357                 return;
1358         }
1359
1360         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1361                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1362
1363                 __swab32s(&rep->mksn_status);
1364                 __swab32s(&rep->mksn_timeout);
1365                 sfw_unpack_sid(rep->mksn_sid);
1366                 return;
1367         }
1368
1369         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1370                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1371
1372                 __swab64s(&req->rmsn_rpyid);
1373                 sfw_unpack_sid(req->rmsn_sid);
1374                 return;
1375         }
1376
1377         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1378                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1379
1380                 __swab32s(&rep->rmsn_status);
1381                 sfw_unpack_sid(rep->rmsn_sid);
1382                 return;
1383         }
1384
1385         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1386                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1387
1388                 __swab64s(&req->dbg_rpyid);
1389                 __swab32s(&req->dbg_flags);
1390                 sfw_unpack_sid(req->dbg_sid);
1391                 return;
1392         }
1393
1394         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1395                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1396
1397                 __swab32s(&rep->dbg_nbatch);
1398                 __swab32s(&rep->dbg_timeout);
1399                 sfw_unpack_sid(rep->dbg_sid);
1400                 return;
1401         }
1402
1403         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1404                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1405
1406                 __swab32s(&req->bar_opc);
1407                 __swab64s(&req->bar_rpyid);
1408                 __swab32s(&req->bar_testidx);
1409                 __swab32s(&req->bar_arg);
1410                 sfw_unpack_sid(req->bar_sid);
1411                 __swab64s(&req->bar_bid.bat_id);
1412                 return;
1413         }
1414
1415         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1416                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1417
1418                 __swab32s(&rep->bar_status);
1419                 sfw_unpack_sid(rep->bar_sid);
1420                 return;
1421         }
1422
1423         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1424                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1425
1426                 __swab64s(&req->tsr_rpyid);
1427                 __swab64s(&req->tsr_bulkid);
1428                 __swab32s(&req->tsr_loop);
1429                 __swab32s(&req->tsr_ndest);
1430                 __swab32s(&req->tsr_concur);
1431                 __swab32s(&req->tsr_service);
1432                 sfw_unpack_sid(req->tsr_sid);
1433                 __swab64s(&req->tsr_bid.bat_id);
1434                 return;
1435         }
1436
1437         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1438                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1439
1440                 __swab32s(&rep->tsr_status);
1441                 sfw_unpack_sid(rep->tsr_sid);
1442                 return;
1443         }
1444
1445         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1446                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1447
1448                 __swab64s(&req->join_rpyid);
1449                 sfw_unpack_sid(req->join_sid);
1450                 return;
1451         }
1452
1453         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1454                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1455
1456                 __swab32s(&rep->join_status);
1457                 __swab32s(&rep->join_timeout);
1458                 sfw_unpack_sid(rep->join_sid);
1459                 return;
1460         }
1461
1462         LBUG ();
1463         return;
1464 }
1465
1466 void
1467 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1468 {
1469         LASSERT (atomic_read(&rpc->crpc_refcount) > 0);
1470         LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1471
1472         spin_lock(&rpc->crpc_lock);
1473         srpc_abort_rpc(rpc, -EINTR);
1474         spin_unlock(&rpc->crpc_lock);
1475         return;
1476 }
1477
1478 void
1479 sfw_post_rpc (srpc_client_rpc_t *rpc)
1480 {
1481         spin_lock(&rpc->crpc_lock);
1482
1483         LASSERT (!rpc->crpc_closed);
1484         LASSERT (!rpc->crpc_aborted);
1485         LASSERT (list_empty(&rpc->crpc_list));
1486         LASSERT (!sfw_data.fw_shuttingdown);
1487
1488         rpc->crpc_timeout = SFW_CLIENT_RPC_TIMEOUT;
1489         srpc_post_rpc(rpc);
1490
1491         spin_unlock(&rpc->crpc_lock);
1492         return;
1493 }
1494
1495 static srpc_service_t sfw_services[] =
1496 {
1497         {
1498                 .sv_name = "debug",
1499                 .sv_id   = SRPC_SERVICE_DEBUG,
1500         },
1501         {
1502                 .sv_name = "query stats",
1503                 .sv_id   = SRPC_SERVICE_QUERY_STAT,
1504         },
1505         {
1506                 .sv_name = "make sessin",
1507                 .sv_id   = SRPC_SERVICE_MAKE_SESSION,
1508         },
1509         {
1510                 .sv_name = "remove session",
1511                 .sv_id   = SRPC_SERVICE_REMOVE_SESSION,
1512         },
1513         {
1514                 .sv_name = "batch service",
1515                 .sv_id   = SRPC_SERVICE_BATCH,
1516         },
1517         {
1518                 .sv_name = "test service",
1519                 .sv_id   = SRPC_SERVICE_TEST,
1520         },
1521         {       .sv_name = NULL, }
1522 };
1523
1524 extern sfw_test_client_ops_t ping_test_client;
1525 extern srpc_service_t        ping_test_service;
1526
1527 extern sfw_test_client_ops_t brw_test_client;
1528 extern srpc_service_t        brw_test_service;
1529
1530 int
1531 sfw_startup (void)
1532 {
1533         int              i;
1534         int              rc;
1535         int              error;
1536         srpc_service_t  *sv;
1537         sfw_test_case_t *tsc;
1538
1539 #ifndef __KERNEL__
1540         char *s;
1541
1542         s = getenv("SESSION_TIMEOUT");
1543         session_timeout = s != NULL ? atoi(s) : session_timeout;
1544
1545         s = getenv("BRW_INJECT_ERRORS");
1546         brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
1547 #endif
1548
1549         if (session_timeout < 0) {
1550                 CERROR ("Session timeout must be non-negative: %d\n",
1551                         session_timeout);
1552                 return -EINVAL;
1553         }
1554
1555         if (session_timeout == 0)
1556                 CWARN ("Zero session_timeout specified "
1557                        "- test sessions never expire.\n");
1558
1559         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1560
1561         sfw_data.fw_session     = NULL;
1562         sfw_data.fw_active_srpc = NULL;
1563         spin_lock_init(&sfw_data.fw_lock);
1564         atomic_set(&sfw_data.fw_nzombies, 0);
1565         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1566         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1567         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1568
1569         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1570         LASSERT (rc == 0);
1571         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1572         LASSERT (rc == 0);
1573
1574         error = 0;
1575         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
1576                 sv = tsc->tsc_srv_service;
1577                 sv->sv_concur = SFW_TEST_CONCURRENCY;
1578
1579                 rc = srpc_add_service(sv);
1580                 LASSERT (rc != -EBUSY);
1581                 if (rc != 0) {
1582                         CWARN ("Failed to add %s service: %d\n",
1583                                sv->sv_name, rc);
1584                         error = rc;
1585                 }
1586         }
1587
1588         for (i = 0; ; i++) {
1589                 sv = &sfw_services[i];
1590                 if (sv->sv_name == NULL) break;
1591
1592                 sv->sv_bulk_ready = NULL;
1593                 sv->sv_handler    = sfw_handle_server_rpc;
1594                 sv->sv_concur     = SFW_SERVICE_CONCURRENCY;
1595                 if (sv->sv_id == SRPC_SERVICE_TEST)
1596                         sv->sv_bulk_ready = sfw_bulk_ready;
1597
1598                 rc = srpc_add_service(sv);
1599                 LASSERT (rc != -EBUSY);
1600                 if (rc != 0) {
1601                         CWARN ("Failed to add %s service: %d\n",
1602                                sv->sv_name, rc);
1603                         error = rc;
1604                 }
1605
1606                 /* about to sfw_shutdown, no need to add buffer */
1607                 if (error) continue;
1608
1609                 rc = srpc_service_add_buffers(sv, SFW_POST_BUFFERS);
1610                 if (rc != SFW_POST_BUFFERS) {
1611                         CWARN ("Failed to reserve enough buffers: "
1612                                "service %s, %d needed, %d reserved\n",
1613                                sv->sv_name, SFW_POST_BUFFERS, rc);
1614                         error = -ENOMEM;
1615                 }
1616         }
1617
1618         if (error != 0)
1619                 sfw_shutdown();
1620         return error;
1621 }
1622
1623 void
1624 sfw_shutdown (void)
1625 {
1626         srpc_service_t  *sv;
1627         sfw_test_case_t *tsc;
1628         int              i;
1629
1630         spin_lock(&sfw_data.fw_lock);
1631
1632         sfw_data.fw_shuttingdown = 1;
1633 #ifdef __KERNEL__
1634         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1635                        "waiting for active RPC to finish.\n");
1636 #else
1637         LASSERT (sfw_data.fw_active_srpc == NULL);
1638 #endif
1639
1640         if (sfw_del_session_timer() != 0)
1641                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1642                                "waiting for session timer to explode.\n");
1643
1644         sfw_deactivate_session();
1645         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1646                        sfw_data.fw_lock,
1647                        "waiting for %d zombie sessions to die.\n",
1648                        atomic_read(&sfw_data.fw_nzombies));
1649
1650         spin_unlock(&sfw_data.fw_lock);
1651
1652         for (i = 0; ; i++) {
1653                 sv = &sfw_services[i];
1654                 if (sv->sv_name == NULL)
1655                         break;
1656
1657                 srpc_shutdown_service(sv);
1658                 srpc_remove_service(sv);
1659         }
1660
1661         list_for_each_entry (tsc, &sfw_data.fw_tests, tsc_list) {
1662                 sv = tsc->tsc_srv_service;
1663                 srpc_shutdown_service(sv);
1664                 srpc_remove_service(sv);
1665         }
1666
1667         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1668                 srpc_client_rpc_t *rpc;
1669
1670                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1671                                  srpc_client_rpc_t, crpc_list);
1672                 list_del(&rpc->crpc_list);
1673
1674                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1675         }
1676
1677         for (i = 0; ; i++) {
1678                 sv = &sfw_services[i];
1679                 if (sv->sv_name == NULL)
1680                         break;
1681
1682                 srpc_wait_service_shutdown(sv);
1683         }
1684
1685         while (!list_empty(&sfw_data.fw_tests)) {
1686                 tsc = list_entry(sfw_data.fw_tests.next,
1687                                  sfw_test_case_t, tsc_list);
1688
1689                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1690
1691                 list_del(&tsc->tsc_list);
1692                 LIBCFS_FREE(tsc, sizeof(*tsc));
1693         }
1694
1695         return;
1696 }