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     1792
  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         sfw_test_case_t *tsc;
 218
 219         if (sn == NULL) return;
 220
 221         LASSERT (!sn->sn_timer_active);
 222
 223         sfw_data.fw_session = NULL;
 224         atomic_inc(&sfw_data.fw_nzombies);
 225         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 226
 227         spin_unlock(&sfw_data.fw_lock);
 228
 229         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 230                                        sfw_test_case_t, tsc_list) {
 231                 srpc_abort_service(tsc->tsc_srv_service);
 232         }
 233
 234         spin_lock(&sfw_data.fw_lock);
 235
 236         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 237                                        sfw_batch_t, bat_list) {
 238                 if (sfw_batch_active(tsb)) {
 239                         nactive++;
 240                         sfw_stop_batch(tsb, 1);
 241                 }
 242         }
 243
 244         if (nactive != 0)
 245                 return;   /* wait for active batches to stop */
 246
 247         list_del_init(&sn->sn_list);
 248         spin_unlock(&sfw_data.fw_lock);
 249
 250         sfw_destroy_session(sn);
 251
 252         spin_lock(&sfw_data.fw_lock);
 253         return;
 254 }
 255
 256 #ifndef __KERNEL__
 257
 258 int
 259 sfw_session_removed (void)
 260 {
 261         return (sfw_data.fw_session == NULL) ? 1 : 0;
 262 }
 263
 264 #endif
 265
 266 void
 267 sfw_session_expired (void *data)
 268 {
 269         sfw_session_t *sn = data;
 270
 271         spin_lock(&sfw_data.fw_lock);
 272
 273         LASSERT (sn->sn_timer_active);
 274         LASSERT (sn == sfw_data.fw_session);
 275
 276         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 277                libcfs_nid2str(sn->sn_id.ses_nid),
 278                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 279
 280         sn->sn_timer_active = 0;
 281         sfw_deactivate_session();
 282
 283         spin_unlock(&sfw_data.fw_lock);
 284         return;
 285 }
 286
 287 static inline void
 288 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
 289 {
 290         stt_timer_t *timer = &sn->sn_timer;
 291
 292         memset(sn, 0, sizeof(sfw_session_t));
 293         CFS_INIT_LIST_HEAD(&sn->sn_list);
 294         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 295         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 296         atomic_set(&sn->sn_brw_errors, 0);
 297         atomic_set(&sn->sn_ping_errors, 0);
 298         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 299
 300         sn->sn_timer_active = 0;
 301         sn->sn_id           = sid;
 302         sn->sn_timeout      = session_timeout;
 303
 304         timer->stt_data = sn;
 305         timer->stt_func = sfw_session_expired;
 306         CFS_INIT_LIST_HEAD(&timer->stt_list);
 307 }
 308
 309 /* completion handler for incoming framework RPCs */
 310 void
 311 sfw_server_rpc_done (srpc_server_rpc_t *rpc)
 312 {
 313         srpc_service_t *sv = rpc->srpc_service;
 314         int             status = rpc->srpc_status;
 315
 316         CDEBUG (D_NET,
 317                 "Incoming framework RPC done: "
 318                 "service %s, peer %s, status %s:%d\n",
 319                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 320                 swi_state2str(rpc->srpc_wi.wi_state),
 321                 status);
 322
 323         if (rpc->srpc_bulk != NULL)
 324                 sfw_free_pages(rpc);
 325         return;
 326 }
 327
 328 void
 329 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 330 {
 331         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 332         LASSERT (list_empty(&rpc->crpc_list));
 333         LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
 334 #ifndef __KERNEL__
 335         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 336 #endif
 337
 338         CDEBUG (D_NET,
 339                 "Outgoing framework RPC done: "
 340                 "service %d, peer %s, status %s:%d:%d\n",
 341                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 342                 swi_state2str(rpc->crpc_wi.wi_state),
 343                 rpc->crpc_aborted, rpc->crpc_status);
 344
 345         spin_lock(&sfw_data.fw_lock);
 346
 347         /* my callers must finish all RPCs before shutting me down */
 348         LASSERT (!sfw_data.fw_shuttingdown);
 349         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 350
 351         spin_unlock(&sfw_data.fw_lock);
 352         return;
 353 }
 354
 355 sfw_batch_t *
 356 sfw_find_batch (lst_bid_t bid)
 357 {
 358         sfw_session_t *sn = sfw_data.fw_session;
 359         sfw_batch_t   *bat;
 360
 361         LASSERT (sn != NULL);
 362
 363         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 364                                        sfw_batch_t, bat_list) {
 365                 if (bat->bat_id.bat_id == bid.bat_id)
 366                         return bat;
 367         }
 368
 369         return NULL;
 370 }
 371
 372 sfw_batch_t *
 373 sfw_bid2batch (lst_bid_t bid)
 374 {
 375         sfw_session_t *sn = sfw_data.fw_session;
 376         sfw_batch_t   *bat;
 377
 378         LASSERT (sn != NULL);
 379
 380         bat = sfw_find_batch(bid);
 381         if (bat != NULL)
 382                 return bat;
 383
 384         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 385         if (bat == NULL)
 386                 return NULL;
 387
 388         bat->bat_error    = 0;
 389         bat->bat_session  = sn;
 390         bat->bat_id       = bid;
 391         atomic_set(&bat->bat_nactive, 0);
 392         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 393
 394         list_add_tail(&bat->bat_list, &sn->sn_batches);
 395         return bat;
 396 }
 397
 398 int
 399 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 400 {
 401         sfw_session_t  *sn = sfw_data.fw_session;
 402         sfw_counters_t *cnt = &reply->str_fw;
 403         sfw_batch_t    *bat;
 404
 405         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 406
 407         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 408                 reply->str_status = EINVAL;
 409                 return 0;
 410         }
 411
 412         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 413                 reply->str_status = ESRCH;
 414                 return 0;
 415         }
 416
 417         LNET_LOCK();
 418         reply->str_lnet = the_lnet.ln_counters;
 419         LNET_UNLOCK();
 420
 421         srpc_get_counters(&reply->str_rpc);
 422
 423         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 424         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 425         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 426
 427         cnt->active_tests = cnt->active_batches = 0;
 428         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 429                                        sfw_batch_t, bat_list) {
 430                 int n = atomic_read(&bat->bat_nactive);
 431
 432                 if (n > 0) {
 433                         cnt->active_batches++;
 434                         cnt->active_tests += n;
 435                 }
 436         }
 437
 438         reply->str_status = 0;
 439         return 0;
 440 }
 441
 442 int
 443 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 444 {
 445         sfw_session_t *sn = sfw_data.fw_session;
 446
 447         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 448                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 449                 reply->mksn_status = EINVAL;
 450                 return 0;
 451         }
 452
 453         if (sn != NULL) {
 454                 reply->mksn_status  = 0;
 455                 reply->mksn_sid     = sn->sn_id;
 456                 reply->mksn_timeout = sn->sn_timeout;
 457
 458                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 459                         atomic_inc(&sn->sn_refcount);
 460                         return 0;
 461                 }
 462
 463                 if (!request->mksn_force) {
 464                         reply->mksn_status = EBUSY;
 465                         strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 466                         return 0;
 467                 }
 468         }
 469
 470         /* brand new or create by force */
 471         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 472         if (sn == NULL) {
 473                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 474                 return -ENOMEM;
 475         }
 476
 477         sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
 478
 479         spin_lock(&sfw_data.fw_lock);
 480
 481         sfw_deactivate_session();
 482         LASSERT (sfw_data.fw_session == NULL);
 483         sfw_data.fw_session = sn;
 484
 485         spin_unlock(&sfw_data.fw_lock);
 486
 487         reply->mksn_status  = 0;
 488         reply->mksn_sid     = sn->sn_id;
 489         reply->mksn_timeout = sn->sn_timeout;
 490         return 0;
 491 }
 492
 493 int
 494 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 495 {
 496         sfw_session_t *sn = sfw_data.fw_session;
 497
 498         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 499
 500         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 501                 reply->rmsn_status = EINVAL;
 502                 return 0;
 503         }
 504
 505         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 506                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 507                 return 0;
 508         }
 509
 510         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 511                 reply->rmsn_status = 0;
 512                 return 0;
 513         }
 514
 515         spin_lock(&sfw_data.fw_lock);
 516         sfw_deactivate_session();
 517         spin_unlock(&sfw_data.fw_lock);
 518
 519         reply->rmsn_status = 0;
 520         reply->rmsn_sid    = LST_INVALID_SID;
 521         LASSERT (sfw_data.fw_session == NULL);
 522         return 0;
 523 }
 524
 525 int
 526 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 527 {
 528         sfw_session_t *sn = sfw_data.fw_session;
 529
 530         if (sn == NULL) {
 531                 reply->dbg_status = ESRCH;
 532                 reply->dbg_sid    = LST_INVALID_SID;
 533                 return 0;
 534         }
 535
 536         reply->dbg_status  = 0;
 537         reply->dbg_sid     = sn->sn_id;
 538         reply->dbg_timeout = sn->sn_timeout;
 539         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 540
 541         return 0;
 542 }
 543
 544 void
 545 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 546 {
 547         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 548         sfw_test_instance_t *tsi = tsu->tsu_instance;
 549
 550         /* Called with hold of tsi->tsi_lock */
 551         LASSERT (list_empty(&rpc->crpc_list));
 552         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 553 }
 554
 555 int
 556 sfw_load_test (sfw_test_instance_t *tsi)
 557 {
 558         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 559         int              nrequired = sfw_test_buffers(tsi);
 560         int              nposted;
 561
 562         LASSERT (tsc != NULL);
 563
 564         if (tsi->tsi_is_client) {
 565                 tsi->tsi_ops = tsc->tsc_cli_ops;
 566                 return 0;
 567         }
 568
 569         nposted = srpc_service_add_buffers(tsc->tsc_srv_service, nrequired);
 570         if (nposted != nrequired) {
 571                 CWARN ("Failed to reserve enough buffers: "
 572                        "service %s, %d needed, %d reserved\n",
 573                        tsc->tsc_srv_service->sv_name, nrequired, nposted);
 574                 srpc_service_remove_buffers(tsc->tsc_srv_service, nposted);
 575                 return -ENOMEM;
 576         }
 577
 578         CDEBUG (D_NET, "Reserved %d buffers for test %s\n",
 579                 nposted, tsc->tsc_srv_service->sv_name);
 580         return 0;
 581 }
 582
 583 void
 584 sfw_unload_test (sfw_test_instance_t *tsi)
 585 {
 586         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 587
 588         LASSERT (tsc != NULL);
 589
 590         if (!tsi->tsi_is_client)
 591                 srpc_service_remove_buffers(tsc->tsc_srv_service,
 592                                             sfw_test_buffers(tsi));
 593         return;
 594 }
 595
 596 void
 597 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 598 {
 599         srpc_client_rpc_t *rpc;
 600         sfw_test_unit_t   *tsu;
 601
 602         if (!tsi->tsi_is_client) goto clean;
 603
 604         tsi->tsi_ops->tso_fini(tsi);
 605
 606         LASSERT (!tsi->tsi_stopping);
 607         LASSERT (list_empty(&tsi->tsi_active_rpcs));
 608         LASSERT (!sfw_test_active(tsi));
 609
 610         while (!list_empty(&tsi->tsi_units)) {
 611                 tsu = list_entry(tsi->tsi_units.next,
 612                                  sfw_test_unit_t, tsu_list);
 613                 list_del(&tsu->tsu_list);
 614                 LIBCFS_FREE(tsu, sizeof(*tsu));
 615         }
 616
 617         while (!list_empty(&tsi->tsi_free_rpcs)) {
 618                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 619                                  srpc_client_rpc_t, crpc_list);
 620                 list_del(&rpc->crpc_list);
 621                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 622         }
 623
 624 clean:
 625         sfw_unload_test(tsi);
 626         LIBCFS_FREE(tsi, sizeof(*tsi));
 627         return;
 628 }
 629
 630 void
 631 sfw_destroy_batch (sfw_batch_t *tsb)
 632 {
 633         sfw_test_instance_t *tsi;
 634
 635         LASSERT (!sfw_batch_active(tsb));
 636         LASSERT (list_empty(&tsb->bat_list));
 637
 638         while (!list_empty(&tsb->bat_tests)) {
 639                 tsi = list_entry(tsb->bat_tests.next,
 640                                  sfw_test_instance_t, tsi_list);
 641                 list_del_init(&tsi->tsi_list);
 642                 sfw_destroy_test_instance(tsi);
 643         }
 644
 645         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 646         return;
 647 }
 648
 649 void
 650 sfw_destroy_session (sfw_session_t *sn)
 651 {
 652         sfw_batch_t *batch;
 653
 654         LASSERT (list_empty(&sn->sn_list));
 655         LASSERT (sn != sfw_data.fw_session);
 656
 657         while (!list_empty(&sn->sn_batches)) {
 658                 batch = list_entry(sn->sn_batches.next,
 659                                    sfw_batch_t, bat_list);
 660                 list_del_init(&batch->bat_list);
 661                 sfw_destroy_batch(batch);
 662         }
 663
 664         LIBCFS_FREE(sn, sizeof(*sn));
 665         atomic_dec(&sfw_data.fw_nzombies);
 666         return;
 667 }
 668
 669 void
 670 sfw_unpack_test_req (srpc_msg_t *msg)
 671 {
 672         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 673
 674         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 675         LASSERT (req->tsr_is_client);
 676
 677         if (msg->msg_magic == SRPC_MSG_MAGIC)
 678                 return; /* no flipping needed */
 679
 680         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 681
 682         if (req->tsr_service == SRPC_SERVICE_BRW) {
 683                 test_bulk_req_t *bulk = &req->tsr_u.bulk;
 684
 685                 __swab32s(&bulk->blk_opc);
 686                 __swab32s(&bulk->blk_npg);
 687                 __swab32s(&bulk->blk_flags);
 688                 return;
 689         }
 690
 691         if (req->tsr_service == SRPC_SERVICE_PING) {
 692                 test_ping_req_t *ping = &req->tsr_u.ping;
 693
 694                 __swab32s(&ping->png_size);
 695                 __swab32s(&ping->png_flags);
 696                 return;
 697         }
 698
 699         LBUG ();
 700         return;
 701 }
 702
 703 int
 704 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 705 {
 706         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 707         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 708         srpc_bulk_t         *bk = rpc->srpc_bulk;
 709         int                  ndest = req->tsr_ndest;
 710         sfw_test_unit_t     *tsu;
 711         sfw_test_instance_t *tsi;
 712         int                  i;
 713         int                  rc;
 714
 715         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 716         if (tsi == NULL) {
 717                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 718                         tsb->bat_id.bat_id);
 719                 return -ENOMEM;
 720         }
 721
 722         memset(tsi, 0, sizeof(*tsi));
 723         spin_lock_init(&tsi->tsi_lock);
 724         atomic_set(&tsi->tsi_nactive, 0);
 725         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 726         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 727         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 728
 729         tsi->tsi_stopping      = 0;
 730         tsi->tsi_batch         = tsb;
 731         tsi->tsi_loop          = req->tsr_loop;
 732         tsi->tsi_concur        = req->tsr_concur;
 733         tsi->tsi_service       = req->tsr_service;
 734         tsi->tsi_is_client     = !!(req->tsr_is_client);
 735         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 736
 737         rc = sfw_load_test(tsi);
 738         if (rc != 0) {
 739                 LIBCFS_FREE(tsi, sizeof(*tsi));
 740                 return rc;
 741         }
 742
 743         LASSERT (!sfw_batch_active(tsb));
 744
 745         if (!tsi->tsi_is_client) {
 746                 /* it's test server, just add it to tsb */
 747                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 748                 return 0;
 749         }
 750
 751         LASSERT (bk != NULL);
 752 #ifndef __KERNEL__
 753         LASSERT (bk->bk_pages != NULL);
 754 #endif
 755         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 756         LASSERT ((unsigned int)bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
 757
 758         sfw_unpack_test_req(msg);
 759         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 760
 761         for (i = 0; i < ndest; i++) {
 762                 lnet_process_id_t *dests;
 763                 lnet_process_id_t  id;
 764                 int                j;
 765
 766 #ifdef __KERNEL__
 767                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 768                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 769 #else
 770                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 771 #endif
 772                 id = dests[i % SFW_ID_PER_PAGE];
 773                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 774                         sfw_unpack_id(id);
 775
 776                 for (j = 0; j < tsi->tsi_concur; j++) {
 777                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 778                         if (tsu == NULL) {
 779                                 rc = -ENOMEM;
 780                                 CERROR ("Can't allocate tsu for %d\n",
 781                                         tsi->tsi_service);
 782                                 goto error;
 783                         }
 784
 785                         tsu->tsu_dest     = id;
 786                         tsu->tsu_instance = tsi;
 787                         tsu->tsu_private  = NULL;
 788                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 789                 }
 790         }
 791
 792         rc = tsi->tsi_ops->tso_init(tsi);
 793         if (rc == 0) {
 794                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 795                 return 0;
 796         }
 797
 798 error:
 799         LASSERT (rc != 0);
 800         sfw_destroy_test_instance(tsi);
 801         return rc;
 802 }
 803
 804 static void
 805 sfw_test_unit_done (sfw_test_unit_t *tsu)
 806 {
 807         sfw_test_instance_t *tsi = tsu->tsu_instance;
 808         sfw_batch_t         *tsb = tsi->tsi_batch;
 809         sfw_session_t       *sn = tsb->bat_session;
 810
 811         LASSERT (sfw_test_active(tsi));
 812
 813         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 814                 return;
 815
 816         /* the test instance is done */
 817         spin_lock(&tsi->tsi_lock);
 818
 819         tsi->tsi_stopping = 0;
 820
 821         spin_unlock(&tsi->tsi_lock);
 822
 823         spin_lock(&sfw_data.fw_lock);
 824
 825         if (!atomic_dec_and_test(&tsb->bat_nactive) || /* tsb still active */
 826             sn == sfw_data.fw_session) {               /* sn also active */
 827                 spin_unlock(&sfw_data.fw_lock);
 828                 return;
 829         }
 830
 831         LASSERT (!list_empty(&sn->sn_list)); /* I'm a zombie! */
 832
 833         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 834                                        sfw_batch_t, bat_list) {
 835                 if (sfw_batch_active(tsb)) {
 836                         spin_unlock(&sfw_data.fw_lock);
 837                         return;
 838                 }
 839         }
 840
 841         list_del_init(&sn->sn_list);
 842         spin_unlock(&sfw_data.fw_lock);
 843
 844         sfw_destroy_session(sn);
 845         return;
 846 }
 847
 848 void
 849 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 850 {
 851         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 852         sfw_test_instance_t *tsi = tsu->tsu_instance;
 853         int                  done = 0;
 854
 855         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 856
 857         spin_lock(&tsi->tsi_lock);
 858
 859         LASSERT (sfw_test_active(tsi));
 860         LASSERT (!list_empty(&rpc->crpc_list));
 861
 862         list_del_init(&rpc->crpc_list);
 863
 864         /* batch is stopping or loop is done or get error */
 865         if (tsi->tsi_stopping ||
 866             tsu->tsu_loop == 0 ||
 867             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 868                 done = 1;
 869
 870         /* dec ref for poster */
 871         srpc_client_rpc_decref(rpc);
 872
 873         spin_unlock(&tsi->tsi_lock);
 874
 875         if (!done) {
 876                 swi_schedule_workitem(&tsu->tsu_worker);
 877                 return;
 878         }
 879
 880         sfw_test_unit_done(tsu);
 881         return;
 882 }
 883
 884 int
 885 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
 886                      int nblk, int blklen, srpc_client_rpc_t **rpcpp)
 887 {
 888         srpc_client_rpc_t   *rpc = NULL;
 889         sfw_test_instance_t *tsi = tsu->tsu_instance;
 890
 891         spin_lock(&tsi->tsi_lock);
 892
 893         LASSERT (sfw_test_active(tsi));
 894
 895         if (!list_empty(&tsi->tsi_free_rpcs)) {
 896                 /* pick request from buffer */
 897                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 898                                  srpc_client_rpc_t, crpc_list);
 899                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 900                 list_del_init(&rpc->crpc_list);
 901
 902                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 903                                      blklen, sfw_test_rpc_done,
 904                                      sfw_test_rpc_fini, tsu);
 905         }
 906
 907         spin_unlock(&tsi->tsi_lock);
 908
 909         if (rpc == NULL)
 910                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 911                                              blklen, sfw_test_rpc_done,
 912                                              sfw_test_rpc_fini, tsu);
 913         if (rpc == NULL) {
 914                 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
 915                 return -ENOMEM;
 916         }
 917
 918         *rpcpp = rpc;
 919         return 0;
 920 }
 921
 922 int
 923 sfw_run_test (swi_workitem_t *wi)
 924 {
 925         sfw_test_unit_t     *tsu = wi->wi_data;
 926         sfw_test_instance_t *tsi = tsu->tsu_instance;
 927         srpc_client_rpc_t   *rpc = NULL;
 928
 929         LASSERT (wi == &tsu->tsu_worker);
 930
 931         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 932                 LASSERT (rpc == NULL);
 933                 goto test_done;
 934         }
 935
 936         LASSERT (rpc != NULL);
 937
 938         spin_lock(&tsi->tsi_lock);
 939
 940         if (tsi->tsi_stopping) {
 941                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 942                 spin_unlock(&tsi->tsi_lock);
 943                 goto test_done;
 944         }
 945
 946         if (tsu->tsu_loop > 0)
 947                 tsu->tsu_loop--;
 948
 949         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 950         spin_unlock(&tsi->tsi_lock);
 951
 952         rpc->crpc_timeout = SFW_TEST_RPC_TIMEOUT;
 953
 954         spin_lock(&rpc->crpc_lock);
 955         srpc_post_rpc(rpc);
 956         spin_unlock(&rpc->crpc_lock);
 957         return 0;
 958
 959 test_done:
 960         /*
 961          * No one can schedule me now since:
 962          * - previous RPC, if any, has done and
 963          * - no new RPC is initiated.
 964          * - my batch is still active; no one can run it again now.
 965          * Cancel pending schedules and prevent future schedule attempts:
 966          */
 967         swi_kill_workitem(wi);
 968         sfw_test_unit_done(tsu);
 969         return 1;
 970 }
 971
 972 int
 973 sfw_run_batch (sfw_batch_t *tsb)
 974 {
 975         swi_workitem_t      *wi;
 976         sfw_test_unit_t     *tsu;
 977         sfw_test_instance_t *tsi;
 978
 979         if (sfw_batch_active(tsb)) {
 980                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
 981                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
 982                 return 0;
 983         }
 984
 985         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
 986                                        sfw_test_instance_t, tsi_list) {
 987                 if (!tsi->tsi_is_client) /* skip server instances */
 988                         continue;
 989
 990                 LASSERT (!tsi->tsi_stopping);
 991                 LASSERT (!sfw_test_active(tsi));
 992
 993                 atomic_inc(&tsb->bat_nactive);
 994
 995                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
 996                                                sfw_test_unit_t, tsu_list) {
 997                         atomic_inc(&tsi->tsi_nactive);
 998                         tsu->tsu_loop = tsi->tsi_loop;
 999                         wi = &tsu->tsu_worker;
1000                         swi_init_workitem(wi, tsu, sfw_run_test);
1001                         swi_schedule_workitem(wi);
1002                 }
1003         }
1004
1005         return 0;
1006 }
1007
1008 int
1009 sfw_stop_batch (sfw_batch_t *tsb, int force)
1010 {
1011         sfw_test_instance_t *tsi;
1012         srpc_client_rpc_t   *rpc;
1013
1014         if (!sfw_batch_active(tsb)) {
1015                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1016                 return 0;
1017         }
1018
1019         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1020                                        sfw_test_instance_t, tsi_list) {
1021                 spin_lock(&tsi->tsi_lock);
1022
1023                 if (!tsi->tsi_is_client ||
1024                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1025                         spin_unlock(&tsi->tsi_lock);
1026                         continue;
1027                 }
1028
1029                 tsi->tsi_stopping = 1;
1030
1031                 if (!force) {
1032                         spin_unlock(&tsi->tsi_lock);
1033                         continue;
1034                 }
1035
1036                 /* abort launched rpcs in the test */
1037                 cfs_list_for_each_entry_typed (rpc, &tsi->tsi_active_rpcs,
1038                                                srpc_client_rpc_t, crpc_list) {
1039                         spin_lock(&rpc->crpc_lock);
1040
1041                         srpc_abort_rpc(rpc, -EINTR);
1042
1043                         spin_unlock(&rpc->crpc_lock);
1044                 }
1045
1046                 spin_unlock(&tsi->tsi_lock);
1047         }
1048
1049         return 0;
1050 }
1051
1052 int
1053 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1054 {
1055         sfw_test_instance_t *tsi;
1056
1057         if (testidx < 0)
1058                 return -EINVAL;
1059
1060         if (testidx == 0) {
1061                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1062                 return 0;
1063         }
1064
1065         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1066                                        sfw_test_instance_t, tsi_list) {
1067                 if (testidx-- > 1)
1068                         continue;
1069
1070                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1071                 return 0;
1072         }
1073
1074         return -ENOENT;
1075 }
1076
1077 void
1078 sfw_free_pages (srpc_server_rpc_t *rpc)
1079 {
1080         srpc_free_bulk(rpc->srpc_bulk);
1081         rpc->srpc_bulk = NULL;
1082 }
1083
1084 int
1085 sfw_alloc_pages (srpc_server_rpc_t *rpc, int npages, int sink)
1086 {
1087         LASSERT (rpc->srpc_bulk == NULL);
1088         LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
1089
1090         rpc->srpc_bulk = srpc_alloc_bulk(npages, sink);
1091         if (rpc->srpc_bulk == NULL) return -ENOMEM;
1092
1093         return 0;
1094 }
1095
1096 int
1097 sfw_add_test (srpc_server_rpc_t *rpc)
1098 {
1099         sfw_session_t     *sn = sfw_data.fw_session;
1100         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1101         srpc_test_reqst_t *request;
1102         int                rc;
1103         sfw_batch_t       *bat;
1104
1105         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1106         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1107
1108         if (request->tsr_loop == 0 ||
1109             request->tsr_concur == 0 ||
1110             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1111             request->tsr_ndest > SFW_MAX_NDESTS ||
1112             (request->tsr_is_client && request->tsr_ndest == 0) ||
1113             request->tsr_concur > SFW_MAX_CONCUR ||
1114             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1115             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1116                 reply->tsr_status = EINVAL;
1117                 return 0;
1118         }
1119
1120         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1121             sfw_find_test_case(request->tsr_service) == NULL) {
1122                 reply->tsr_status = ENOENT;
1123                 return 0;
1124         }
1125
1126         bat = sfw_bid2batch(request->tsr_bid);
1127         if (bat == NULL) {
1128                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1129                         rpc->srpc_service->sv_name,
1130                         libcfs_id2str(rpc->srpc_peer));
1131                 return -ENOMEM;
1132         }
1133
1134         if (sfw_batch_active(bat)) {
1135                 reply->tsr_status = EBUSY;
1136                 return 0;
1137         }
1138
1139         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1140                 /* rpc will be resumed later in sfw_bulk_ready */
1141                 return sfw_alloc_pages(rpc,
1142                                        sfw_id_pages(request->tsr_ndest), 1);
1143         }
1144
1145         rc = sfw_add_test_instance(bat, rpc);
1146         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1147                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1148                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1149                 request->tsr_is_client ? "client" : "server",
1150                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1151
1152         reply->tsr_status = (rc < 0) ? -rc : rc;
1153         return 0;
1154 }
1155
1156 int
1157 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1158 {
1159         sfw_session_t *sn = sfw_data.fw_session;
1160         int            rc = 0;
1161         sfw_batch_t   *bat;
1162
1163         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1164
1165         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1166                 reply->bar_status = ESRCH;
1167                 return 0;
1168         }
1169
1170         bat = sfw_find_batch(request->bar_bid);
1171         if (bat == NULL) {
1172                 reply->bar_status = ENOENT;
1173                 return 0;
1174         }
1175
1176         switch (request->bar_opc) {
1177         case SRPC_BATCH_OPC_RUN:
1178                 rc = sfw_run_batch(bat);
1179                 break;
1180
1181         case SRPC_BATCH_OPC_STOP:
1182                 rc = sfw_stop_batch(bat, request->bar_arg);
1183                 break;
1184
1185         case SRPC_BATCH_OPC_QUERY:
1186                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1187                 break;
1188
1189         default:
1190                 return -EINVAL; /* drop it */
1191         }
1192
1193         reply->bar_status = (rc < 0) ? -rc : rc;
1194         return 0;
1195 }
1196
1197 int
1198 sfw_handle_server_rpc (srpc_server_rpc_t *rpc)
1199 {
1200         srpc_service_t *sv = rpc->srpc_service;
1201         srpc_msg_t     *reply = &rpc->srpc_replymsg;
1202         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1203         int             rc = 0;
1204
1205         LASSERT (sfw_data.fw_active_srpc == NULL);
1206         LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1207
1208         spin_lock(&sfw_data.fw_lock);
1209
1210         if (sfw_data.fw_shuttingdown) {
1211                 spin_unlock(&sfw_data.fw_lock);
1212                 return -ESHUTDOWN;
1213         }
1214
1215         /* Remove timer to avoid racing with it or expiring active session */
1216         if (sfw_del_session_timer() != 0) {
1217                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1218                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1219                 spin_unlock(&sfw_data.fw_lock);
1220                 return -EAGAIN;
1221         }
1222
1223         sfw_data.fw_active_srpc = rpc;
1224         spin_unlock(&sfw_data.fw_lock);
1225
1226         sfw_unpack_message(request);
1227         LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1228
1229         switch(sv->sv_id) {
1230         default:
1231                 LBUG ();
1232         case SRPC_SERVICE_TEST:
1233                 rc = sfw_add_test(rpc);
1234                 break;
1235
1236         case SRPC_SERVICE_BATCH:
1237                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1238                                        &reply->msg_body.bat_reply);
1239                 break;
1240
1241         case SRPC_SERVICE_QUERY_STAT:
1242                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1243                                    &reply->msg_body.stat_reply);
1244                 break;
1245
1246         case SRPC_SERVICE_DEBUG:
1247                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1248                                        &reply->msg_body.dbg_reply);
1249                 break;
1250
1251         case SRPC_SERVICE_MAKE_SESSION:
1252                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1253                                       &reply->msg_body.mksn_reply);
1254                 break;
1255
1256         case SRPC_SERVICE_REMOVE_SESSION:
1257                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1258                                         &reply->msg_body.rmsn_reply);
1259                 break;
1260         }
1261
1262         rpc->srpc_done = sfw_server_rpc_done;
1263         spin_lock(&sfw_data.fw_lock);
1264
1265 #ifdef __KERNEL__
1266         if (!sfw_data.fw_shuttingdown)
1267                 sfw_add_session_timer();
1268 #else
1269         LASSERT (!sfw_data.fw_shuttingdown);
1270         sfw_add_session_timer();
1271 #endif
1272
1273         sfw_data.fw_active_srpc = NULL;
1274         spin_unlock(&sfw_data.fw_lock);
1275         return rc;
1276 }
1277
1278 int
1279 sfw_bulk_ready (srpc_server_rpc_t *rpc, int status)
1280 {
1281         srpc_service_t *sv = rpc->srpc_service;
1282         int             rc;
1283
1284         LASSERT (rpc->srpc_bulk != NULL);
1285         LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1286         LASSERT (sfw_data.fw_active_srpc == NULL);
1287         LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1288
1289         spin_lock(&sfw_data.fw_lock);
1290
1291         if (status != 0) {
1292                 CERROR ("Bulk transfer failed for RPC: "
1293                         "service %s, peer %s, status %d\n",
1294                         sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1295                 spin_unlock(&sfw_data.fw_lock);
1296                 return -EIO;
1297         }
1298
1299         if (sfw_data.fw_shuttingdown) {
1300                 spin_unlock(&sfw_data.fw_lock);
1301                 return -ESHUTDOWN;
1302         }
1303
1304         if (sfw_del_session_timer() != 0) {
1305                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1306                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1307                 spin_unlock(&sfw_data.fw_lock);
1308                 return -EAGAIN;
1309         }
1310
1311         sfw_data.fw_active_srpc = rpc;
1312         spin_unlock(&sfw_data.fw_lock);
1313
1314         rc = sfw_add_test(rpc);
1315
1316         spin_lock(&sfw_data.fw_lock);
1317
1318 #ifdef __KERNEL__
1319         if (!sfw_data.fw_shuttingdown)
1320                 sfw_add_session_timer();
1321 #else
1322         LASSERT (!sfw_data.fw_shuttingdown);
1323         sfw_add_session_timer();
1324 #endif
1325
1326         sfw_data.fw_active_srpc = NULL;
1327         spin_unlock(&sfw_data.fw_lock);
1328         return rc;
1329 }
1330
1331 srpc_client_rpc_t *
1332 sfw_create_rpc (lnet_process_id_t peer, int service,
1333                 int nbulkiov, int bulklen,
1334                 void (*done) (srpc_client_rpc_t *), void *priv)
1335 {
1336         srpc_client_rpc_t *rpc;
1337
1338         spin_lock(&sfw_data.fw_lock);
1339
1340         LASSERT (!sfw_data.fw_shuttingdown);
1341         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1342
1343         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1344                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1345                                  srpc_client_rpc_t, crpc_list);
1346                 list_del(&rpc->crpc_list);
1347                 spin_unlock(&sfw_data.fw_lock);
1348
1349                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1350                                      done, sfw_client_rpc_fini, priv);
1351                 return rpc;
1352         }
1353
1354         spin_unlock(&sfw_data.fw_lock);
1355
1356         rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1357                                      nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1358                                      priv);
1359         return rpc;
1360 }
1361
1362 void
1363 sfw_unpack_message (srpc_msg_t *msg)
1364 {
1365         if (msg->msg_magic == SRPC_MSG_MAGIC)
1366                 return; /* no flipping needed */
1367
1368         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1369
1370         __swab32s(&msg->msg_type);
1371
1372         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1373                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1374
1375                 __swab32s(&req->str_type);
1376                 __swab64s(&req->str_rpyid);
1377                 sfw_unpack_sid(req->str_sid);
1378                 return;
1379         }
1380
1381         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1382                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1383
1384                 __swab32s(&rep->str_status);
1385                 sfw_unpack_sid(rep->str_sid);
1386                 sfw_unpack_fw_counters(rep->str_fw);
1387                 sfw_unpack_rpc_counters(rep->str_rpc);
1388                 sfw_unpack_lnet_counters(rep->str_lnet);
1389                 return;
1390         }
1391
1392         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1393                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1394
1395                 __swab64s(&req->mksn_rpyid);
1396                 __swab32s(&req->mksn_force);
1397                 sfw_unpack_sid(req->mksn_sid);
1398                 return;
1399         }
1400
1401         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1402                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1403
1404                 __swab32s(&rep->mksn_status);
1405                 __swab32s(&rep->mksn_timeout);
1406                 sfw_unpack_sid(rep->mksn_sid);
1407                 return;
1408         }
1409
1410         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1411                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1412
1413                 __swab64s(&req->rmsn_rpyid);
1414                 sfw_unpack_sid(req->rmsn_sid);
1415                 return;
1416         }
1417
1418         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1419                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1420
1421                 __swab32s(&rep->rmsn_status);
1422                 sfw_unpack_sid(rep->rmsn_sid);
1423                 return;
1424         }
1425
1426         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1427                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1428
1429                 __swab64s(&req->dbg_rpyid);
1430                 __swab32s(&req->dbg_flags);
1431                 sfw_unpack_sid(req->dbg_sid);
1432                 return;
1433         }
1434
1435         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1436                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1437
1438                 __swab32s(&rep->dbg_nbatch);
1439                 __swab32s(&rep->dbg_timeout);
1440                 sfw_unpack_sid(rep->dbg_sid);
1441                 return;
1442         }
1443
1444         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1445                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1446
1447                 __swab32s(&req->bar_opc);
1448                 __swab64s(&req->bar_rpyid);
1449                 __swab32s(&req->bar_testidx);
1450                 __swab32s(&req->bar_arg);
1451                 sfw_unpack_sid(req->bar_sid);
1452                 __swab64s(&req->bar_bid.bat_id);
1453                 return;
1454         }
1455
1456         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1457                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1458
1459                 __swab32s(&rep->bar_status);
1460                 sfw_unpack_sid(rep->bar_sid);
1461                 return;
1462         }
1463
1464         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1465                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1466
1467                 __swab64s(&req->tsr_rpyid);
1468                 __swab64s(&req->tsr_bulkid);
1469                 __swab32s(&req->tsr_loop);
1470                 __swab32s(&req->tsr_ndest);
1471                 __swab32s(&req->tsr_concur);
1472                 __swab32s(&req->tsr_service);
1473                 sfw_unpack_sid(req->tsr_sid);
1474                 __swab64s(&req->tsr_bid.bat_id);
1475                 return;
1476         }
1477
1478         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1479                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1480
1481                 __swab32s(&rep->tsr_status);
1482                 sfw_unpack_sid(rep->tsr_sid);
1483                 return;
1484         }
1485
1486         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1487                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1488
1489                 __swab64s(&req->join_rpyid);
1490                 sfw_unpack_sid(req->join_sid);
1491                 return;
1492         }
1493
1494         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1495                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1496
1497                 __swab32s(&rep->join_status);
1498                 __swab32s(&rep->join_timeout);
1499                 sfw_unpack_sid(rep->join_sid);
1500                 return;
1501         }
1502
1503         LBUG ();
1504         return;
1505 }
1506
1507 void
1508 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1509 {
1510         LASSERT (atomic_read(&rpc->crpc_refcount) > 0);
1511         LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1512
1513         spin_lock(&rpc->crpc_lock);
1514         srpc_abort_rpc(rpc, -EINTR);
1515         spin_unlock(&rpc->crpc_lock);
1516         return;
1517 }
1518
1519 void
1520 sfw_post_rpc (srpc_client_rpc_t *rpc)
1521 {
1522         spin_lock(&rpc->crpc_lock);
1523
1524         LASSERT (!rpc->crpc_closed);
1525         LASSERT (!rpc->crpc_aborted);
1526         LASSERT (list_empty(&rpc->crpc_list));
1527         LASSERT (!sfw_data.fw_shuttingdown);
1528
1529         rpc->crpc_timeout = SFW_CLIENT_RPC_TIMEOUT;
1530         srpc_post_rpc(rpc);
1531
1532         spin_unlock(&rpc->crpc_lock);
1533         return;
1534 }
1535
1536 static srpc_service_t sfw_services[] =
1537 {
1538         {
1539                 /* sv_id */    SRPC_SERVICE_DEBUG,
1540                 /* sv_name */  "debug",
1541                 0
1542         },
1543         {
1544                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1545                 /* sv_name */  "query stats",
1546                 0
1547         },
1548         {
1549                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1550                 /* sv_name */  "make session",
1551                 0
1552         },
1553         {
1554                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1555                 /* sv_name */  "remove session",
1556                 0
1557         },
1558         {
1559                 /* sv_id */    SRPC_SERVICE_BATCH,
1560                 /* sv_name */  "batch service",
1561                 0
1562         },
1563         {
1564                 /* sv_id */    SRPC_SERVICE_TEST,
1565                 /* sv_name */  "test service",
1566                 0
1567         },
1568         {
1569                 /* sv_id */    0,
1570                 /* sv_name */  NULL,
1571                 0
1572         }
1573 };
1574
1575 extern sfw_test_client_ops_t ping_test_client;
1576 extern srpc_service_t        ping_test_service;
1577 extern void ping_init_test_client(void);
1578 extern void ping_init_test_service(void);
1579
1580 extern sfw_test_client_ops_t brw_test_client;
1581 extern srpc_service_t        brw_test_service;
1582 extern void brw_init_test_client(void);
1583 extern void brw_init_test_service(void);
1584
1585
1586 int
1587 sfw_startup (void)
1588 {
1589         int              i;
1590         int              rc;
1591         int              error;
1592         srpc_service_t  *sv;
1593         sfw_test_case_t *tsc;
1594
1595 #ifndef __KERNEL__
1596         char *s;
1597
1598         s = getenv("SESSION_TIMEOUT");
1599         session_timeout = s != NULL ? atoi(s) : session_timeout;
1600
1601         s = getenv("BRW_INJECT_ERRORS");
1602         brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
1603 #endif
1604
1605         if (session_timeout < 0) {
1606                 CERROR ("Session timeout must be non-negative: %d\n",
1607                         session_timeout);
1608                 return -EINVAL;
1609         }
1610
1611         if (session_timeout == 0)
1612                 CWARN ("Zero session_timeout specified "
1613                        "- test sessions never expire.\n");
1614
1615         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1616
1617         sfw_data.fw_session     = NULL;
1618         sfw_data.fw_active_srpc = NULL;
1619         spin_lock_init(&sfw_data.fw_lock);
1620         atomic_set(&sfw_data.fw_nzombies, 0);
1621         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1622         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1623         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1624
1625         brw_init_test_client();
1626         brw_init_test_service();
1627         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1628         LASSERT (rc == 0);
1629
1630         ping_init_test_client();
1631         ping_init_test_service();
1632         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1633         LASSERT (rc == 0);
1634
1635         error = 0;
1636         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1637                                        sfw_test_case_t, tsc_list) {
1638                 sv = tsc->tsc_srv_service;
1639                 sv->sv_concur = SFW_TEST_CONCURRENCY;
1640
1641                 rc = srpc_add_service(sv);
1642                 LASSERT (rc != -EBUSY);
1643                 if (rc != 0) {
1644                         CWARN ("Failed to add %s service: %d\n",
1645                                sv->sv_name, rc);
1646                         error = rc;
1647                 }
1648         }
1649
1650         for (i = 0; ; i++) {
1651                 sv = &sfw_services[i];
1652                 if (sv->sv_name == NULL) break;
1653
1654                 sv->sv_bulk_ready = NULL;
1655                 sv->sv_handler    = sfw_handle_server_rpc;
1656                 sv->sv_concur     = SFW_SERVICE_CONCURRENCY;
1657                 if (sv->sv_id == SRPC_SERVICE_TEST)
1658                         sv->sv_bulk_ready = sfw_bulk_ready;
1659
1660                 rc = srpc_add_service(sv);
1661                 LASSERT (rc != -EBUSY);
1662                 if (rc != 0) {
1663                         CWARN ("Failed to add %s service: %d\n",
1664                                sv->sv_name, rc);
1665                         error = rc;
1666                 }
1667
1668                 /* about to sfw_shutdown, no need to add buffer */
1669                 if (error) continue;
1670
1671                 rc = srpc_service_add_buffers(sv, SFW_POST_BUFFERS);
1672                 if (rc != SFW_POST_BUFFERS) {
1673                         CWARN ("Failed to reserve enough buffers: "
1674                                "service %s, %d needed, %d reserved\n",
1675                                sv->sv_name, SFW_POST_BUFFERS, rc);
1676                         error = -ENOMEM;
1677                 }
1678         }
1679
1680         if (error != 0)
1681                 sfw_shutdown();
1682         return error;
1683 }
1684
1685 void
1686 sfw_shutdown (void)
1687 {
1688         srpc_service_t  *sv;
1689         sfw_test_case_t *tsc;
1690         int              i;
1691
1692         spin_lock(&sfw_data.fw_lock);
1693
1694         sfw_data.fw_shuttingdown = 1;
1695 #ifdef __KERNEL__
1696         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1697                        "waiting for active RPC to finish.\n");
1698 #else
1699         LASSERT (sfw_data.fw_active_srpc == NULL);
1700 #endif
1701
1702         if (sfw_del_session_timer() != 0)
1703                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1704                                "waiting for session timer to explode.\n");
1705
1706         sfw_deactivate_session();
1707         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1708                        sfw_data.fw_lock,
1709                        "waiting for %d zombie sessions to die.\n",
1710                        atomic_read(&sfw_data.fw_nzombies));
1711
1712         spin_unlock(&sfw_data.fw_lock);
1713
1714         for (i = 0; ; i++) {
1715                 sv = &sfw_services[i];
1716                 if (sv->sv_name == NULL)
1717                         break;
1718
1719                 srpc_shutdown_service(sv);
1720                 srpc_remove_service(sv);
1721         }
1722
1723         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1724                                        sfw_test_case_t, tsc_list) {
1725                 sv = tsc->tsc_srv_service;
1726                 srpc_shutdown_service(sv);
1727                 srpc_remove_service(sv);
1728         }
1729
1730         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1731                 srpc_client_rpc_t *rpc;
1732
1733                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1734                                  srpc_client_rpc_t, crpc_list);
1735                 list_del(&rpc->crpc_list);
1736
1737                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1738         }
1739
1740         for (i = 0; ; i++) {
1741                 sv = &sfw_services[i];
1742                 if (sv->sv_name == NULL)
1743                         break;
1744
1745                 srpc_wait_service_shutdown(sv);
1746         }
1747
1748         while (!list_empty(&sfw_data.fw_tests)) {
1749                 tsc = list_entry(sfw_data.fw_tests.next,
1750                                  sfw_test_case_t, tsc_list);
1751
1752                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1753
1754                 list_del(&tsc->tsc_list);
1755                 LIBCFS_FREE(tsc, sizeof(*tsc));
1756         }
1757
1758         return;
1759 }