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  * Copyright (c) 2012, Intel Corporation.
  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 static int session_timeout = 100;
  49 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  50                 "test session timeout in seconds (100 by default, 0 == never)");
  51
  52 static int rpc_timeout = 64;
  53 CFS_MODULE_PARM(rpc_timeout, "i", int, 0644,
  54                 "rpc timeout in seconds (64 by default, 0 == never)");
  55
  56 #define sfw_unpack_id(id)               \
  57 do {                                    \
  58         __swab64s(&(id).nid);           \
  59         __swab32s(&(id).pid);           \
  60 } while (0)
  61
  62 #define sfw_unpack_sid(sid)             \
  63 do {                                    \
  64         __swab64s(&(sid).ses_nid);      \
  65         __swab64s(&(sid).ses_stamp);    \
  66 } while (0)
  67
  68 #define sfw_unpack_fw_counters(fc)        \
  69 do {                                      \
  70         __swab32s(&(fc).running_ms);      \
  71         __swab32s(&(fc).active_batches);  \
  72         __swab32s(&(fc).zombie_sessions); \
  73         __swab32s(&(fc).brw_errors);      \
  74         __swab32s(&(fc).ping_errors);     \
  75 } while (0)
  76
  77 #define sfw_unpack_rpc_counters(rc)     \
  78 do {                                    \
  79         __swab32s(&(rc).errors);        \
  80         __swab32s(&(rc).rpcs_sent);     \
  81         __swab32s(&(rc).rpcs_rcvd);     \
  82         __swab32s(&(rc).rpcs_dropped);  \
  83         __swab32s(&(rc).rpcs_expired);  \
  84         __swab64s(&(rc).bulk_get);      \
  85         __swab64s(&(rc).bulk_put);      \
  86 } while (0)
  87
  88 #define sfw_unpack_lnet_counters(lc)    \
  89 do {                                    \
  90         __swab32s(&(lc).errors);        \
  91         __swab32s(&(lc).msgs_max);      \
  92         __swab32s(&(lc).msgs_alloc);    \
  93         __swab32s(&(lc).send_count);    \
  94         __swab32s(&(lc).recv_count);    \
  95         __swab32s(&(lc).drop_count);    \
  96         __swab32s(&(lc).route_count);   \
  97         __swab64s(&(lc).send_length);   \
  98         __swab64s(&(lc).recv_length);   \
  99         __swab64s(&(lc).drop_length);   \
 100         __swab64s(&(lc).route_length);  \
 101 } while (0)
 102
 103 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 104 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 105
 106 struct smoketest_framework {
 107         cfs_list_t         fw_zombie_rpcs;     /* RPCs to be recycled */
 108         cfs_list_t         fw_zombie_sessions; /* stopping sessions */
 109         cfs_list_t         fw_tests;           /* registered test cases */
 110         atomic_t       fw_nzombies;        /* # zombie sessions */
 111         spinlock_t         fw_lock;             /* serialise */
 112         sfw_session_t     *fw_session;          /* _the_ session */
 113         int                fw_shuttingdown;     /* shutdown in progress */
 114         srpc_server_rpc_t *fw_active_srpc;      /* running RPC */
 115 } sfw_data;
 116
 117 /* forward ref's */
 118 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 119 void sfw_destroy_session (sfw_session_t *sn);
 120
 121 static inline sfw_test_case_t *
 122 sfw_find_test_case(int id)
 123 {
 124         sfw_test_case_t *tsc;
 125
 126         LASSERT (id <= SRPC_SERVICE_MAX_ID);
 127         LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 128
 129         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 130                                        sfw_test_case_t, tsc_list) {
 131                 if (tsc->tsc_srv_service->sv_id == id)
 132                         return tsc;
 133         }
 134
 135         return NULL;
 136 }
 137
 138 static int
 139 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 140 {
 141         sfw_test_case_t *tsc;
 142
 143         if (sfw_find_test_case(service->sv_id) != NULL) {
 144                 CERROR ("Failed to register test %s (%d)\n",
 145                         service->sv_name, service->sv_id);
 146                 return -EEXIST;
 147         }
 148
 149         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 150         if (tsc == NULL)
 151                 return -ENOMEM;
 152
 153         memset(tsc, 0, sizeof(sfw_test_case_t));
 154         tsc->tsc_cli_ops     = cliops;
 155         tsc->tsc_srv_service = service;
 156
 157         cfs_list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 158         return 0;
 159 }
 160
 161 void
 162 sfw_add_session_timer (void)
 163 {
 164         sfw_session_t *sn = sfw_data.fw_session;
 165         stt_timer_t   *timer = &sn->sn_timer;
 166
 167         LASSERT (!sfw_data.fw_shuttingdown);
 168
 169         if (sn == NULL || sn->sn_timeout == 0)
 170                 return;
 171
 172         LASSERT (!sn->sn_timer_active);
 173
 174         sn->sn_timer_active = 1;
 175         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 176                                           cfs_time_current_sec());
 177         stt_add_timer(timer);
 178         return;
 179 }
 180
 181 int
 182 sfw_del_session_timer (void)
 183 {
 184         sfw_session_t *sn = sfw_data.fw_session;
 185
 186         if (sn == NULL || !sn->sn_timer_active)
 187                 return 0;
 188
 189         LASSERT (sn->sn_timeout != 0);
 190
 191         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 192                 sn->sn_timer_active = 0;
 193                 return 0;
 194         }
 195
 196 #ifndef __KERNEL__
 197         /* Racing is impossible in single-threaded userland selftest */
 198         LBUG();
 199 #endif
 200         return EBUSY; /* racing with sfw_session_expired() */
 201 }
 202
 203 /* called with sfw_data.fw_lock held */
 204 static void
 205 sfw_deactivate_session (void)
 206 {
 207         sfw_session_t *sn = sfw_data.fw_session;
 208         int            nactive = 0;
 209         sfw_batch_t   *tsb;
 210         sfw_test_case_t *tsc;
 211
 212         if (sn == NULL) return;
 213
 214         LASSERT (!sn->sn_timer_active);
 215
 216         sfw_data.fw_session = NULL;
 217         atomic_inc(&sfw_data.fw_nzombies);
 218         cfs_list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 219
 220         spin_unlock(&sfw_data.fw_lock);
 221
 222         cfs_list_for_each_entry_typed(tsc, &sfw_data.fw_tests,
 223                                       sfw_test_case_t, tsc_list) {
 224                 srpc_abort_service(tsc->tsc_srv_service);
 225         }
 226
 227         spin_lock(&sfw_data.fw_lock);
 228
 229         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 230                                        sfw_batch_t, bat_list) {
 231                 if (sfw_batch_active(tsb)) {
 232                         nactive++;
 233                         sfw_stop_batch(tsb, 1);
 234                 }
 235         }
 236
 237         if (nactive != 0)
 238                 return;   /* wait for active batches to stop */
 239
 240         cfs_list_del_init(&sn->sn_list);
 241         spin_unlock(&sfw_data.fw_lock);
 242
 243         sfw_destroy_session(sn);
 244
 245         spin_lock(&sfw_data.fw_lock);
 246 }
 247
 248 #ifndef __KERNEL__
 249
 250 int
 251 sfw_session_removed (void)
 252 {
 253         return (sfw_data.fw_session == NULL) ? 1 : 0;
 254 }
 255
 256 #endif
 257
 258 void
 259 sfw_session_expired (void *data)
 260 {
 261         sfw_session_t *sn = data;
 262
 263         spin_lock(&sfw_data.fw_lock);
 264
 265         LASSERT (sn->sn_timer_active);
 266         LASSERT (sn == sfw_data.fw_session);
 267
 268         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 269                libcfs_nid2str(sn->sn_id.ses_nid),
 270                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 271
 272         sn->sn_timer_active = 0;
 273         sfw_deactivate_session();
 274
 275         spin_unlock(&sfw_data.fw_lock);
 276 }
 277
 278 static inline void
 279 sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
 280                  unsigned features, const char *name)
 281 {
 282         stt_timer_t *timer = &sn->sn_timer;
 283
 284         memset(sn, 0, sizeof(sfw_session_t));
 285         CFS_INIT_LIST_HEAD(&sn->sn_list);
 286         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 287         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 288         atomic_set(&sn->sn_brw_errors, 0);
 289         atomic_set(&sn->sn_ping_errors, 0);
 290         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 291
 292         sn->sn_timer_active = 0;
 293         sn->sn_id           = sid;
 294         sn->sn_features     = features;
 295         sn->sn_timeout      = session_timeout;
 296         sn->sn_started      = cfs_time_current();
 297
 298         timer->stt_data = sn;
 299         timer->stt_func = sfw_session_expired;
 300         CFS_INIT_LIST_HEAD(&timer->stt_list);
 301 }
 302
 303 /* completion handler for incoming framework RPCs */
 304 void
 305 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 306 {
 307         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 308         int                     status  = rpc->srpc_status;
 309
 310         CDEBUG (D_NET,
 311                 "Incoming framework RPC done: "
 312                 "service %s, peer %s, status %s:%d\n",
 313                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 314                 swi_state2str(rpc->srpc_wi.swi_state),
 315                 status);
 316
 317         if (rpc->srpc_bulk != NULL)
 318                 sfw_free_pages(rpc);
 319         return;
 320 }
 321
 322 void
 323 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 324 {
 325         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 326         LASSERT (cfs_list_empty(&rpc->crpc_list));
 327         LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
 328 #ifndef __KERNEL__
 329         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 330 #endif
 331
 332         CDEBUG (D_NET,
 333                 "Outgoing framework RPC done: "
 334                 "service %d, peer %s, status %s:%d:%d\n",
 335                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 336                 swi_state2str(rpc->crpc_wi.swi_state),
 337                 rpc->crpc_aborted, rpc->crpc_status);
 338
 339         spin_lock(&sfw_data.fw_lock);
 340
 341         /* my callers must finish all RPCs before shutting me down */
 342         LASSERT(!sfw_data.fw_shuttingdown);
 343         cfs_list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 344
 345         spin_unlock(&sfw_data.fw_lock);
 346 }
 347
 348 sfw_batch_t *
 349 sfw_find_batch (lst_bid_t bid)
 350 {
 351         sfw_session_t *sn = sfw_data.fw_session;
 352         sfw_batch_t   *bat;
 353
 354         LASSERT (sn != NULL);
 355
 356         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 357                                        sfw_batch_t, bat_list) {
 358                 if (bat->bat_id.bat_id == bid.bat_id)
 359                         return bat;
 360         }
 361
 362         return NULL;
 363 }
 364
 365 sfw_batch_t *
 366 sfw_bid2batch (lst_bid_t bid)
 367 {
 368         sfw_session_t *sn = sfw_data.fw_session;
 369         sfw_batch_t   *bat;
 370
 371         LASSERT (sn != NULL);
 372
 373         bat = sfw_find_batch(bid);
 374         if (bat != NULL)
 375                 return bat;
 376
 377         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 378         if (bat == NULL)
 379                 return NULL;
 380
 381         bat->bat_error    = 0;
 382         bat->bat_session  = sn;
 383         bat->bat_id       = bid;
 384         atomic_set(&bat->bat_nactive, 0);
 385         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 386
 387         cfs_list_add_tail(&bat->bat_list, &sn->sn_batches);
 388         return bat;
 389 }
 390
 391 int
 392 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 393 {
 394         sfw_session_t  *sn = sfw_data.fw_session;
 395         sfw_counters_t *cnt = &reply->str_fw;
 396         sfw_batch_t    *bat;
 397         struct timeval  tv;
 398
 399         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 400
 401         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 402                 reply->str_status = EINVAL;
 403                 return 0;
 404         }
 405
 406         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 407                 reply->str_status = ESRCH;
 408                 return 0;
 409         }
 410
 411         lnet_counters_get(&reply->str_lnet);
 412         srpc_get_counters(&reply->str_rpc);
 413
 414         /* send over the msecs since the session was started
 415          - with 32 bits to send, this is ~49 days */
 416         cfs_duration_usec(cfs_time_sub(cfs_time_current(),
 417                                        sn->sn_started), &tv);
 418
 419         cnt->running_ms      = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
 420         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 421         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 422         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 423
 424         cnt->active_batches = 0;
 425         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 426                                        sfw_batch_t, bat_list) {
 427                 if (atomic_read(&bat->bat_nactive) > 0)
 428                         cnt->active_batches++;
 429         }
 430
 431         reply->str_status = 0;
 432         return 0;
 433 }
 434
 435 int
 436 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 437 {
 438         sfw_session_t *sn = sfw_data.fw_session;
 439         srpc_msg_t    *msg = container_of(request, srpc_msg_t,
 440                                           msg_body.mksn_reqst);
 441         int            cplen = 0;
 442
 443         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 444                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 445                 reply->mksn_status = EINVAL;
 446                 return 0;
 447         }
 448
 449         if (sn != NULL) {
 450                 reply->mksn_status  = 0;
 451                 reply->mksn_sid     = sn->sn_id;
 452                 reply->mksn_timeout = sn->sn_timeout;
 453
 454                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 455                         atomic_inc(&sn->sn_refcount);
 456                         return 0;
 457                 }
 458
 459                 if (!request->mksn_force) {
 460                         reply->mksn_status = EBUSY;
 461                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 462                                         sizeof(reply->mksn_name));
 463                         if (cplen >= sizeof(reply->mksn_name))
 464                                 return -E2BIG;
 465                         return 0;
 466                 }
 467         }
 468
 469         /* reject the request if it requires unknown features
 470          * NB: old version will always accept all features because it's not
 471          * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
 472          * harmless because it will return zero feature to console, and it's
 473          * console's responsibility to make sure all nodes in a session have
 474          * same feature mask. */
 475         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 476                 reply->mksn_status = EPROTO;
 477                 return 0;
 478         }
 479
 480         /* brand new or create by force */
 481         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 482         if (sn == NULL) {
 483                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 484                 return -ENOMEM;
 485         }
 486
 487         sfw_init_session(sn, request->mksn_sid,
 488                          msg->msg_ses_feats, &request->mksn_name[0]);
 489
 490         spin_lock(&sfw_data.fw_lock);
 491
 492         sfw_deactivate_session();
 493         LASSERT(sfw_data.fw_session == NULL);
 494         sfw_data.fw_session = sn;
 495
 496         spin_unlock(&sfw_data.fw_lock);
 497
 498         reply->mksn_status  = 0;
 499         reply->mksn_sid     = sn->sn_id;
 500         reply->mksn_timeout = sn->sn_timeout;
 501         return 0;
 502 }
 503
 504 int
 505 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 506 {
 507         sfw_session_t *sn = sfw_data.fw_session;
 508
 509         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 510
 511         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 512                 reply->rmsn_status = EINVAL;
 513                 return 0;
 514         }
 515
 516         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 517                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 518                 return 0;
 519         }
 520
 521         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 522                 reply->rmsn_status = 0;
 523                 return 0;
 524         }
 525
 526         spin_lock(&sfw_data.fw_lock);
 527         sfw_deactivate_session();
 528         spin_unlock(&sfw_data.fw_lock);
 529
 530         reply->rmsn_status = 0;
 531         reply->rmsn_sid    = LST_INVALID_SID;
 532         LASSERT(sfw_data.fw_session == NULL);
 533         return 0;
 534 }
 535
 536 int
 537 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 538 {
 539         sfw_session_t *sn = sfw_data.fw_session;
 540
 541         if (sn == NULL) {
 542                 reply->dbg_status = ESRCH;
 543                 reply->dbg_sid    = LST_INVALID_SID;
 544                 return 0;
 545         }
 546
 547         reply->dbg_status  = 0;
 548         reply->dbg_sid     = sn->sn_id;
 549         reply->dbg_timeout = sn->sn_timeout;
 550         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 551             >= sizeof(reply->dbg_name))
 552                 return -E2BIG;
 553
 554         return 0;
 555 }
 556
 557 void
 558 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 559 {
 560         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 561         sfw_test_instance_t *tsi = tsu->tsu_instance;
 562
 563         /* Called with hold of tsi->tsi_lock */
 564         LASSERT (cfs_list_empty(&rpc->crpc_list));
 565         cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 566 }
 567
 568 static inline int
 569 sfw_test_buffers(sfw_test_instance_t *tsi)
 570 {
 571         struct sfw_test_case    *tsc;
 572         struct srpc_service     *svc;
 573         int                     nbuf;
 574
 575         LASSERT(tsi != NULL);
 576         tsc = sfw_find_test_case(tsi->tsi_service);
 577         LASSERT(tsc != NULL);
 578         svc = tsc->tsc_srv_service;
 579         LASSERT(svc != NULL);
 580
 581         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 582         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 583 }
 584
 585 int
 586 sfw_load_test(struct sfw_test_instance *tsi)
 587 {
 588         struct sfw_test_case    *tsc;
 589         struct srpc_service     *svc;
 590         int                     nbuf;
 591         int                     rc;
 592
 593         LASSERT(tsi != NULL);
 594         tsc = sfw_find_test_case(tsi->tsi_service);
 595         nbuf = sfw_test_buffers(tsi);
 596         LASSERT(tsc != NULL);
 597         svc = tsc->tsc_srv_service;
 598
 599         if (tsi->tsi_is_client) {
 600                 tsi->tsi_ops = tsc->tsc_cli_ops;
 601                 return 0;
 602         }
 603
 604         rc = srpc_service_add_buffers(svc, nbuf);
 605         if (rc != 0) {
 606                 CWARN("Failed to reserve enough buffers: "
 607                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 608                 /* NB: this error handler is not strictly correct, because
 609                  * it may release more buffers than already allocated,
 610                  * but it doesn't matter because request portal should
 611                  * be lazy portal and will grow buffers if necessary. */
 612                 srpc_service_remove_buffers(svc, nbuf);
 613                 return -ENOMEM;
 614         }
 615
 616         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 617                nbuf * (srpc_serv_is_framework(svc) ?
 618                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 619         return 0;
 620 }
 621
 622 void
 623 sfw_unload_test(struct sfw_test_instance *tsi)
 624 {
 625         struct sfw_test_case *tsc;
 626
 627         LASSERT(tsi != NULL);
 628         tsc = sfw_find_test_case(tsi->tsi_service);
 629         LASSERT(tsc != NULL);
 630
 631         if (tsi->tsi_is_client)
 632                 return;
 633
 634         /* shrink buffers, because request portal is lazy portal
 635          * which can grow buffers at runtime so we may leave
 636          * some buffers behind, but never mind... */
 637         srpc_service_remove_buffers(tsc->tsc_srv_service,
 638                                     sfw_test_buffers(tsi));
 639         return;
 640 }
 641
 642 void
 643 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 644 {
 645         srpc_client_rpc_t *rpc;
 646         sfw_test_unit_t   *tsu;
 647
 648         if (!tsi->tsi_is_client) goto clean;
 649
 650         tsi->tsi_ops->tso_fini(tsi);
 651
 652         LASSERT (!tsi->tsi_stopping);
 653         LASSERT (cfs_list_empty(&tsi->tsi_active_rpcs));
 654         LASSERT (!sfw_test_active(tsi));
 655
 656         while (!cfs_list_empty(&tsi->tsi_units)) {
 657                 tsu = cfs_list_entry(tsi->tsi_units.next,
 658                                      sfw_test_unit_t, tsu_list);
 659                 cfs_list_del(&tsu->tsu_list);
 660                 LIBCFS_FREE(tsu, sizeof(*tsu));
 661         }
 662
 663         while (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 664                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 665                                      srpc_client_rpc_t, crpc_list);
 666                 cfs_list_del(&rpc->crpc_list);
 667                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 668         }
 669
 670 clean:
 671         sfw_unload_test(tsi);
 672         LIBCFS_FREE(tsi, sizeof(*tsi));
 673         return;
 674 }
 675
 676 void
 677 sfw_destroy_batch (sfw_batch_t *tsb)
 678 {
 679         sfw_test_instance_t *tsi;
 680
 681         LASSERT (!sfw_batch_active(tsb));
 682         LASSERT (cfs_list_empty(&tsb->bat_list));
 683
 684         while (!cfs_list_empty(&tsb->bat_tests)) {
 685                 tsi = cfs_list_entry(tsb->bat_tests.next,
 686                                      sfw_test_instance_t, tsi_list);
 687                 cfs_list_del_init(&tsi->tsi_list);
 688                 sfw_destroy_test_instance(tsi);
 689         }
 690
 691         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 692         return;
 693 }
 694
 695 void
 696 sfw_destroy_session (sfw_session_t *sn)
 697 {
 698         sfw_batch_t *batch;
 699
 700         LASSERT (cfs_list_empty(&sn->sn_list));
 701         LASSERT (sn != sfw_data.fw_session);
 702
 703         while (!cfs_list_empty(&sn->sn_batches)) {
 704                 batch = cfs_list_entry(sn->sn_batches.next,
 705                                        sfw_batch_t, bat_list);
 706                 cfs_list_del_init(&batch->bat_list);
 707                 sfw_destroy_batch(batch);
 708         }
 709
 710         LIBCFS_FREE(sn, sizeof(*sn));
 711         atomic_dec(&sfw_data.fw_nzombies);
 712         return;
 713 }
 714
 715 void
 716 sfw_unpack_addtest_req(srpc_msg_t *msg)
 717 {
 718         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 719
 720         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 721         LASSERT (req->tsr_is_client);
 722
 723         if (msg->msg_magic == SRPC_MSG_MAGIC)
 724                 return; /* no flipping needed */
 725
 726         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 727
 728         if (req->tsr_service == SRPC_SERVICE_BRW) {
 729                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 730                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 731
 732                         __swab32s(&bulk->blk_opc);
 733                         __swab32s(&bulk->blk_npg);
 734                         __swab32s(&bulk->blk_flags);
 735
 736                 } else {
 737                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 738
 739                         __swab16s(&bulk->blk_opc);
 740                         __swab16s(&bulk->blk_flags);
 741                         __swab32s(&bulk->blk_offset);
 742                         __swab32s(&bulk->blk_len);
 743                 }
 744
 745                 return;
 746         }
 747
 748         if (req->tsr_service == SRPC_SERVICE_PING) {
 749                 test_ping_req_t *ping = &req->tsr_u.ping;
 750
 751                 __swab32s(&ping->png_size);
 752                 __swab32s(&ping->png_flags);
 753                 return;
 754         }
 755
 756         LBUG ();
 757         return;
 758 }
 759
 760 int
 761 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 762 {
 763         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 764         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 765         srpc_bulk_t         *bk = rpc->srpc_bulk;
 766         int                  ndest = req->tsr_ndest;
 767         sfw_test_unit_t     *tsu;
 768         sfw_test_instance_t *tsi;
 769         int                  i;
 770         int                  rc;
 771
 772         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 773         if (tsi == NULL) {
 774                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 775                         tsb->bat_id.bat_id);
 776                 return -ENOMEM;
 777         }
 778
 779         memset(tsi, 0, sizeof(*tsi));
 780         spin_lock_init(&tsi->tsi_lock);
 781         atomic_set(&tsi->tsi_nactive, 0);
 782         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 783         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 784         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 785
 786         tsi->tsi_stopping      = 0;
 787         tsi->tsi_batch         = tsb;
 788         tsi->tsi_loop          = req->tsr_loop;
 789         tsi->tsi_concur        = req->tsr_concur;
 790         tsi->tsi_service       = req->tsr_service;
 791         tsi->tsi_is_client     = !!(req->tsr_is_client);
 792         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 793
 794         rc = sfw_load_test(tsi);
 795         if (rc != 0) {
 796                 LIBCFS_FREE(tsi, sizeof(*tsi));
 797                 return rc;
 798         }
 799
 800         LASSERT (!sfw_batch_active(tsb));
 801
 802         if (!tsi->tsi_is_client) {
 803                 /* it's test server, just add it to tsb */
 804                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 805                 return 0;
 806         }
 807
 808         LASSERT (bk != NULL);
 809 #ifndef __KERNEL__
 810         LASSERT (bk->bk_pages != NULL);
 811 #endif
 812         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 813         LASSERT((unsigned int)bk->bk_len >=
 814                 sizeof(lnet_process_id_packed_t) * ndest);
 815
 816         sfw_unpack_addtest_req(msg);
 817         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 818
 819         for (i = 0; i < ndest; i++) {
 820                 lnet_process_id_packed_t *dests;
 821                 lnet_process_id_packed_t  id;
 822                 int                       j;
 823
 824 #ifdef __KERNEL__
 825                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 826                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 827 #else
 828                 dests = page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 829 #endif
 830                 id = dests[i % SFW_ID_PER_PAGE];
 831                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 832                         sfw_unpack_id(id);
 833
 834                 for (j = 0; j < tsi->tsi_concur; j++) {
 835                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 836                         if (tsu == NULL) {
 837                                 rc = -ENOMEM;
 838                                 CERROR ("Can't allocate tsu for %d\n",
 839                                         tsi->tsi_service);
 840                                 goto error;
 841                         }
 842
 843                         tsu->tsu_dest.nid = id.nid;
 844                         tsu->tsu_dest.pid = id.pid;
 845                         tsu->tsu_instance = tsi;
 846                         tsu->tsu_private  = NULL;
 847                         cfs_list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 848                 }
 849         }
 850
 851         rc = tsi->tsi_ops->tso_init(tsi);
 852         if (rc == 0) {
 853                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 854                 return 0;
 855         }
 856
 857 error:
 858         LASSERT (rc != 0);
 859         sfw_destroy_test_instance(tsi);
 860         return rc;
 861 }
 862
 863 static void
 864 sfw_test_unit_done (sfw_test_unit_t *tsu)
 865 {
 866         sfw_test_instance_t *tsi = tsu->tsu_instance;
 867         sfw_batch_t         *tsb = tsi->tsi_batch;
 868         sfw_session_t       *sn = tsb->bat_session;
 869
 870         LASSERT (sfw_test_active(tsi));
 871
 872         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 873                 return;
 874
 875         /* the test instance is done */
 876         spin_lock(&tsi->tsi_lock);
 877
 878         tsi->tsi_stopping = 0;
 879
 880         spin_unlock(&tsi->tsi_lock);
 881
 882         spin_lock(&sfw_data.fw_lock);
 883
 884         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 885             sn == sfw_data.fw_session) {                  /* sn also active */
 886                 spin_unlock(&sfw_data.fw_lock);
 887                 return;
 888         }
 889
 890         LASSERT (!cfs_list_empty(&sn->sn_list)); /* I'm a zombie! */
 891
 892         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 893                                        sfw_batch_t, bat_list) {
 894                 if (sfw_batch_active(tsb)) {
 895                         spin_unlock(&sfw_data.fw_lock);
 896                         return;
 897                 }
 898         }
 899
 900         cfs_list_del_init(&sn->sn_list);
 901         spin_unlock(&sfw_data.fw_lock);
 902
 903         sfw_destroy_session(sn);
 904         return;
 905 }
 906
 907 void
 908 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 909 {
 910         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 911         sfw_test_instance_t *tsi = tsu->tsu_instance;
 912         int                  done = 0;
 913
 914         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 915
 916         spin_lock(&tsi->tsi_lock);
 917
 918         LASSERT (sfw_test_active(tsi));
 919         LASSERT (!cfs_list_empty(&rpc->crpc_list));
 920
 921         cfs_list_del_init(&rpc->crpc_list);
 922
 923         /* batch is stopping or loop is done or get error */
 924         if (tsi->tsi_stopping ||
 925             tsu->tsu_loop == 0 ||
 926             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 927                 done = 1;
 928
 929         /* dec ref for poster */
 930         srpc_client_rpc_decref(rpc);
 931
 932         spin_unlock(&tsi->tsi_lock);
 933
 934         if (!done) {
 935                 swi_schedule_workitem(&tsu->tsu_worker);
 936                 return;
 937         }
 938
 939         sfw_test_unit_done(tsu);
 940         return;
 941 }
 942
 943 int
 944 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
 945                     unsigned features, int nblk, int blklen,
 946                     srpc_client_rpc_t **rpcpp)
 947 {
 948         srpc_client_rpc_t   *rpc = NULL;
 949         sfw_test_instance_t *tsi = tsu->tsu_instance;
 950
 951         spin_lock(&tsi->tsi_lock);
 952
 953         LASSERT (sfw_test_active(tsi));
 954
 955         if (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 956                 /* pick request from buffer */
 957                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 958                                      srpc_client_rpc_t, crpc_list);
 959                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 960                 cfs_list_del_init(&rpc->crpc_list);
 961         }
 962
 963         spin_unlock(&tsi->tsi_lock);
 964
 965         if (rpc == NULL) {
 966                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 967                                              blklen, sfw_test_rpc_done,
 968                                              sfw_test_rpc_fini, tsu);
 969         } else {
 970                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 971                                      blklen, sfw_test_rpc_done,
 972                                      sfw_test_rpc_fini, tsu);
 973         }
 974
 975         if (rpc == NULL) {
 976                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 977                 return -ENOMEM;
 978         }
 979
 980         rpc->crpc_reqstmsg.msg_ses_feats = features;
 981         *rpcpp = rpc;
 982
 983         return 0;
 984 }
 985
 986 int
 987 sfw_run_test (swi_workitem_t *wi)
 988 {
 989         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 990         sfw_test_instance_t *tsi = tsu->tsu_instance;
 991         srpc_client_rpc_t   *rpc = NULL;
 992
 993         LASSERT (wi == &tsu->tsu_worker);
 994
 995         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 996                 LASSERT (rpc == NULL);
 997                 goto test_done;
 998         }
 999
1000         LASSERT (rpc != NULL);
1001
1002         spin_lock(&tsi->tsi_lock);
1003
1004         if (tsi->tsi_stopping) {
1005                 cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
1006                 spin_unlock(&tsi->tsi_lock);
1007                 goto test_done;
1008         }
1009
1010         if (tsu->tsu_loop > 0)
1011                 tsu->tsu_loop--;
1012
1013         cfs_list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
1014         spin_unlock(&tsi->tsi_lock);
1015
1016         spin_lock(&rpc->crpc_lock);
1017         rpc->crpc_timeout = rpc_timeout;
1018         srpc_post_rpc(rpc);
1019         spin_unlock(&rpc->crpc_lock);
1020         return 0;
1021
1022 test_done:
1023         /*
1024          * No one can schedule me now since:
1025          * - previous RPC, if any, has done and
1026          * - no new RPC is initiated.
1027          * - my batch is still active; no one can run it again now.
1028          * Cancel pending schedules and prevent future schedule attempts:
1029          */
1030         swi_exit_workitem(wi);
1031         sfw_test_unit_done(tsu);
1032         return 1;
1033 }
1034
1035 int
1036 sfw_run_batch (sfw_batch_t *tsb)
1037 {
1038         swi_workitem_t      *wi;
1039         sfw_test_unit_t     *tsu;
1040         sfw_test_instance_t *tsi;
1041
1042         if (sfw_batch_active(tsb)) {
1043                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
1044                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1045                 return 0;
1046         }
1047
1048         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1049                                        sfw_test_instance_t, tsi_list) {
1050                 if (!tsi->tsi_is_client) /* skip server instances */
1051                         continue;
1052
1053                 LASSERT (!tsi->tsi_stopping);
1054                 LASSERT (!sfw_test_active(tsi));
1055
1056                 atomic_inc(&tsb->bat_nactive);
1057
1058                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
1059                                                sfw_test_unit_t, tsu_list) {
1060                         atomic_inc(&tsi->tsi_nactive);
1061                         tsu->tsu_loop = tsi->tsi_loop;
1062                         wi = &tsu->tsu_worker;
1063                         swi_init_workitem(wi, tsu, sfw_run_test,
1064                                           lst_sched_test[\
1065                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1066                         swi_schedule_workitem(wi);
1067                 }
1068         }
1069
1070         return 0;
1071 }
1072
1073 int
1074 sfw_stop_batch (sfw_batch_t *tsb, int force)
1075 {
1076         sfw_test_instance_t *tsi;
1077         srpc_client_rpc_t   *rpc;
1078
1079         if (!sfw_batch_active(tsb)) {
1080                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1081                 return 0;
1082         }
1083
1084         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1085                                        sfw_test_instance_t, tsi_list) {
1086                 spin_lock(&tsi->tsi_lock);
1087
1088                 if (!tsi->tsi_is_client ||
1089                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1090                         spin_unlock(&tsi->tsi_lock);
1091                         continue;
1092                 }
1093
1094                 tsi->tsi_stopping = 1;
1095
1096                 if (!force) {
1097                         spin_unlock(&tsi->tsi_lock);
1098                         continue;
1099                 }
1100
1101                 /* abort launched rpcs in the test */
1102                 cfs_list_for_each_entry_typed(rpc, &tsi->tsi_active_rpcs,
1103                                               srpc_client_rpc_t, crpc_list) {
1104                         spin_lock(&rpc->crpc_lock);
1105
1106                         srpc_abort_rpc(rpc, -EINTR);
1107
1108                         spin_unlock(&rpc->crpc_lock);
1109                 }
1110
1111                 spin_unlock(&tsi->tsi_lock);
1112         }
1113
1114         return 0;
1115 }
1116
1117 int
1118 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1119 {
1120         sfw_test_instance_t *tsi;
1121
1122         if (testidx < 0)
1123                 return -EINVAL;
1124
1125         if (testidx == 0) {
1126                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1127                 return 0;
1128         }
1129
1130         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1131                                        sfw_test_instance_t, tsi_list) {
1132                 if (testidx-- > 1)
1133                         continue;
1134
1135                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1136                 return 0;
1137         }
1138
1139         return -ENOENT;
1140 }
1141
1142 void
1143 sfw_free_pages (srpc_server_rpc_t *rpc)
1144 {
1145         srpc_free_bulk(rpc->srpc_bulk);
1146         rpc->srpc_bulk = NULL;
1147 }
1148
1149 int
1150 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1151                 int sink)
1152 {
1153         LASSERT(rpc->srpc_bulk == NULL);
1154         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1155
1156         rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1157         if (rpc->srpc_bulk == NULL)
1158                 return -ENOMEM;
1159
1160         return 0;
1161 }
1162
1163 int
1164 sfw_add_test (srpc_server_rpc_t *rpc)
1165 {
1166         sfw_session_t     *sn = sfw_data.fw_session;
1167         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1168         srpc_test_reqst_t *request;
1169         int                rc;
1170         sfw_batch_t       *bat;
1171
1172         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1173         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1174
1175         if (request->tsr_loop == 0 ||
1176             request->tsr_concur == 0 ||
1177             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1178             request->tsr_ndest > SFW_MAX_NDESTS ||
1179             (request->tsr_is_client && request->tsr_ndest == 0) ||
1180             request->tsr_concur > SFW_MAX_CONCUR ||
1181             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1182             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1183                 reply->tsr_status = EINVAL;
1184                 return 0;
1185         }
1186
1187         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1188             sfw_find_test_case(request->tsr_service) == NULL) {
1189                 reply->tsr_status = ENOENT;
1190                 return 0;
1191         }
1192
1193         bat = sfw_bid2batch(request->tsr_bid);
1194         if (bat == NULL) {
1195                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1196                         rpc->srpc_scd->scd_svc->sv_name,
1197                         libcfs_id2str(rpc->srpc_peer));
1198                 return -ENOMEM;
1199         }
1200
1201         if (sfw_batch_active(bat)) {
1202                 reply->tsr_status = EBUSY;
1203                 return 0;
1204         }
1205
1206         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1207                 /* rpc will be resumed later in sfw_bulk_ready */
1208                 int     npg = sfw_id_pages(request->tsr_ndest);
1209                 int     len;
1210
1211                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1212                         len = npg * PAGE_CACHE_SIZE;
1213
1214                 } else  {
1215                         len = sizeof(lnet_process_id_packed_t) *
1216                               request->tsr_ndest;
1217                 }
1218
1219                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1220         }
1221
1222         rc = sfw_add_test_instance(bat, rpc);
1223         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1224                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1225                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1226                 request->tsr_is_client ? "client" : "server",
1227                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1228
1229         reply->tsr_status = (rc < 0) ? -rc : rc;
1230         return 0;
1231 }
1232
1233 int
1234 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1235 {
1236         sfw_session_t *sn = sfw_data.fw_session;
1237         int            rc = 0;
1238         sfw_batch_t   *bat;
1239
1240         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1241
1242         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1243                 reply->bar_status = ESRCH;
1244                 return 0;
1245         }
1246
1247         bat = sfw_find_batch(request->bar_bid);
1248         if (bat == NULL) {
1249                 reply->bar_status = ENOENT;
1250                 return 0;
1251         }
1252
1253         switch (request->bar_opc) {
1254         case SRPC_BATCH_OPC_RUN:
1255                 rc = sfw_run_batch(bat);
1256                 break;
1257
1258         case SRPC_BATCH_OPC_STOP:
1259                 rc = sfw_stop_batch(bat, request->bar_arg);
1260                 break;
1261
1262         case SRPC_BATCH_OPC_QUERY:
1263                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1264                 break;
1265
1266         default:
1267                 return -EINVAL; /* drop it */
1268         }
1269
1270         reply->bar_status = (rc < 0) ? -rc : rc;
1271         return 0;
1272 }
1273
1274 int
1275 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1276 {
1277         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1278         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1279         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1280         unsigned        features = LST_FEATS_MASK;
1281         int             rc = 0;
1282
1283         LASSERT(sfw_data.fw_active_srpc == NULL);
1284         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1285
1286         spin_lock(&sfw_data.fw_lock);
1287
1288         if (sfw_data.fw_shuttingdown) {
1289                 spin_unlock(&sfw_data.fw_lock);
1290                 return -ESHUTDOWN;
1291         }
1292
1293         /* Remove timer to avoid racing with it or expiring active session */
1294         if (sfw_del_session_timer() != 0) {
1295                 CERROR("Dropping RPC (%s) from %s: racing with expiry timer.",
1296                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1297                 spin_unlock(&sfw_data.fw_lock);
1298                 return -EAGAIN;
1299         }
1300
1301         sfw_data.fw_active_srpc = rpc;
1302         spin_unlock(&sfw_data.fw_lock);
1303
1304         sfw_unpack_message(request);
1305         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1306
1307         /* rpc module should have checked this */
1308         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1309
1310         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1311             sv->sv_id != SRPC_SERVICE_DEBUG) {
1312                 sfw_session_t *sn = sfw_data.fw_session;
1313
1314                 if (sn != NULL &&
1315                     sn->sn_features != request->msg_ses_feats) {
1316                         CNETERR("Features of framework RPC don't match "
1317                                 "features of current session: %x/%x\n",
1318                                 request->msg_ses_feats, sn->sn_features);
1319                         reply->msg_body.reply.status = EPROTO;
1320                         reply->msg_body.reply.sid    = sn->sn_id;
1321                         goto out;
1322                 }
1323
1324         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1325                 /* NB: at this point, old version will ignore features and
1326                  * create new session anyway, so console should be able
1327                  * to handle this */
1328                 reply->msg_body.reply.status = EPROTO;
1329                 goto out;
1330         }
1331
1332         switch(sv->sv_id) {
1333         default:
1334                 LBUG ();
1335         case SRPC_SERVICE_TEST:
1336                 rc = sfw_add_test(rpc);
1337                 break;
1338
1339         case SRPC_SERVICE_BATCH:
1340                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1341                                        &reply->msg_body.bat_reply);
1342                 break;
1343
1344         case SRPC_SERVICE_QUERY_STAT:
1345                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1346                                    &reply->msg_body.stat_reply);
1347                 break;
1348
1349         case SRPC_SERVICE_DEBUG:
1350                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1351                                        &reply->msg_body.dbg_reply);
1352                 break;
1353
1354         case SRPC_SERVICE_MAKE_SESSION:
1355                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1356                                       &reply->msg_body.mksn_reply);
1357                 break;
1358
1359         case SRPC_SERVICE_REMOVE_SESSION:
1360                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1361                                         &reply->msg_body.rmsn_reply);
1362                 break;
1363         }
1364
1365         if (sfw_data.fw_session != NULL)
1366                 features = sfw_data.fw_session->sn_features;
1367  out:
1368         reply->msg_ses_feats = features;
1369         rpc->srpc_done = sfw_server_rpc_done;
1370         spin_lock(&sfw_data.fw_lock);
1371
1372 #ifdef __KERNEL__
1373         if (!sfw_data.fw_shuttingdown)
1374                 sfw_add_session_timer();
1375 #else
1376         LASSERT(!sfw_data.fw_shuttingdown);
1377         sfw_add_session_timer();
1378 #endif
1379
1380         sfw_data.fw_active_srpc = NULL;
1381         spin_unlock(&sfw_data.fw_lock);
1382         return rc;
1383 }
1384
1385 int
1386 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1387 {
1388         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1389         int                     rc;
1390
1391         LASSERT(rpc->srpc_bulk != NULL);
1392         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1393         LASSERT(sfw_data.fw_active_srpc == NULL);
1394         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1395
1396         spin_lock(&sfw_data.fw_lock);
1397
1398         if (status != 0) {
1399                 CERROR("Bulk transfer failed for RPC: "
1400                        "service %s, peer %s, status %d\n",
1401                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1402                 spin_unlock(&sfw_data.fw_lock);
1403                 return -EIO;
1404         }
1405
1406         if (sfw_data.fw_shuttingdown) {
1407                 spin_unlock(&sfw_data.fw_lock);
1408                 return -ESHUTDOWN;
1409         }
1410
1411         if (sfw_del_session_timer() != 0) {
1412                 CERROR("Dropping RPC (%s) from %s: racing with expiry timer",
1413                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1414                 spin_unlock(&sfw_data.fw_lock);
1415                 return -EAGAIN;
1416         }
1417
1418         sfw_data.fw_active_srpc = rpc;
1419         spin_unlock(&sfw_data.fw_lock);
1420
1421         rc = sfw_add_test(rpc);
1422
1423         spin_lock(&sfw_data.fw_lock);
1424
1425 #ifdef __KERNEL__
1426         if (!sfw_data.fw_shuttingdown)
1427                 sfw_add_session_timer();
1428 #else
1429         LASSERT(!sfw_data.fw_shuttingdown);
1430         sfw_add_session_timer();
1431 #endif
1432
1433         sfw_data.fw_active_srpc = NULL;
1434         spin_unlock(&sfw_data.fw_lock);
1435         return rc;
1436 }
1437
1438 srpc_client_rpc_t *
1439 sfw_create_rpc(lnet_process_id_t peer, int service,
1440                unsigned features, int nbulkiov, int bulklen,
1441                void (*done)(srpc_client_rpc_t *), void *priv)
1442 {
1443         srpc_client_rpc_t *rpc = NULL;
1444
1445         spin_lock(&sfw_data.fw_lock);
1446
1447         LASSERT (!sfw_data.fw_shuttingdown);
1448         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1449
1450         if (nbulkiov == 0 && !cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1451                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1452                                      srpc_client_rpc_t, crpc_list);
1453                 cfs_list_del(&rpc->crpc_list);
1454
1455                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1456                                      done, sfw_client_rpc_fini, priv);
1457         }
1458
1459         spin_unlock(&sfw_data.fw_lock);
1460
1461         if (rpc == NULL) {
1462                 rpc = srpc_create_client_rpc(peer, service,
1463                                              nbulkiov, bulklen, done,
1464                                              nbulkiov != 0 ?  NULL :
1465                                              sfw_client_rpc_fini,
1466                                              priv);
1467         }
1468
1469         if (rpc != NULL) /* "session" is concept in framework */
1470                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1471
1472         return rpc;
1473 }
1474
1475 void
1476 sfw_unpack_message (srpc_msg_t *msg)
1477 {
1478         if (msg->msg_magic == SRPC_MSG_MAGIC)
1479                 return; /* no flipping needed */
1480
1481         /* srpc module should guarantee I wouldn't get crap */
1482         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1483
1484         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1485                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1486
1487                 __swab32s(&req->str_type);
1488                 __swab64s(&req->str_rpyid);
1489                 sfw_unpack_sid(req->str_sid);
1490                 return;
1491         }
1492
1493         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1494                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1495
1496                 __swab32s(&rep->str_status);
1497                 sfw_unpack_sid(rep->str_sid);
1498                 sfw_unpack_fw_counters(rep->str_fw);
1499                 sfw_unpack_rpc_counters(rep->str_rpc);
1500                 sfw_unpack_lnet_counters(rep->str_lnet);
1501                 return;
1502         }
1503
1504         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1505                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1506
1507                 __swab64s(&req->mksn_rpyid);
1508                 __swab32s(&req->mksn_force);
1509                 sfw_unpack_sid(req->mksn_sid);
1510                 return;
1511         }
1512
1513         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1514                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1515
1516                 __swab32s(&rep->mksn_status);
1517                 __swab32s(&rep->mksn_timeout);
1518                 sfw_unpack_sid(rep->mksn_sid);
1519                 return;
1520         }
1521
1522         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1523                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1524
1525                 __swab64s(&req->rmsn_rpyid);
1526                 sfw_unpack_sid(req->rmsn_sid);
1527                 return;
1528         }
1529
1530         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1531                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1532
1533                 __swab32s(&rep->rmsn_status);
1534                 sfw_unpack_sid(rep->rmsn_sid);
1535                 return;
1536         }
1537
1538         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1539                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1540
1541                 __swab64s(&req->dbg_rpyid);
1542                 __swab32s(&req->dbg_flags);
1543                 sfw_unpack_sid(req->dbg_sid);
1544                 return;
1545         }
1546
1547         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1548                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1549
1550                 __swab32s(&rep->dbg_nbatch);
1551                 __swab32s(&rep->dbg_timeout);
1552                 sfw_unpack_sid(rep->dbg_sid);
1553                 return;
1554         }
1555
1556         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1557                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1558
1559                 __swab32s(&req->bar_opc);
1560                 __swab64s(&req->bar_rpyid);
1561                 __swab32s(&req->bar_testidx);
1562                 __swab32s(&req->bar_arg);
1563                 sfw_unpack_sid(req->bar_sid);
1564                 __swab64s(&req->bar_bid.bat_id);
1565                 return;
1566         }
1567
1568         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1569                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1570
1571                 __swab32s(&rep->bar_status);
1572                 sfw_unpack_sid(rep->bar_sid);
1573                 return;
1574         }
1575
1576         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1577                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1578
1579                 __swab64s(&req->tsr_rpyid);
1580                 __swab64s(&req->tsr_bulkid);
1581                 __swab32s(&req->tsr_loop);
1582                 __swab32s(&req->tsr_ndest);
1583                 __swab32s(&req->tsr_concur);
1584                 __swab32s(&req->tsr_service);
1585                 sfw_unpack_sid(req->tsr_sid);
1586                 __swab64s(&req->tsr_bid.bat_id);
1587                 return;
1588         }
1589
1590         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1591                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1592
1593                 __swab32s(&rep->tsr_status);
1594                 sfw_unpack_sid(rep->tsr_sid);
1595                 return;
1596         }
1597
1598         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1599                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1600
1601                 __swab64s(&req->join_rpyid);
1602                 sfw_unpack_sid(req->join_sid);
1603                 return;
1604         }
1605
1606         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1607                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1608
1609                 __swab32s(&rep->join_status);
1610                 __swab32s(&rep->join_timeout);
1611                 sfw_unpack_sid(rep->join_sid);
1612                 return;
1613         }
1614
1615         LBUG ();
1616         return;
1617 }
1618
1619 void
1620 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1621 {
1622         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1623         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1624
1625         spin_lock(&rpc->crpc_lock);
1626         srpc_abort_rpc(rpc, -EINTR);
1627         spin_unlock(&rpc->crpc_lock);
1628         return;
1629 }
1630
1631 void
1632 sfw_post_rpc (srpc_client_rpc_t *rpc)
1633 {
1634         spin_lock(&rpc->crpc_lock);
1635
1636         LASSERT (!rpc->crpc_closed);
1637         LASSERT (!rpc->crpc_aborted);
1638         LASSERT (cfs_list_empty(&rpc->crpc_list));
1639         LASSERT (!sfw_data.fw_shuttingdown);
1640
1641         rpc->crpc_timeout = rpc_timeout;
1642         srpc_post_rpc(rpc);
1643
1644         spin_unlock(&rpc->crpc_lock);
1645         return;
1646 }
1647
1648 static srpc_service_t sfw_services[] =
1649 {
1650         {
1651                 /* sv_id */    SRPC_SERVICE_DEBUG,
1652                 /* sv_name */  "debug",
1653                 0
1654         },
1655         {
1656                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1657                 /* sv_name */  "query stats",
1658                 0
1659         },
1660         {
1661                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1662                 /* sv_name */  "make session",
1663                 0
1664         },
1665         {
1666                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1667                 /* sv_name */  "remove session",
1668                 0
1669         },
1670         {
1671                 /* sv_id */    SRPC_SERVICE_BATCH,
1672                 /* sv_name */  "batch service",
1673                 0
1674         },
1675         {
1676                 /* sv_id */    SRPC_SERVICE_TEST,
1677                 /* sv_name */  "test service",
1678                 0
1679         },
1680         {
1681                 /* sv_id */    0,
1682                 /* sv_name */  NULL,
1683                 0
1684         }
1685 };
1686
1687 extern sfw_test_client_ops_t ping_test_client;
1688 extern srpc_service_t        ping_test_service;
1689 extern void ping_init_test_client(void);
1690 extern void ping_init_test_service(void);
1691
1692 extern sfw_test_client_ops_t brw_test_client;
1693 extern srpc_service_t        brw_test_service;
1694 extern void brw_init_test_client(void);
1695 extern void brw_init_test_service(void);
1696
1697
1698 int
1699 sfw_startup (void)
1700 {
1701         int              i;
1702         int              rc;
1703         int              error;
1704         srpc_service_t  *sv;
1705         sfw_test_case_t *tsc;
1706
1707 #ifndef __KERNEL__
1708         char *s;
1709
1710         s = getenv("SESSION_TIMEOUT");
1711         session_timeout = s != NULL ? atoi(s) : session_timeout;
1712
1713         s = getenv("RPC_TIMEOUT");
1714         rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1715 #endif
1716
1717         if (session_timeout < 0) {
1718                 CERROR ("Session timeout must be non-negative: %d\n",
1719                         session_timeout);
1720                 return -EINVAL;
1721         }
1722
1723         if (rpc_timeout < 0) {
1724                 CERROR ("RPC timeout must be non-negative: %d\n",
1725                         rpc_timeout);
1726                 return -EINVAL;
1727         }
1728
1729         if (session_timeout == 0)
1730                 CWARN ("Zero session_timeout specified "
1731                        "- test sessions never expire.\n");
1732
1733         if (rpc_timeout == 0)
1734                 CWARN ("Zero rpc_timeout specified "
1735                        "- test RPC never expire.\n");
1736
1737         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1738
1739         sfw_data.fw_session     = NULL;
1740         sfw_data.fw_active_srpc = NULL;
1741         spin_lock_init(&sfw_data.fw_lock);
1742         atomic_set(&sfw_data.fw_nzombies, 0);
1743         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1744         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1745         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1746
1747         brw_init_test_client();
1748         brw_init_test_service();
1749         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1750         LASSERT (rc == 0);
1751
1752         ping_init_test_client();
1753         ping_init_test_service();
1754         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1755         LASSERT (rc == 0);
1756
1757         error = 0;
1758         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1759                                        sfw_test_case_t, tsc_list) {
1760                 sv = tsc->tsc_srv_service;
1761
1762                 rc = srpc_add_service(sv);
1763                 LASSERT (rc != -EBUSY);
1764                 if (rc != 0) {
1765                         CWARN ("Failed to add %s service: %d\n",
1766                                sv->sv_name, rc);
1767                         error = rc;
1768                 }
1769         }
1770
1771         for (i = 0; ; i++) {
1772                 sv = &sfw_services[i];
1773                 if (sv->sv_name == NULL) break;
1774
1775                 sv->sv_bulk_ready = NULL;
1776                 sv->sv_handler    = sfw_handle_server_rpc;
1777                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1778                 if (sv->sv_id == SRPC_SERVICE_TEST)
1779                         sv->sv_bulk_ready = sfw_bulk_ready;
1780
1781                 rc = srpc_add_service(sv);
1782                 LASSERT (rc != -EBUSY);
1783                 if (rc != 0) {
1784                         CWARN ("Failed to add %s service: %d\n",
1785                                sv->sv_name, rc);
1786                         error = rc;
1787                 }
1788
1789                 /* about to sfw_shutdown, no need to add buffer */
1790                 if (error) continue;
1791
1792                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1793                 if (rc != 0) {
1794                         CWARN("Failed to reserve enough buffers: "
1795                               "service %s, %d needed: %d\n",
1796                               sv->sv_name, sv->sv_wi_total, rc);
1797                         error = -ENOMEM;
1798                 }
1799         }
1800
1801         if (error != 0)
1802                 sfw_shutdown();
1803         return error;
1804 }
1805
1806 void
1807 sfw_shutdown (void)
1808 {
1809         srpc_service_t  *sv;
1810         sfw_test_case_t *tsc;
1811         int              i;
1812
1813         spin_lock(&sfw_data.fw_lock);
1814
1815         sfw_data.fw_shuttingdown = 1;
1816 #ifdef __KERNEL__
1817         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1818                        "waiting for active RPC to finish.\n");
1819 #else
1820         LASSERT (sfw_data.fw_active_srpc == NULL);
1821 #endif
1822
1823         if (sfw_del_session_timer() != 0)
1824                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1825                                "waiting for session timer to explode.\n");
1826
1827         sfw_deactivate_session();
1828         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1829                        sfw_data.fw_lock,
1830                        "waiting for %d zombie sessions to die.\n",
1831                        atomic_read(&sfw_data.fw_nzombies));
1832
1833         spin_unlock(&sfw_data.fw_lock);
1834
1835         for (i = 0; ; i++) {
1836                 sv = &sfw_services[i];
1837                 if (sv->sv_name == NULL)
1838                         break;
1839
1840                 srpc_shutdown_service(sv);
1841                 srpc_remove_service(sv);
1842         }
1843
1844         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1845                                        sfw_test_case_t, tsc_list) {
1846                 sv = tsc->tsc_srv_service;
1847                 srpc_shutdown_service(sv);
1848                 srpc_remove_service(sv);
1849         }
1850
1851         while (!cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1852                 srpc_client_rpc_t *rpc;
1853
1854                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1855                                      srpc_client_rpc_t, crpc_list);
1856                 cfs_list_del(&rpc->crpc_list);
1857
1858                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1859         }
1860
1861         for (i = 0; ; i++) {
1862                 sv = &sfw_services[i];
1863                 if (sv->sv_name == NULL)
1864                         break;
1865
1866                 srpc_wait_service_shutdown(sv);
1867         }
1868
1869         while (!cfs_list_empty(&sfw_data.fw_tests)) {
1870                 tsc = cfs_list_entry(sfw_data.fw_tests.next,
1871                                      sfw_test_case_t, tsc_list);
1872
1873                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1874
1875                 cfs_list_del(&tsc->tsc_list);
1876                 LIBCFS_FREE(tsc, sizeof(*tsc));
1877         }
1878
1879         return;
1880 }