lnet/selftest/framework.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  */
  30 /*
  31  * This file is part of Lustre, http://www.lustre.org/
  32  * Lustre is a trademark of Sun Microsystems, Inc.
  33  *
  34  * lnet/selftest/framework.c
  35  *
  36  * Author: Isaac Huang <isaac@clusterfs.com>
  37  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_LNET
  41
  42 #include "selftest.h"
  43
  44 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
  45
  46 static int session_timeout = 100;
  47 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  48                 "test session timeout in seconds (100 by default, 0 == never)");
  49
  50 static int rpc_timeout = 64;
  51 CFS_MODULE_PARM(rpc_timeout, "i", int, 0644,
  52                 "rpc timeout in seconds (64 by default, 0 == never)");
  53
  54 #define sfw_unpack_id(id)               \
  55 do {                                    \
  56         __swab64s(&(id).nid);           \
  57         __swab32s(&(id).pid);           \
  58 } while (0)
  59
  60 #define sfw_unpack_sid(sid)             \
  61 do {                                    \
  62         __swab64s(&(sid).ses_nid);      \
  63         __swab64s(&(sid).ses_stamp);    \
  64 } while (0)
  65
  66 #define sfw_unpack_fw_counters(fc)        \
  67 do {                                      \
  68         __swab32s(&(fc).running_ms);      \
  69         __swab32s(&(fc).active_batches);  \
  70         __swab32s(&(fc).zombie_sessions); \
  71         __swab32s(&(fc).brw_errors);      \
  72         __swab32s(&(fc).ping_errors);     \
  73 } while (0)
  74
  75 #define sfw_unpack_rpc_counters(rc)     \
  76 do {                                    \
  77         __swab32s(&(rc).errors);        \
  78         __swab32s(&(rc).rpcs_sent);     \
  79         __swab32s(&(rc).rpcs_rcvd);     \
  80         __swab32s(&(rc).rpcs_dropped);  \
  81         __swab32s(&(rc).rpcs_expired);  \
  82         __swab64s(&(rc).bulk_get);      \
  83         __swab64s(&(rc).bulk_put);      \
  84 } while (0)
  85
  86 #define sfw_unpack_lnet_counters(lc)    \
  87 do {                                    \
  88         __swab32s(&(lc).errors);        \
  89         __swab32s(&(lc).msgs_max);      \
  90         __swab32s(&(lc).msgs_alloc);    \
  91         __swab32s(&(lc).send_count);    \
  92         __swab32s(&(lc).recv_count);    \
  93         __swab32s(&(lc).drop_count);    \
  94         __swab32s(&(lc).route_count);   \
  95         __swab64s(&(lc).send_length);   \
  96         __swab64s(&(lc).recv_length);   \
  97         __swab64s(&(lc).drop_length);   \
  98         __swab64s(&(lc).route_length);  \
  99 } while (0)
 100
 101 #define sfw_test_active(t)      (cfs_atomic_read(&(t)->tsi_nactive) != 0)
 102 #define sfw_batch_active(b)     (cfs_atomic_read(&(b)->bat_nactive) != 0)
 103
 104 struct smoketest_framework {
 105         cfs_list_t         fw_zombie_rpcs;     /* RPCs to be recycled */
 106         cfs_list_t         fw_zombie_sessions; /* stopping sessions */
 107         cfs_list_t         fw_tests;           /* registered test cases */
 108         cfs_atomic_t       fw_nzombies;        /* # zombie sessions */
 109         cfs_spinlock_t     fw_lock;            /* serialise */
 110         sfw_session_t     *fw_session;         /* _the_ session */
 111         int                fw_shuttingdown;    /* shutdown in progress */
 112         srpc_server_rpc_t *fw_active_srpc;     /* running RPC */
 113 } sfw_data;
 114
 115 /* forward ref's */
 116 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 117 void sfw_destroy_session (sfw_session_t *sn);
 118
 119 static inline sfw_test_case_t *
 120 sfw_find_test_case(int id)
 121 {
 122         sfw_test_case_t *tsc;
 123
 124         LASSERT (id <= SRPC_SERVICE_MAX_ID);
 125         LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 126
 127         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 128                                        sfw_test_case_t, tsc_list) {
 129                 if (tsc->tsc_srv_service->sv_id == id)
 130                         return tsc;
 131         }
 132
 133         return NULL;
 134 }
 135
 136 static int
 137 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 138 {
 139         sfw_test_case_t *tsc;
 140
 141         if (sfw_find_test_case(service->sv_id) != NULL) {
 142                 CERROR ("Failed to register test %s (%d)\n",
 143                         service->sv_name, service->sv_id);
 144                 return -EEXIST;
 145         }
 146
 147         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 148         if (tsc == NULL)
 149                 return -ENOMEM;
 150
 151         memset(tsc, 0, sizeof(sfw_test_case_t));
 152         tsc->tsc_cli_ops     = cliops;
 153         tsc->tsc_srv_service = service;
 154
 155         cfs_list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 156         return 0;
 157 }
 158
 159 void
 160 sfw_add_session_timer (void)
 161 {
 162         sfw_session_t *sn = sfw_data.fw_session;
 163         stt_timer_t   *timer = &sn->sn_timer;
 164
 165         LASSERT (!sfw_data.fw_shuttingdown);
 166
 167         if (sn == NULL || sn->sn_timeout == 0)
 168                 return;
 169
 170         LASSERT (!sn->sn_timer_active);
 171
 172         sn->sn_timer_active = 1;
 173         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 174                                           cfs_time_current_sec());
 175         stt_add_timer(timer);
 176         return;
 177 }
 178
 179 int
 180 sfw_del_session_timer (void)
 181 {
 182         sfw_session_t *sn = sfw_data.fw_session;
 183
 184         if (sn == NULL || !sn->sn_timer_active)
 185                 return 0;
 186
 187         LASSERT (sn->sn_timeout != 0);
 188
 189         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 190                 sn->sn_timer_active = 0;
 191                 return 0;
 192         }
 193
 194 #ifndef __KERNEL__
 195         /* Racing is impossible in single-threaded userland selftest */
 196         LBUG();
 197 #endif
 198         return EBUSY; /* racing with sfw_session_expired() */
 199 }
 200
 201 /* called with sfw_data.fw_lock held */
 202 static void
 203 sfw_deactivate_session (void)
 204 {
 205         sfw_session_t *sn = sfw_data.fw_session;
 206         int            nactive = 0;
 207         sfw_batch_t   *tsb;
 208         sfw_test_case_t *tsc;
 209
 210         if (sn == NULL) return;
 211
 212         LASSERT (!sn->sn_timer_active);
 213
 214         sfw_data.fw_session = NULL;
 215         cfs_atomic_inc(&sfw_data.fw_nzombies);
 216         cfs_list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 217
 218         cfs_spin_unlock(&sfw_data.fw_lock);
 219
 220         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 221                                        sfw_test_case_t, tsc_list) {
 222                 srpc_abort_service(tsc->tsc_srv_service);
 223         }
 224
 225         cfs_spin_lock(&sfw_data.fw_lock);
 226
 227         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 228                                        sfw_batch_t, bat_list) {
 229                 if (sfw_batch_active(tsb)) {
 230                         nactive++;
 231                         sfw_stop_batch(tsb, 1);
 232                 }
 233         }
 234
 235         if (nactive != 0)
 236                 return;   /* wait for active batches to stop */
 237
 238         cfs_list_del_init(&sn->sn_list);
 239         cfs_spin_unlock(&sfw_data.fw_lock);
 240
 241         sfw_destroy_session(sn);
 242
 243         cfs_spin_lock(&sfw_data.fw_lock);
 244         return;
 245 }
 246
 247 #ifndef __KERNEL__
 248
 249 int
 250 sfw_session_removed (void)
 251 {
 252         return (sfw_data.fw_session == NULL) ? 1 : 0;
 253 }
 254
 255 #endif
 256
 257 void
 258 sfw_session_expired (void *data)
 259 {
 260         sfw_session_t *sn = data;
 261
 262         cfs_spin_lock(&sfw_data.fw_lock);
 263
 264         LASSERT (sn->sn_timer_active);
 265         LASSERT (sn == sfw_data.fw_session);
 266
 267         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 268                libcfs_nid2str(sn->sn_id.ses_nid),
 269                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 270
 271         sn->sn_timer_active = 0;
 272         sfw_deactivate_session();
 273
 274         cfs_spin_unlock(&sfw_data.fw_lock);
 275         return;
 276 }
 277
 278 static inline void
 279 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
 280 {
 281         stt_timer_t *timer = &sn->sn_timer;
 282
 283         memset(sn, 0, sizeof(sfw_session_t));
 284         CFS_INIT_LIST_HEAD(&sn->sn_list);
 285         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 286         cfs_atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 287         cfs_atomic_set(&sn->sn_brw_errors, 0);
 288         cfs_atomic_set(&sn->sn_ping_errors, 0);
 289         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 290
 291         sn->sn_timer_active = 0;
 292         sn->sn_id           = sid;
 293         sn->sn_timeout      = session_timeout;
 294         sn->sn_started      = cfs_time_current();
 295
 296         timer->stt_data = sn;
 297         timer->stt_func = sfw_session_expired;
 298         CFS_INIT_LIST_HEAD(&timer->stt_list);
 299 }
 300
 301 /* completion handler for incoming framework RPCs */
 302 void
 303 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 304 {
 305         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 306         int                     status  = rpc->srpc_status;
 307
 308         CDEBUG (D_NET,
 309                 "Incoming framework RPC done: "
 310                 "service %s, peer %s, status %s:%d\n",
 311                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 312                 swi_state2str(rpc->srpc_wi.swi_state),
 313                 status);
 314
 315         if (rpc->srpc_bulk != NULL)
 316                 sfw_free_pages(rpc);
 317         return;
 318 }
 319
 320 void
 321 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 322 {
 323         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 324         LASSERT (cfs_list_empty(&rpc->crpc_list));
 325         LASSERT (cfs_atomic_read(&rpc->crpc_refcount) == 0);
 326 #ifndef __KERNEL__
 327         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 328 #endif
 329
 330         CDEBUG (D_NET,
 331                 "Outgoing framework RPC done: "
 332                 "service %d, peer %s, status %s:%d:%d\n",
 333                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 334                 swi_state2str(rpc->crpc_wi.swi_state),
 335                 rpc->crpc_aborted, rpc->crpc_status);
 336
 337         cfs_spin_lock(&sfw_data.fw_lock);
 338
 339         /* my callers must finish all RPCs before shutting me down */
 340         LASSERT (!sfw_data.fw_shuttingdown);
 341         cfs_list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 342
 343         cfs_spin_unlock(&sfw_data.fw_lock);
 344         return;
 345 }
 346
 347 sfw_batch_t *
 348 sfw_find_batch (lst_bid_t bid)
 349 {
 350         sfw_session_t *sn = sfw_data.fw_session;
 351         sfw_batch_t   *bat;
 352
 353         LASSERT (sn != NULL);
 354
 355         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 356                                        sfw_batch_t, bat_list) {
 357                 if (bat->bat_id.bat_id == bid.bat_id)
 358                         return bat;
 359         }
 360
 361         return NULL;
 362 }
 363
 364 sfw_batch_t *
 365 sfw_bid2batch (lst_bid_t bid)
 366 {
 367         sfw_session_t *sn = sfw_data.fw_session;
 368         sfw_batch_t   *bat;
 369
 370         LASSERT (sn != NULL);
 371
 372         bat = sfw_find_batch(bid);
 373         if (bat != NULL)
 374                 return bat;
 375
 376         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 377         if (bat == NULL)
 378                 return NULL;
 379
 380         bat->bat_error    = 0;
 381         bat->bat_session  = sn;
 382         bat->bat_id       = bid;
 383         cfs_atomic_set(&bat->bat_nactive, 0);
 384         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 385
 386         cfs_list_add_tail(&bat->bat_list, &sn->sn_batches);
 387         return bat;
 388 }
 389
 390 int
 391 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 392 {
 393         sfw_session_t  *sn = sfw_data.fw_session;
 394         sfw_counters_t *cnt = &reply->str_fw;
 395         sfw_batch_t    *bat;
 396         struct timeval  tv;
 397
 398         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 399
 400         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 401                 reply->str_status = EINVAL;
 402                 return 0;
 403         }
 404
 405         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 406                 reply->str_status = ESRCH;
 407                 return 0;
 408         }
 409
 410         lnet_counters_get(&reply->str_lnet);
 411         srpc_get_counters(&reply->str_rpc);
 412
 413         /* send over the msecs since the session was started
 414          - with 32 bits to send, this is ~49 days */
 415         cfs_duration_usec(cfs_time_sub(cfs_time_current(),
 416                                        sn->sn_started), &tv);
 417
 418         cnt->running_ms      = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
 419         cnt->brw_errors      = cfs_atomic_read(&sn->sn_brw_errors);
 420         cnt->ping_errors     = cfs_atomic_read(&sn->sn_ping_errors);
 421         cnt->zombie_sessions = cfs_atomic_read(&sfw_data.fw_nzombies);
 422
 423         cnt->active_batches = 0;
 424         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 425                                        sfw_batch_t, bat_list) {
 426                 if (cfs_atomic_read(&bat->bat_nactive) > 0)
 427                         cnt->active_batches++;
 428         }
 429
 430         reply->str_status = 0;
 431         return 0;
 432 }
 433
 434 int
 435 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 436 {
 437         sfw_session_t *sn = sfw_data.fw_session;
 438
 439         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 440                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 441                 reply->mksn_status = EINVAL;
 442                 return 0;
 443         }
 444
 445         if (sn != NULL) {
 446                 reply->mksn_status  = 0;
 447                 reply->mksn_sid     = sn->sn_id;
 448                 reply->mksn_timeout = sn->sn_timeout;
 449
 450                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 451                         cfs_atomic_inc(&sn->sn_refcount);
 452                         return 0;
 453                 }
 454
 455                 if (!request->mksn_force) {
 456                         reply->mksn_status = EBUSY;
 457                         strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 458                         return 0;
 459                 }
 460         }
 461
 462         /* brand new or create by force */
 463         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 464         if (sn == NULL) {
 465                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 466                 return -ENOMEM;
 467         }
 468
 469         sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
 470
 471         cfs_spin_lock(&sfw_data.fw_lock);
 472
 473         sfw_deactivate_session();
 474         LASSERT (sfw_data.fw_session == NULL);
 475         sfw_data.fw_session = sn;
 476
 477         cfs_spin_unlock(&sfw_data.fw_lock);
 478
 479         reply->mksn_status  = 0;
 480         reply->mksn_sid     = sn->sn_id;
 481         reply->mksn_timeout = sn->sn_timeout;
 482         return 0;
 483 }
 484
 485 int
 486 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 487 {
 488         sfw_session_t *sn = sfw_data.fw_session;
 489
 490         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 491
 492         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 493                 reply->rmsn_status = EINVAL;
 494                 return 0;
 495         }
 496
 497         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 498                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 499                 return 0;
 500         }
 501
 502         if (!cfs_atomic_dec_and_test(&sn->sn_refcount)) {
 503                 reply->rmsn_status = 0;
 504                 return 0;
 505         }
 506
 507         cfs_spin_lock(&sfw_data.fw_lock);
 508         sfw_deactivate_session();
 509         cfs_spin_unlock(&sfw_data.fw_lock);
 510
 511         reply->rmsn_status = 0;
 512         reply->rmsn_sid    = LST_INVALID_SID;
 513         LASSERT (sfw_data.fw_session == NULL);
 514         return 0;
 515 }
 516
 517 int
 518 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 519 {
 520         sfw_session_t *sn = sfw_data.fw_session;
 521
 522         if (sn == NULL) {
 523                 reply->dbg_status = ESRCH;
 524                 reply->dbg_sid    = LST_INVALID_SID;
 525                 return 0;
 526         }
 527
 528         reply->dbg_status  = 0;
 529         reply->dbg_sid     = sn->sn_id;
 530         reply->dbg_timeout = sn->sn_timeout;
 531         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 532
 533         return 0;
 534 }
 535
 536 void
 537 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 538 {
 539         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 540         sfw_test_instance_t *tsi = tsu->tsu_instance;
 541
 542         /* Called with hold of tsi->tsi_lock */
 543         LASSERT (cfs_list_empty(&rpc->crpc_list));
 544         cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 545 }
 546
 547 static inline int
 548 sfw_test_buffers(sfw_test_instance_t *tsi)
 549 {
 550         struct sfw_test_case    *tsc = sfw_find_test_case(tsi->tsi_service);
 551         struct srpc_service     *svc = tsc->tsc_srv_service;
 552         int                     nbuf;
 553
 554         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 555         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 556 }
 557
 558 int
 559 sfw_load_test(struct sfw_test_instance *tsi)
 560 {
 561         struct sfw_test_case    *tsc = sfw_find_test_case(tsi->tsi_service);
 562         struct srpc_service     *svc = tsc->tsc_srv_service;
 563         int                     nbuf = sfw_test_buffers(tsi);
 564         int                     rc;
 565
 566         LASSERT(tsc != NULL);
 567
 568         if (tsi->tsi_is_client) {
 569                 tsi->tsi_ops = tsc->tsc_cli_ops;
 570                 return 0;
 571         }
 572
 573         rc = srpc_service_add_buffers(svc, nbuf);
 574         if (rc != 0) {
 575                 CWARN("Failed to reserve enough buffers: "
 576                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 577                 /* NB: this error handler is not strictly correct, because
 578                  * it may release more buffers than already allocated,
 579                  * but it doesn't matter because request portal should
 580                  * be lazy portal and will grow buffers if necessary. */
 581                 srpc_service_remove_buffers(svc, nbuf);
 582                 return -ENOMEM;
 583         }
 584
 585         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 586                nbuf * (srpc_serv_is_framework(svc) ?
 587                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 588         return 0;
 589 }
 590
 591 void
 592 sfw_unload_test(struct sfw_test_instance *tsi)
 593 {
 594         struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
 595
 596         LASSERT(tsc != NULL);
 597
 598         if (tsi->tsi_is_client)
 599                 return;
 600
 601         /* shrink buffers, because request portal is lazy portal
 602          * which can grow buffers at runtime so we may leave
 603          * some buffers behind, but never mind... */
 604         srpc_service_remove_buffers(tsc->tsc_srv_service,
 605                                     sfw_test_buffers(tsi));
 606         return;
 607 }
 608
 609 void
 610 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 611 {
 612         srpc_client_rpc_t *rpc;
 613         sfw_test_unit_t   *tsu;
 614
 615         if (!tsi->tsi_is_client) goto clean;
 616
 617         tsi->tsi_ops->tso_fini(tsi);
 618
 619         LASSERT (!tsi->tsi_stopping);
 620         LASSERT (cfs_list_empty(&tsi->tsi_active_rpcs));
 621         LASSERT (!sfw_test_active(tsi));
 622
 623         while (!cfs_list_empty(&tsi->tsi_units)) {
 624                 tsu = cfs_list_entry(tsi->tsi_units.next,
 625                                      sfw_test_unit_t, tsu_list);
 626                 cfs_list_del(&tsu->tsu_list);
 627                 LIBCFS_FREE(tsu, sizeof(*tsu));
 628         }
 629
 630         while (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 631                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 632                                      srpc_client_rpc_t, crpc_list);
 633                 cfs_list_del(&rpc->crpc_list);
 634                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 635         }
 636
 637 clean:
 638         sfw_unload_test(tsi);
 639         LIBCFS_FREE(tsi, sizeof(*tsi));
 640         return;
 641 }
 642
 643 void
 644 sfw_destroy_batch (sfw_batch_t *tsb)
 645 {
 646         sfw_test_instance_t *tsi;
 647
 648         LASSERT (!sfw_batch_active(tsb));
 649         LASSERT (cfs_list_empty(&tsb->bat_list));
 650
 651         while (!cfs_list_empty(&tsb->bat_tests)) {
 652                 tsi = cfs_list_entry(tsb->bat_tests.next,
 653                                      sfw_test_instance_t, tsi_list);
 654                 cfs_list_del_init(&tsi->tsi_list);
 655                 sfw_destroy_test_instance(tsi);
 656         }
 657
 658         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 659         return;
 660 }
 661
 662 void
 663 sfw_destroy_session (sfw_session_t *sn)
 664 {
 665         sfw_batch_t *batch;
 666
 667         LASSERT (cfs_list_empty(&sn->sn_list));
 668         LASSERT (sn != sfw_data.fw_session);
 669
 670         while (!cfs_list_empty(&sn->sn_batches)) {
 671                 batch = cfs_list_entry(sn->sn_batches.next,
 672                                        sfw_batch_t, bat_list);
 673                 cfs_list_del_init(&batch->bat_list);
 674                 sfw_destroy_batch(batch);
 675         }
 676
 677         LIBCFS_FREE(sn, sizeof(*sn));
 678         cfs_atomic_dec(&sfw_data.fw_nzombies);
 679         return;
 680 }
 681
 682 void
 683 sfw_unpack_test_req (srpc_msg_t *msg)
 684 {
 685         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 686
 687         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 688         LASSERT (req->tsr_is_client);
 689
 690         if (msg->msg_magic == SRPC_MSG_MAGIC)
 691                 return; /* no flipping needed */
 692
 693         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 694
 695         if (req->tsr_service == SRPC_SERVICE_BRW) {
 696                 test_bulk_req_t *bulk = &req->tsr_u.bulk;
 697
 698                 __swab32s(&bulk->blk_opc);
 699                 __swab32s(&bulk->blk_npg);
 700                 __swab32s(&bulk->blk_flags);
 701                 return;
 702         }
 703
 704         if (req->tsr_service == SRPC_SERVICE_PING) {
 705                 test_ping_req_t *ping = &req->tsr_u.ping;
 706
 707                 __swab32s(&ping->png_size);
 708                 __swab32s(&ping->png_flags);
 709                 return;
 710         }
 711
 712         LBUG ();
 713         return;
 714 }
 715
 716 int
 717 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 718 {
 719         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 720         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 721         srpc_bulk_t         *bk = rpc->srpc_bulk;
 722         int                  ndest = req->tsr_ndest;
 723         sfw_test_unit_t     *tsu;
 724         sfw_test_instance_t *tsi;
 725         int                  i;
 726         int                  rc;
 727
 728         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 729         if (tsi == NULL) {
 730                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 731                         tsb->bat_id.bat_id);
 732                 return -ENOMEM;
 733         }
 734
 735         memset(tsi, 0, sizeof(*tsi));
 736         cfs_spin_lock_init(&tsi->tsi_lock);
 737         cfs_atomic_set(&tsi->tsi_nactive, 0);
 738         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 739         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 740         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 741
 742         tsi->tsi_stopping      = 0;
 743         tsi->tsi_batch         = tsb;
 744         tsi->tsi_loop          = req->tsr_loop;
 745         tsi->tsi_concur        = req->tsr_concur;
 746         tsi->tsi_service       = req->tsr_service;
 747         tsi->tsi_is_client     = !!(req->tsr_is_client);
 748         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 749
 750         rc = sfw_load_test(tsi);
 751         if (rc != 0) {
 752                 LIBCFS_FREE(tsi, sizeof(*tsi));
 753                 return rc;
 754         }
 755
 756         LASSERT (!sfw_batch_active(tsb));
 757
 758         if (!tsi->tsi_is_client) {
 759                 /* it's test server, just add it to tsb */
 760                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 761                 return 0;
 762         }
 763
 764         LASSERT (bk != NULL);
 765 #ifndef __KERNEL__
 766         LASSERT (bk->bk_pages != NULL);
 767 #endif
 768         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 769         LASSERT ((unsigned int)bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
 770
 771         sfw_unpack_test_req(msg);
 772         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 773
 774         for (i = 0; i < ndest; i++) {
 775                 lnet_process_id_packed_t *dests;
 776                 lnet_process_id_packed_t  id;
 777                 int                       j;
 778
 779 #ifdef __KERNEL__
 780                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 781                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 782 #else
 783                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 784 #endif
 785                 id = dests[i % SFW_ID_PER_PAGE];
 786                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 787                         sfw_unpack_id(id);
 788
 789                 for (j = 0; j < tsi->tsi_concur; j++) {
 790                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 791                         if (tsu == NULL) {
 792                                 rc = -ENOMEM;
 793                                 CERROR ("Can't allocate tsu for %d\n",
 794                                         tsi->tsi_service);
 795                                 goto error;
 796                         }
 797
 798                         tsu->tsu_dest.nid = id.nid;
 799                         tsu->tsu_dest.pid = id.pid;
 800                         tsu->tsu_instance = tsi;
 801                         tsu->tsu_private  = NULL;
 802                         cfs_list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 803                 }
 804         }
 805
 806         rc = tsi->tsi_ops->tso_init(tsi);
 807         if (rc == 0) {
 808                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 809                 return 0;
 810         }
 811
 812 error:
 813         LASSERT (rc != 0);
 814         sfw_destroy_test_instance(tsi);
 815         return rc;
 816 }
 817
 818 static void
 819 sfw_test_unit_done (sfw_test_unit_t *tsu)
 820 {
 821         sfw_test_instance_t *tsi = tsu->tsu_instance;
 822         sfw_batch_t         *tsb = tsi->tsi_batch;
 823         sfw_session_t       *sn = tsb->bat_session;
 824
 825         LASSERT (sfw_test_active(tsi));
 826
 827         if (!cfs_atomic_dec_and_test(&tsi->tsi_nactive))
 828                 return;
 829
 830         /* the test instance is done */
 831         cfs_spin_lock(&tsi->tsi_lock);
 832
 833         tsi->tsi_stopping = 0;
 834
 835         cfs_spin_unlock(&tsi->tsi_lock);
 836
 837         cfs_spin_lock(&sfw_data.fw_lock);
 838
 839         if (!cfs_atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 840             sn == sfw_data.fw_session) {               /* sn also active */
 841                 cfs_spin_unlock(&sfw_data.fw_lock);
 842                 return;
 843         }
 844
 845         LASSERT (!cfs_list_empty(&sn->sn_list)); /* I'm a zombie! */
 846
 847         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 848                                        sfw_batch_t, bat_list) {
 849                 if (sfw_batch_active(tsb)) {
 850                         cfs_spin_unlock(&sfw_data.fw_lock);
 851                         return;
 852                 }
 853         }
 854
 855         cfs_list_del_init(&sn->sn_list);
 856         cfs_spin_unlock(&sfw_data.fw_lock);
 857
 858         sfw_destroy_session(sn);
 859         return;
 860 }
 861
 862 void
 863 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 864 {
 865         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 866         sfw_test_instance_t *tsi = tsu->tsu_instance;
 867         int                  done = 0;
 868
 869         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 870
 871         cfs_spin_lock(&tsi->tsi_lock);
 872
 873         LASSERT (sfw_test_active(tsi));
 874         LASSERT (!cfs_list_empty(&rpc->crpc_list));
 875
 876         cfs_list_del_init(&rpc->crpc_list);
 877
 878         /* batch is stopping or loop is done or get error */
 879         if (tsi->tsi_stopping ||
 880             tsu->tsu_loop == 0 ||
 881             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 882                 done = 1;
 883
 884         /* dec ref for poster */
 885         srpc_client_rpc_decref(rpc);
 886
 887         cfs_spin_unlock(&tsi->tsi_lock);
 888
 889         if (!done) {
 890                 swi_schedule_workitem(&tsu->tsu_worker);
 891                 return;
 892         }
 893
 894         sfw_test_unit_done(tsu);
 895         return;
 896 }
 897
 898 int
 899 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
 900                      int nblk, int blklen, srpc_client_rpc_t **rpcpp)
 901 {
 902         srpc_client_rpc_t   *rpc = NULL;
 903         sfw_test_instance_t *tsi = tsu->tsu_instance;
 904
 905         cfs_spin_lock(&tsi->tsi_lock);
 906
 907         LASSERT (sfw_test_active(tsi));
 908
 909         if (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 910                 /* pick request from buffer */
 911                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 912                                      srpc_client_rpc_t, crpc_list);
 913                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 914                 cfs_list_del_init(&rpc->crpc_list);
 915         }
 916
 917         cfs_spin_unlock(&tsi->tsi_lock);
 918
 919         if (rpc == NULL) {
 920                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 921                                              blklen, sfw_test_rpc_done,
 922                                              sfw_test_rpc_fini, tsu);
 923         } else {
 924                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 925                                      blklen, sfw_test_rpc_done,
 926                                      sfw_test_rpc_fini, tsu);
 927         }
 928
 929         if (rpc == NULL) {
 930                 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
 931                 return -ENOMEM;
 932         }
 933
 934         *rpcpp = rpc;
 935         return 0;
 936 }
 937
 938 int
 939 sfw_run_test (swi_workitem_t *wi)
 940 {
 941         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 942         sfw_test_instance_t *tsi = tsu->tsu_instance;
 943         srpc_client_rpc_t   *rpc = NULL;
 944
 945         LASSERT (wi == &tsu->tsu_worker);
 946
 947         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 948                 LASSERT (rpc == NULL);
 949                 goto test_done;
 950         }
 951
 952         LASSERT (rpc != NULL);
 953
 954         cfs_spin_lock(&tsi->tsi_lock);
 955
 956         if (tsi->tsi_stopping) {
 957                 cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 958                 cfs_spin_unlock(&tsi->tsi_lock);
 959                 goto test_done;
 960         }
 961
 962         if (tsu->tsu_loop > 0)
 963                 tsu->tsu_loop--;
 964
 965         cfs_list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 966         cfs_spin_unlock(&tsi->tsi_lock);
 967
 968         rpc->crpc_timeout = rpc_timeout;
 969
 970         cfs_spin_lock(&rpc->crpc_lock);
 971         srpc_post_rpc(rpc);
 972         cfs_spin_unlock(&rpc->crpc_lock);
 973         return 0;
 974
 975 test_done:
 976         /*
 977          * No one can schedule me now since:
 978          * - previous RPC, if any, has done and
 979          * - no new RPC is initiated.
 980          * - my batch is still active; no one can run it again now.
 981          * Cancel pending schedules and prevent future schedule attempts:
 982          */
 983         swi_exit_workitem(wi);
 984         sfw_test_unit_done(tsu);
 985         return 1;
 986 }
 987
 988 int
 989 sfw_run_batch (sfw_batch_t *tsb)
 990 {
 991         swi_workitem_t      *wi;
 992         sfw_test_unit_t     *tsu;
 993         sfw_test_instance_t *tsi;
 994
 995         if (sfw_batch_active(tsb)) {
 996                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
 997                        tsb->bat_id.bat_id, cfs_atomic_read(&tsb->bat_nactive));
 998                 return 0;
 999         }
1000
1001         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1002                                        sfw_test_instance_t, tsi_list) {
1003                 if (!tsi->tsi_is_client) /* skip server instances */
1004                         continue;
1005
1006                 LASSERT (!tsi->tsi_stopping);
1007                 LASSERT (!sfw_test_active(tsi));
1008
1009                 cfs_atomic_inc(&tsb->bat_nactive);
1010
1011                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
1012                                                sfw_test_unit_t, tsu_list) {
1013                         cfs_atomic_inc(&tsi->tsi_nactive);
1014                         tsu->tsu_loop = tsi->tsi_loop;
1015                         wi = &tsu->tsu_worker;
1016                         swi_init_workitem(wi, tsu, sfw_run_test,
1017                                           lst_sched_test[\
1018                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1019                         swi_schedule_workitem(wi);
1020                 }
1021         }
1022
1023         return 0;
1024 }
1025
1026 int
1027 sfw_stop_batch (sfw_batch_t *tsb, int force)
1028 {
1029         sfw_test_instance_t *tsi;
1030         srpc_client_rpc_t   *rpc;
1031
1032         if (!sfw_batch_active(tsb)) {
1033                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1034                 return 0;
1035         }
1036
1037         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1038                                        sfw_test_instance_t, tsi_list) {
1039                 cfs_spin_lock(&tsi->tsi_lock);
1040
1041                 if (!tsi->tsi_is_client ||
1042                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1043                         cfs_spin_unlock(&tsi->tsi_lock);
1044                         continue;
1045                 }
1046
1047                 tsi->tsi_stopping = 1;
1048
1049                 if (!force) {
1050                         cfs_spin_unlock(&tsi->tsi_lock);
1051                         continue;
1052                 }
1053
1054                 /* abort launched rpcs in the test */
1055                 cfs_list_for_each_entry_typed (rpc, &tsi->tsi_active_rpcs,
1056                                                srpc_client_rpc_t, crpc_list) {
1057                         cfs_spin_lock(&rpc->crpc_lock);
1058
1059                         srpc_abort_rpc(rpc, -EINTR);
1060
1061                         cfs_spin_unlock(&rpc->crpc_lock);
1062                 }
1063
1064                 cfs_spin_unlock(&tsi->tsi_lock);
1065         }
1066
1067         return 0;
1068 }
1069
1070 int
1071 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1072 {
1073         sfw_test_instance_t *tsi;
1074
1075         if (testidx < 0)
1076                 return -EINVAL;
1077
1078         if (testidx == 0) {
1079                 reply->bar_active = cfs_atomic_read(&tsb->bat_nactive);
1080                 return 0;
1081         }
1082
1083         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1084                                        sfw_test_instance_t, tsi_list) {
1085                 if (testidx-- > 1)
1086                         continue;
1087
1088                 reply->bar_active = cfs_atomic_read(&tsi->tsi_nactive);
1089                 return 0;
1090         }
1091
1092         return -ENOENT;
1093 }
1094
1095 void
1096 sfw_free_pages (srpc_server_rpc_t *rpc)
1097 {
1098         srpc_free_bulk(rpc->srpc_bulk);
1099         rpc->srpc_bulk = NULL;
1100 }
1101
1102 int
1103 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int sink)
1104 {
1105         LASSERT(rpc->srpc_bulk == NULL);
1106         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1107
1108         rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, sink);
1109         if (rpc->srpc_bulk == NULL)
1110                 return -ENOMEM;
1111
1112         return 0;
1113 }
1114
1115 int
1116 sfw_add_test (srpc_server_rpc_t *rpc)
1117 {
1118         sfw_session_t     *sn = sfw_data.fw_session;
1119         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1120         srpc_test_reqst_t *request;
1121         int                rc;
1122         sfw_batch_t       *bat;
1123
1124         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1125         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1126
1127         if (request->tsr_loop == 0 ||
1128             request->tsr_concur == 0 ||
1129             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1130             request->tsr_ndest > SFW_MAX_NDESTS ||
1131             (request->tsr_is_client && request->tsr_ndest == 0) ||
1132             request->tsr_concur > SFW_MAX_CONCUR ||
1133             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1134             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1135                 reply->tsr_status = EINVAL;
1136                 return 0;
1137         }
1138
1139         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1140             sfw_find_test_case(request->tsr_service) == NULL) {
1141                 reply->tsr_status = ENOENT;
1142                 return 0;
1143         }
1144
1145         bat = sfw_bid2batch(request->tsr_bid);
1146         if (bat == NULL) {
1147                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1148                         rpc->srpc_scd->scd_svc->sv_name,
1149                         libcfs_id2str(rpc->srpc_peer));
1150                 return -ENOMEM;
1151         }
1152
1153         if (sfw_batch_active(bat)) {
1154                 reply->tsr_status = EBUSY;
1155                 return 0;
1156         }
1157
1158         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1159                 /* rpc will be resumed later in sfw_bulk_ready */
1160                 return sfw_alloc_pages(rpc, CFS_CPT_ANY,
1161                                        sfw_id_pages(request->tsr_ndest), 1);
1162         }
1163
1164         rc = sfw_add_test_instance(bat, rpc);
1165         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1166                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1167                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1168                 request->tsr_is_client ? "client" : "server",
1169                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1170
1171         reply->tsr_status = (rc < 0) ? -rc : rc;
1172         return 0;
1173 }
1174
1175 int
1176 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1177 {
1178         sfw_session_t *sn = sfw_data.fw_session;
1179         int            rc = 0;
1180         sfw_batch_t   *bat;
1181
1182         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1183
1184         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1185                 reply->bar_status = ESRCH;
1186                 return 0;
1187         }
1188
1189         bat = sfw_find_batch(request->bar_bid);
1190         if (bat == NULL) {
1191                 reply->bar_status = ENOENT;
1192                 return 0;
1193         }
1194
1195         switch (request->bar_opc) {
1196         case SRPC_BATCH_OPC_RUN:
1197                 rc = sfw_run_batch(bat);
1198                 break;
1199
1200         case SRPC_BATCH_OPC_STOP:
1201                 rc = sfw_stop_batch(bat, request->bar_arg);
1202                 break;
1203
1204         case SRPC_BATCH_OPC_QUERY:
1205                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1206                 break;
1207
1208         default:
1209                 return -EINVAL; /* drop it */
1210         }
1211
1212         reply->bar_status = (rc < 0) ? -rc : rc;
1213         return 0;
1214 }
1215
1216 int
1217 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1218 {
1219         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1220         srpc_msg_t     *reply = &rpc->srpc_replymsg;
1221         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1222         int             rc = 0;
1223
1224         LASSERT (sfw_data.fw_active_srpc == NULL);
1225         LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1226
1227         cfs_spin_lock(&sfw_data.fw_lock);
1228
1229         if (sfw_data.fw_shuttingdown) {
1230                 cfs_spin_unlock(&sfw_data.fw_lock);
1231                 return -ESHUTDOWN;
1232         }
1233
1234         /* Remove timer to avoid racing with it or expiring active session */
1235         if (sfw_del_session_timer() != 0) {
1236                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1237                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1238                 cfs_spin_unlock(&sfw_data.fw_lock);
1239                 return -EAGAIN;
1240         }
1241
1242         sfw_data.fw_active_srpc = rpc;
1243         cfs_spin_unlock(&sfw_data.fw_lock);
1244
1245         sfw_unpack_message(request);
1246         LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1247
1248         switch(sv->sv_id) {
1249         default:
1250                 LBUG ();
1251         case SRPC_SERVICE_TEST:
1252                 rc = sfw_add_test(rpc);
1253                 break;
1254
1255         case SRPC_SERVICE_BATCH:
1256                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1257                                        &reply->msg_body.bat_reply);
1258                 break;
1259
1260         case SRPC_SERVICE_QUERY_STAT:
1261                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1262                                    &reply->msg_body.stat_reply);
1263                 break;
1264
1265         case SRPC_SERVICE_DEBUG:
1266                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1267                                        &reply->msg_body.dbg_reply);
1268                 break;
1269
1270         case SRPC_SERVICE_MAKE_SESSION:
1271                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1272                                       &reply->msg_body.mksn_reply);
1273                 break;
1274
1275         case SRPC_SERVICE_REMOVE_SESSION:
1276                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1277                                         &reply->msg_body.rmsn_reply);
1278                 break;
1279         }
1280
1281         rpc->srpc_done = sfw_server_rpc_done;
1282         cfs_spin_lock(&sfw_data.fw_lock);
1283
1284 #ifdef __KERNEL__
1285         if (!sfw_data.fw_shuttingdown)
1286                 sfw_add_session_timer();
1287 #else
1288         LASSERT (!sfw_data.fw_shuttingdown);
1289         sfw_add_session_timer();
1290 #endif
1291
1292         sfw_data.fw_active_srpc = NULL;
1293         cfs_spin_unlock(&sfw_data.fw_lock);
1294         return rc;
1295 }
1296
1297 int
1298 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1299 {
1300         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1301         int                     rc;
1302
1303         LASSERT (rpc->srpc_bulk != NULL);
1304         LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1305         LASSERT (sfw_data.fw_active_srpc == NULL);
1306         LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1307
1308         cfs_spin_lock(&sfw_data.fw_lock);
1309
1310         if (status != 0) {
1311                 CERROR ("Bulk transfer failed for RPC: "
1312                         "service %s, peer %s, status %d\n",
1313                         sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1314                 cfs_spin_unlock(&sfw_data.fw_lock);
1315                 return -EIO;
1316         }
1317
1318         if (sfw_data.fw_shuttingdown) {
1319                 cfs_spin_unlock(&sfw_data.fw_lock);
1320                 return -ESHUTDOWN;
1321         }
1322
1323         if (sfw_del_session_timer() != 0) {
1324                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1325                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1326                 cfs_spin_unlock(&sfw_data.fw_lock);
1327                 return -EAGAIN;
1328         }
1329
1330         sfw_data.fw_active_srpc = rpc;
1331         cfs_spin_unlock(&sfw_data.fw_lock);
1332
1333         rc = sfw_add_test(rpc);
1334
1335         cfs_spin_lock(&sfw_data.fw_lock);
1336
1337 #ifdef __KERNEL__
1338         if (!sfw_data.fw_shuttingdown)
1339                 sfw_add_session_timer();
1340 #else
1341         LASSERT (!sfw_data.fw_shuttingdown);
1342         sfw_add_session_timer();
1343 #endif
1344
1345         sfw_data.fw_active_srpc = NULL;
1346         cfs_spin_unlock(&sfw_data.fw_lock);
1347         return rc;
1348 }
1349
1350 srpc_client_rpc_t *
1351 sfw_create_rpc (lnet_process_id_t peer, int service,
1352                 int nbulkiov, int bulklen,
1353                 void (*done) (srpc_client_rpc_t *), void *priv)
1354 {
1355         srpc_client_rpc_t *rpc;
1356
1357         cfs_spin_lock(&sfw_data.fw_lock);
1358
1359         LASSERT (!sfw_data.fw_shuttingdown);
1360         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1361
1362         if (nbulkiov == 0 && !cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1363                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1364                                      srpc_client_rpc_t, crpc_list);
1365                 cfs_list_del(&rpc->crpc_list);
1366                 cfs_spin_unlock(&sfw_data.fw_lock);
1367
1368                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1369                                      done, sfw_client_rpc_fini, priv);
1370                 return rpc;
1371         }
1372
1373         cfs_spin_unlock(&sfw_data.fw_lock);
1374
1375         rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1376                                      nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1377                                      priv);
1378         return rpc;
1379 }
1380
1381 void
1382 sfw_unpack_message (srpc_msg_t *msg)
1383 {
1384         if (msg->msg_magic == SRPC_MSG_MAGIC)
1385                 return; /* no flipping needed */
1386
1387         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1388
1389         __swab32s(&msg->msg_type);
1390
1391         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1392                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1393
1394                 __swab32s(&req->str_type);
1395                 __swab64s(&req->str_rpyid);
1396                 sfw_unpack_sid(req->str_sid);
1397                 return;
1398         }
1399
1400         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1401                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1402
1403                 __swab32s(&rep->str_status);
1404                 sfw_unpack_sid(rep->str_sid);
1405                 sfw_unpack_fw_counters(rep->str_fw);
1406                 sfw_unpack_rpc_counters(rep->str_rpc);
1407                 sfw_unpack_lnet_counters(rep->str_lnet);
1408                 return;
1409         }
1410
1411         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1412                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1413
1414                 __swab64s(&req->mksn_rpyid);
1415                 __swab32s(&req->mksn_force);
1416                 sfw_unpack_sid(req->mksn_sid);
1417                 return;
1418         }
1419
1420         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1421                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1422
1423                 __swab32s(&rep->mksn_status);
1424                 __swab32s(&rep->mksn_timeout);
1425                 sfw_unpack_sid(rep->mksn_sid);
1426                 return;
1427         }
1428
1429         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1430                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1431
1432                 __swab64s(&req->rmsn_rpyid);
1433                 sfw_unpack_sid(req->rmsn_sid);
1434                 return;
1435         }
1436
1437         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1438                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1439
1440                 __swab32s(&rep->rmsn_status);
1441                 sfw_unpack_sid(rep->rmsn_sid);
1442                 return;
1443         }
1444
1445         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1446                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1447
1448                 __swab64s(&req->dbg_rpyid);
1449                 __swab32s(&req->dbg_flags);
1450                 sfw_unpack_sid(req->dbg_sid);
1451                 return;
1452         }
1453
1454         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1455                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1456
1457                 __swab32s(&rep->dbg_nbatch);
1458                 __swab32s(&rep->dbg_timeout);
1459                 sfw_unpack_sid(rep->dbg_sid);
1460                 return;
1461         }
1462
1463         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1464                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1465
1466                 __swab32s(&req->bar_opc);
1467                 __swab64s(&req->bar_rpyid);
1468                 __swab32s(&req->bar_testidx);
1469                 __swab32s(&req->bar_arg);
1470                 sfw_unpack_sid(req->bar_sid);
1471                 __swab64s(&req->bar_bid.bat_id);
1472                 return;
1473         }
1474
1475         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1476                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1477
1478                 __swab32s(&rep->bar_status);
1479                 sfw_unpack_sid(rep->bar_sid);
1480                 return;
1481         }
1482
1483         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1484                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1485
1486                 __swab64s(&req->tsr_rpyid);
1487                 __swab64s(&req->tsr_bulkid);
1488                 __swab32s(&req->tsr_loop);
1489                 __swab32s(&req->tsr_ndest);
1490                 __swab32s(&req->tsr_concur);
1491                 __swab32s(&req->tsr_service);
1492                 sfw_unpack_sid(req->tsr_sid);
1493                 __swab64s(&req->tsr_bid.bat_id);
1494                 return;
1495         }
1496
1497         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1498                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1499
1500                 __swab32s(&rep->tsr_status);
1501                 sfw_unpack_sid(rep->tsr_sid);
1502                 return;
1503         }
1504
1505         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1506                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1507
1508                 __swab64s(&req->join_rpyid);
1509                 sfw_unpack_sid(req->join_sid);
1510                 return;
1511         }
1512
1513         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1514                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1515
1516                 __swab32s(&rep->join_status);
1517                 __swab32s(&rep->join_timeout);
1518                 sfw_unpack_sid(rep->join_sid);
1519                 return;
1520         }
1521
1522         LBUG ();
1523         return;
1524 }
1525
1526 void
1527 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1528 {
1529         LASSERT (cfs_atomic_read(&rpc->crpc_refcount) > 0);
1530         LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1531
1532         cfs_spin_lock(&rpc->crpc_lock);
1533         srpc_abort_rpc(rpc, -EINTR);
1534         cfs_spin_unlock(&rpc->crpc_lock);
1535         return;
1536 }
1537
1538 void
1539 sfw_post_rpc (srpc_client_rpc_t *rpc)
1540 {
1541         cfs_spin_lock(&rpc->crpc_lock);
1542
1543         LASSERT (!rpc->crpc_closed);
1544         LASSERT (!rpc->crpc_aborted);
1545         LASSERT (cfs_list_empty(&rpc->crpc_list));
1546         LASSERT (!sfw_data.fw_shuttingdown);
1547
1548         rpc->crpc_timeout = rpc_timeout;
1549         srpc_post_rpc(rpc);
1550
1551         cfs_spin_unlock(&rpc->crpc_lock);
1552         return;
1553 }
1554
1555 static srpc_service_t sfw_services[] =
1556 {
1557         {
1558                 /* sv_id */    SRPC_SERVICE_DEBUG,
1559                 /* sv_name */  "debug",
1560                 0
1561         },
1562         {
1563                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1564                 /* sv_name */  "query stats",
1565                 0
1566         },
1567         {
1568                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1569                 /* sv_name */  "make session",
1570                 0
1571         },
1572         {
1573                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1574                 /* sv_name */  "remove session",
1575                 0
1576         },
1577         {
1578                 /* sv_id */    SRPC_SERVICE_BATCH,
1579                 /* sv_name */  "batch service",
1580                 0
1581         },
1582         {
1583                 /* sv_id */    SRPC_SERVICE_TEST,
1584                 /* sv_name */  "test service",
1585                 0
1586         },
1587         {
1588                 /* sv_id */    0,
1589                 /* sv_name */  NULL,
1590                 0
1591         }
1592 };
1593
1594 extern sfw_test_client_ops_t ping_test_client;
1595 extern srpc_service_t        ping_test_service;
1596 extern void ping_init_test_client(void);
1597 extern void ping_init_test_service(void);
1598
1599 extern sfw_test_client_ops_t brw_test_client;
1600 extern srpc_service_t        brw_test_service;
1601 extern void brw_init_test_client(void);
1602 extern void brw_init_test_service(void);
1603
1604
1605 int
1606 sfw_startup (void)
1607 {
1608         int              i;
1609         int              rc;
1610         int              error;
1611         srpc_service_t  *sv;
1612         sfw_test_case_t *tsc;
1613
1614 #ifndef __KERNEL__
1615         char *s;
1616
1617         s = getenv("SESSION_TIMEOUT");
1618         session_timeout = s != NULL ? atoi(s) : session_timeout;
1619
1620         s = getenv("RPC_TIMEOUT");
1621         rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1622 #endif
1623
1624         if (session_timeout < 0) {
1625                 CERROR ("Session timeout must be non-negative: %d\n",
1626                         session_timeout);
1627                 return -EINVAL;
1628         }
1629
1630         if (rpc_timeout < 0) {
1631                 CERROR ("RPC timeout must be non-negative: %d\n",
1632                         rpc_timeout);
1633                 return -EINVAL;
1634         }
1635
1636         if (session_timeout == 0)
1637                 CWARN ("Zero session_timeout specified "
1638                        "- test sessions never expire.\n");
1639
1640         if (rpc_timeout == 0)
1641                 CWARN ("Zero rpc_timeout specified "
1642                        "- test RPC never expire.\n");
1643
1644         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1645
1646         sfw_data.fw_session     = NULL;
1647         sfw_data.fw_active_srpc = NULL;
1648         cfs_spin_lock_init(&sfw_data.fw_lock);
1649         cfs_atomic_set(&sfw_data.fw_nzombies, 0);
1650         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1651         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1652         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1653
1654         brw_init_test_client();
1655         brw_init_test_service();
1656         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1657         LASSERT (rc == 0);
1658
1659         ping_init_test_client();
1660         ping_init_test_service();
1661         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1662         LASSERT (rc == 0);
1663
1664         error = 0;
1665         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1666                                        sfw_test_case_t, tsc_list) {
1667                 sv = tsc->tsc_srv_service;
1668
1669                 rc = srpc_add_service(sv);
1670                 LASSERT (rc != -EBUSY);
1671                 if (rc != 0) {
1672                         CWARN ("Failed to add %s service: %d\n",
1673                                sv->sv_name, rc);
1674                         error = rc;
1675                 }
1676         }
1677
1678         for (i = 0; ; i++) {
1679                 sv = &sfw_services[i];
1680                 if (sv->sv_name == NULL) break;
1681
1682                 sv->sv_bulk_ready = NULL;
1683                 sv->sv_handler    = sfw_handle_server_rpc;
1684                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1685                 if (sv->sv_id == SRPC_SERVICE_TEST)
1686                         sv->sv_bulk_ready = sfw_bulk_ready;
1687
1688                 rc = srpc_add_service(sv);
1689                 LASSERT (rc != -EBUSY);
1690                 if (rc != 0) {
1691                         CWARN ("Failed to add %s service: %d\n",
1692                                sv->sv_name, rc);
1693                         error = rc;
1694                 }
1695
1696                 /* about to sfw_shutdown, no need to add buffer */
1697                 if (error) continue;
1698
1699                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1700                 if (rc != 0) {
1701                         CWARN("Failed to reserve enough buffers: "
1702                               "service %s, %d needed: %d\n",
1703                               sv->sv_name, sv->sv_wi_total, rc);
1704                         error = -ENOMEM;
1705                 }
1706         }
1707
1708         if (error != 0)
1709                 sfw_shutdown();
1710         return error;
1711 }
1712
1713 void
1714 sfw_shutdown (void)
1715 {
1716         srpc_service_t  *sv;
1717         sfw_test_case_t *tsc;
1718         int              i;
1719
1720         cfs_spin_lock(&sfw_data.fw_lock);
1721
1722         sfw_data.fw_shuttingdown = 1;
1723 #ifdef __KERNEL__
1724         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1725                        "waiting for active RPC to finish.\n");
1726 #else
1727         LASSERT (sfw_data.fw_active_srpc == NULL);
1728 #endif
1729
1730         if (sfw_del_session_timer() != 0)
1731                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1732                                "waiting for session timer to explode.\n");
1733
1734         sfw_deactivate_session();
1735         lst_wait_until(cfs_atomic_read(&sfw_data.fw_nzombies) == 0,
1736                        sfw_data.fw_lock,
1737                        "waiting for %d zombie sessions to die.\n",
1738                        cfs_atomic_read(&sfw_data.fw_nzombies));
1739
1740         cfs_spin_unlock(&sfw_data.fw_lock);
1741
1742         for (i = 0; ; i++) {
1743                 sv = &sfw_services[i];
1744                 if (sv->sv_name == NULL)
1745                         break;
1746
1747                 srpc_shutdown_service(sv);
1748                 srpc_remove_service(sv);
1749         }
1750
1751         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1752                                        sfw_test_case_t, tsc_list) {
1753                 sv = tsc->tsc_srv_service;
1754                 srpc_shutdown_service(sv);
1755                 srpc_remove_service(sv);
1756         }
1757
1758         while (!cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1759                 srpc_client_rpc_t *rpc;
1760
1761                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1762                                      srpc_client_rpc_t, crpc_list);
1763                 cfs_list_del(&rpc->crpc_list);
1764
1765                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1766         }
1767
1768         for (i = 0; ; i++) {
1769                 sv = &sfw_services[i];
1770                 if (sv->sv_name == NULL)
1771                         break;
1772
1773                 srpc_wait_service_shutdown(sv);
1774         }
1775
1776         while (!cfs_list_empty(&sfw_data.fw_tests)) {
1777                 tsc = cfs_list_entry(sfw_data.fw_tests.next,
1778                                      sfw_test_case_t, tsc_list);
1779
1780                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1781
1782                 cfs_list_del(&tsc->tsc_list);
1783                 LIBCFS_FREE(tsc, sizeof(*tsc));
1784         }
1785
1786         return;
1787 }