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, 2014, 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 module_param(session_timeout, int, 0444);
  50 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
  51
  52 static int rpc_timeout = 64;
  53 module_param(rpc_timeout, int, 0644);
  54 MODULE_PARM_DESC(rpc_timeout, "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 static struct smoketest_framework {
 107         /* RPCs to be recycled */
 108         struct list_head        fw_zombie_rpcs;
 109         /* stopping sessions */
 110         struct list_head        fw_zombie_sessions;
 111         /* registered test cases */
 112         struct list_head        fw_tests;
 113         /* # zombie sessions */
 114         atomic_t                fw_nzombies;
 115         /* serialise */
 116         spinlock_t              fw_lock;
 117         /* _the_ session */
 118         sfw_session_t           *fw_session;
 119         /* shutdown in progress */
 120         int                     fw_shuttingdown;
 121         /* running RPC */
 122         srpc_server_rpc_t       *fw_active_srpc;
 123 } sfw_data;
 124
 125 /* forward ref's */
 126 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 127 void sfw_destroy_session (sfw_session_t *sn);
 128
 129 static inline sfw_test_case_t *
 130 sfw_find_test_case(int id)
 131 {
 132         sfw_test_case_t *tsc;
 133
 134         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 135         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 136
 137         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 138                 if (tsc->tsc_srv_service->sv_id == id)
 139                         return tsc;
 140         }
 141
 142         return NULL;
 143 }
 144
 145 static int
 146 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 147 {
 148         sfw_test_case_t *tsc;
 149
 150         if (sfw_find_test_case(service->sv_id) != NULL) {
 151                 CERROR ("Failed to register test %s (%d)\n",
 152                         service->sv_name, service->sv_id);
 153                 return -EEXIST;
 154         }
 155
 156         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 157         if (tsc == NULL)
 158                 return -ENOMEM;
 159
 160         tsc->tsc_cli_ops     = cliops;
 161         tsc->tsc_srv_service = service;
 162
 163         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 164         return 0;
 165 }
 166
 167 static void
 168 sfw_add_session_timer (void)
 169 {
 170         sfw_session_t *sn = sfw_data.fw_session;
 171         stt_timer_t   *timer = &sn->sn_timer;
 172
 173         LASSERT (!sfw_data.fw_shuttingdown);
 174
 175         if (sn == NULL || sn->sn_timeout == 0)
 176                 return;
 177
 178         LASSERT (!sn->sn_timer_active);
 179
 180         sn->sn_timer_active = 1;
 181         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 182                                           cfs_time_current_sec());
 183         stt_add_timer(timer);
 184         return;
 185 }
 186
 187 static int
 188 sfw_del_session_timer (void)
 189 {
 190         sfw_session_t *sn = sfw_data.fw_session;
 191
 192         if (sn == NULL || !sn->sn_timer_active)
 193                 return 0;
 194
 195         LASSERT (sn->sn_timeout != 0);
 196
 197         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 198                 sn->sn_timer_active = 0;
 199                 return 0;
 200         }
 201
 202         return EBUSY; /* racing with sfw_session_expired() */
 203 }
 204
 205 /* called with sfw_data.fw_lock held */
 206 static void
 207 sfw_deactivate_session (void)
 208 __must_hold(&sfw_data.fw_lock)
 209 {
 210         sfw_session_t *sn = sfw_data.fw_session;
 211         int            nactive = 0;
 212         sfw_batch_t   *tsb;
 213         sfw_test_case_t *tsc;
 214
 215         if (sn == NULL) return;
 216
 217         LASSERT(!sn->sn_timer_active);
 218
 219         sfw_data.fw_session = NULL;
 220         atomic_inc(&sfw_data.fw_nzombies);
 221         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 222
 223         spin_unlock(&sfw_data.fw_lock);
 224
 225         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 226                 srpc_abort_service(tsc->tsc_srv_service);
 227         }
 228
 229         spin_lock(&sfw_data.fw_lock);
 230
 231         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 232                 if (sfw_batch_active(tsb)) {
 233                         nactive++;
 234                         sfw_stop_batch(tsb, 1);
 235                 }
 236         }
 237
 238         if (nactive != 0)
 239                 return; /* wait for active batches to stop */
 240
 241         list_del_init(&sn->sn_list);
 242         spin_unlock(&sfw_data.fw_lock);
 243
 244         sfw_destroy_session(sn);
 245
 246         spin_lock(&sfw_data.fw_lock);
 247 }
 248
 249
 250 static void
 251 sfw_session_expired (void *data)
 252 {
 253         sfw_session_t *sn = data;
 254
 255         spin_lock(&sfw_data.fw_lock);
 256
 257         LASSERT (sn->sn_timer_active);
 258         LASSERT (sn == sfw_data.fw_session);
 259
 260         CWARN ("Session expired! sid: %s-%llu, name: %s\n",
 261                libcfs_nid2str(sn->sn_id.ses_nid),
 262                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 263
 264         sn->sn_timer_active = 0;
 265         sfw_deactivate_session();
 266
 267         spin_unlock(&sfw_data.fw_lock);
 268 }
 269
 270 static inline void
 271 sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
 272                  unsigned features, const char *name)
 273 {
 274         stt_timer_t *timer = &sn->sn_timer;
 275
 276         memset(sn, 0, sizeof(sfw_session_t));
 277         INIT_LIST_HEAD(&sn->sn_list);
 278         INIT_LIST_HEAD(&sn->sn_batches);
 279         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 280         atomic_set(&sn->sn_brw_errors, 0);
 281         atomic_set(&sn->sn_ping_errors, 0);
 282         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 283
 284         sn->sn_timer_active = 0;
 285         sn->sn_id           = sid;
 286         sn->sn_features     = features;
 287         sn->sn_timeout      = session_timeout;
 288         sn->sn_started      = cfs_time_current();
 289
 290         timer->stt_data = sn;
 291         timer->stt_func = sfw_session_expired;
 292         INIT_LIST_HEAD(&timer->stt_list);
 293 }
 294
 295 /* completion handler for incoming framework RPCs */
 296 static void
 297 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 298 {
 299         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 300         int                     status  = rpc->srpc_status;
 301
 302         CDEBUG (D_NET,
 303                 "Incoming framework RPC done: "
 304                 "service %s, peer %s, status %s:%d\n",
 305                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 306                 swi_state2str(rpc->srpc_wi.swi_state),
 307                 status);
 308
 309         if (rpc->srpc_bulk != NULL)
 310                 sfw_free_pages(rpc);
 311         return;
 312 }
 313
 314 static void
 315 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 316 {
 317         LASSERT(rpc->crpc_bulk.bk_niov == 0);
 318         LASSERT(list_empty(&rpc->crpc_list));
 319         LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
 320
 321         CDEBUG(D_NET, "Outgoing framework RPC done: "
 322                "service %d, peer %s, status %s:%d:%d\n",
 323                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 324                swi_state2str(rpc->crpc_wi.swi_state),
 325                rpc->crpc_aborted, rpc->crpc_status);
 326
 327         spin_lock(&sfw_data.fw_lock);
 328
 329         /* my callers must finish all RPCs before shutting me down */
 330         LASSERT(!sfw_data.fw_shuttingdown);
 331         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 332
 333         spin_unlock(&sfw_data.fw_lock);
 334 }
 335
 336 static sfw_batch_t *
 337 sfw_find_batch (lst_bid_t bid)
 338 {
 339         sfw_session_t *sn = sfw_data.fw_session;
 340         sfw_batch_t   *bat;
 341
 342         LASSERT(sn != NULL);
 343
 344         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 345                 if (bat->bat_id.bat_id == bid.bat_id)
 346                         return bat;
 347         }
 348
 349         return NULL;
 350 }
 351
 352 static sfw_batch_t *
 353 sfw_bid2batch (lst_bid_t bid)
 354 {
 355         sfw_session_t *sn = sfw_data.fw_session;
 356         sfw_batch_t   *bat;
 357
 358         LASSERT (sn != NULL);
 359
 360         bat = sfw_find_batch(bid);
 361         if (bat != NULL)
 362                 return bat;
 363
 364         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 365         if (bat == NULL)
 366                 return NULL;
 367
 368         bat->bat_error   = 0;
 369         bat->bat_session = sn;
 370         bat->bat_id      = bid;
 371         atomic_set(&bat->bat_nactive, 0);
 372         INIT_LIST_HEAD(&bat->bat_tests);
 373
 374         list_add_tail(&bat->bat_list, &sn->sn_batches);
 375         return bat;
 376 }
 377
 378 static int
 379 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 380 {
 381         sfw_session_t  *sn = sfw_data.fw_session;
 382         sfw_counters_t *cnt = &reply->str_fw;
 383         sfw_batch_t    *bat;
 384         struct timeval  tv;
 385
 386         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 387
 388         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 389                 reply->str_status = EINVAL;
 390                 return 0;
 391         }
 392
 393         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 394                 reply->str_status = ESRCH;
 395                 return 0;
 396         }
 397
 398         lnet_counters_get(&reply->str_lnet);
 399         srpc_get_counters(&reply->str_rpc);
 400
 401         /* send over the msecs since the session was started
 402          - with 32 bits to send, this is ~49 days */
 403         cfs_duration_usec(cfs_time_sub(cfs_time_current(),
 404                                        sn->sn_started), &tv);
 405
 406         cnt->running_ms      = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
 407         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 408         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 409         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 410
 411         cnt->active_batches = 0;
 412         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 413                 if (atomic_read(&bat->bat_nactive) > 0)
 414                         cnt->active_batches++;
 415         }
 416
 417         reply->str_status = 0;
 418         return 0;
 419 }
 420
 421 int
 422 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 423 {
 424         sfw_session_t *sn = sfw_data.fw_session;
 425         srpc_msg_t    *msg = container_of(request, srpc_msg_t,
 426                                           msg_body.mksn_reqst);
 427         int            cplen = 0;
 428
 429         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 430                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 431                 reply->mksn_status = EINVAL;
 432                 return 0;
 433         }
 434
 435         if (sn != NULL) {
 436                 reply->mksn_status  = 0;
 437                 reply->mksn_sid     = sn->sn_id;
 438                 reply->mksn_timeout = sn->sn_timeout;
 439
 440                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 441                         atomic_inc(&sn->sn_refcount);
 442                         return 0;
 443                 }
 444
 445                 if (!request->mksn_force) {
 446                         reply->mksn_status = EBUSY;
 447                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 448                                         sizeof(reply->mksn_name));
 449                         if (cplen >= sizeof(reply->mksn_name))
 450                                 return -E2BIG;
 451                         return 0;
 452                 }
 453         }
 454
 455         /* reject the request if it requires unknown features
 456          * NB: old version will always accept all features because it's not
 457          * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
 458          * harmless because it will return zero feature to console, and it's
 459          * console's responsibility to make sure all nodes in a session have
 460          * same feature mask. */
 461         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 462                 reply->mksn_status = EPROTO;
 463                 return 0;
 464         }
 465
 466         /* brand new or create by force */
 467         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 468         if (sn == NULL) {
 469                 CERROR("dropping RPC mksn under memory pressure\n");
 470                 return -ENOMEM;
 471         }
 472
 473         sfw_init_session(sn, request->mksn_sid,
 474                          msg->msg_ses_feats, &request->mksn_name[0]);
 475
 476         spin_lock(&sfw_data.fw_lock);
 477
 478         sfw_deactivate_session();
 479         LASSERT(sfw_data.fw_session == NULL);
 480         sfw_data.fw_session = sn;
 481
 482         spin_unlock(&sfw_data.fw_lock);
 483
 484         reply->mksn_status  = 0;
 485         reply->mksn_sid     = sn->sn_id;
 486         reply->mksn_timeout = sn->sn_timeout;
 487         return 0;
 488 }
 489
 490 static int
 491 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 492 {
 493         sfw_session_t *sn = sfw_data.fw_session;
 494
 495         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 496
 497         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 498                 reply->rmsn_status = EINVAL;
 499                 return 0;
 500         }
 501
 502         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 503                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 504                 return 0;
 505         }
 506
 507         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 508                 reply->rmsn_status = 0;
 509                 return 0;
 510         }
 511
 512         spin_lock(&sfw_data.fw_lock);
 513         sfw_deactivate_session();
 514         spin_unlock(&sfw_data.fw_lock);
 515
 516         reply->rmsn_status = 0;
 517         reply->rmsn_sid    = LST_INVALID_SID;
 518         LASSERT(sfw_data.fw_session == NULL);
 519         return 0;
 520 }
 521
 522 static int
 523 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 524 {
 525         sfw_session_t *sn = sfw_data.fw_session;
 526
 527         if (sn == NULL) {
 528                 reply->dbg_status = ESRCH;
 529                 reply->dbg_sid    = LST_INVALID_SID;
 530                 return 0;
 531         }
 532
 533         reply->dbg_status  = 0;
 534         reply->dbg_sid     = sn->sn_id;
 535         reply->dbg_timeout = sn->sn_timeout;
 536         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 537             >= sizeof(reply->dbg_name))
 538                 return -E2BIG;
 539
 540         return 0;
 541 }
 542
 543 static void
 544 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 545 {
 546         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 547         sfw_test_instance_t *tsi = tsu->tsu_instance;
 548
 549         /* Called with hold of tsi->tsi_lock */
 550         LASSERT(list_empty(&rpc->crpc_list));
 551         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 552 }
 553
 554 static inline int
 555 sfw_test_buffers(sfw_test_instance_t *tsi)
 556 {
 557         struct sfw_test_case    *tsc;
 558         struct srpc_service     *svc;
 559         int                     nbuf;
 560
 561         LASSERT(tsi != NULL);
 562         tsc = sfw_find_test_case(tsi->tsi_service);
 563         LASSERT(tsc != NULL);
 564         svc = tsc->tsc_srv_service;
 565         LASSERT(svc != NULL);
 566
 567         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 568         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 569 }
 570
 571 static int
 572 sfw_load_test(struct sfw_test_instance *tsi)
 573 {
 574         struct sfw_test_case    *tsc;
 575         struct srpc_service     *svc;
 576         int                     nbuf;
 577         int                     rc;
 578
 579         LASSERT(tsi != NULL);
 580         tsc = sfw_find_test_case(tsi->tsi_service);
 581         nbuf = sfw_test_buffers(tsi);
 582         LASSERT(tsc != NULL);
 583         svc = tsc->tsc_srv_service;
 584
 585         if (tsi->tsi_is_client) {
 586                 tsi->tsi_ops = tsc->tsc_cli_ops;
 587                 return 0;
 588         }
 589
 590         rc = srpc_service_add_buffers(svc, nbuf);
 591         if (rc != 0) {
 592                 CWARN("Failed to reserve enough buffers: "
 593                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 594                 /* NB: this error handler is not strictly correct, because
 595                  * it may release more buffers than already allocated,
 596                  * but it doesn't matter because request portal should
 597                  * be lazy portal and will grow buffers if necessary. */
 598                 srpc_service_remove_buffers(svc, nbuf);
 599                 return -ENOMEM;
 600         }
 601
 602         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 603                nbuf * (srpc_serv_is_framework(svc) ?
 604                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 605         return 0;
 606 }
 607
 608 static void
 609 sfw_unload_test(struct sfw_test_instance *tsi)
 610 {
 611         struct sfw_test_case *tsc;
 612
 613         LASSERT(tsi != NULL);
 614         tsc = sfw_find_test_case(tsi->tsi_service);
 615         LASSERT(tsc != NULL);
 616
 617         if (tsi->tsi_is_client)
 618                 return;
 619
 620         /* shrink buffers, because request portal is lazy portal
 621          * which can grow buffers at runtime so we may leave
 622          * some buffers behind, but never mind... */
 623         srpc_service_remove_buffers(tsc->tsc_srv_service,
 624                                     sfw_test_buffers(tsi));
 625         return;
 626 }
 627
 628 static void
 629 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 630 {
 631         srpc_client_rpc_t *rpc;
 632         sfw_test_unit_t   *tsu;
 633
 634         if (!tsi->tsi_is_client) goto clean;
 635
 636         tsi->tsi_ops->tso_fini(tsi);
 637
 638         LASSERT(!tsi->tsi_stopping);
 639         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 640         LASSERT(!sfw_test_active(tsi));
 641
 642         while (!list_empty(&tsi->tsi_units)) {
 643                 tsu = list_entry(tsi->tsi_units.next,
 644                                  sfw_test_unit_t, tsu_list);
 645                 list_del(&tsu->tsu_list);
 646                 LIBCFS_FREE(tsu, sizeof(*tsu));
 647         }
 648
 649         while (!list_empty(&tsi->tsi_free_rpcs)) {
 650                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 651                                  srpc_client_rpc_t, crpc_list);
 652                 list_del(&rpc->crpc_list);
 653                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 654         }
 655
 656 clean:
 657         sfw_unload_test(tsi);
 658         LIBCFS_FREE(tsi, sizeof(*tsi));
 659         return;
 660 }
 661
 662 static void
 663 sfw_destroy_batch (sfw_batch_t *tsb)
 664 {
 665         sfw_test_instance_t *tsi;
 666
 667         LASSERT(!sfw_batch_active(tsb));
 668         LASSERT(list_empty(&tsb->bat_list));
 669
 670         while (!list_empty(&tsb->bat_tests)) {
 671                 tsi = list_entry(tsb->bat_tests.next,
 672                                  sfw_test_instance_t, tsi_list);
 673                 list_del_init(&tsi->tsi_list);
 674                 sfw_destroy_test_instance(tsi);
 675         }
 676
 677         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 678         return;
 679 }
 680
 681 void
 682 sfw_destroy_session (sfw_session_t *sn)
 683 {
 684         sfw_batch_t *batch;
 685
 686         LASSERT(list_empty(&sn->sn_list));
 687         LASSERT(sn != sfw_data.fw_session);
 688
 689         while (!list_empty(&sn->sn_batches)) {
 690                 batch = list_entry(sn->sn_batches.next,
 691                                    sfw_batch_t, bat_list);
 692                 list_del_init(&batch->bat_list);
 693                 sfw_destroy_batch(batch);
 694         }
 695
 696         LIBCFS_FREE(sn, sizeof(*sn));
 697         atomic_dec(&sfw_data.fw_nzombies);
 698         return;
 699 }
 700
 701 static void
 702 sfw_unpack_addtest_req(srpc_msg_t *msg)
 703 {
 704         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 705
 706         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 707         LASSERT (req->tsr_is_client);
 708
 709         if (msg->msg_magic == SRPC_MSG_MAGIC)
 710                 return; /* no flipping needed */
 711
 712         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 713
 714         if (req->tsr_service == SRPC_SERVICE_BRW) {
 715                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 716                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 717
 718                         __swab32s(&bulk->blk_opc);
 719                         __swab32s(&bulk->blk_npg);
 720                         __swab32s(&bulk->blk_flags);
 721
 722                 } else {
 723                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 724
 725                         __swab16s(&bulk->blk_opc);
 726                         __swab16s(&bulk->blk_flags);
 727                         __swab32s(&bulk->blk_offset);
 728                         __swab32s(&bulk->blk_len);
 729                 }
 730
 731                 return;
 732         }
 733
 734         if (req->tsr_service == SRPC_SERVICE_PING) {
 735                 test_ping_req_t *ping = &req->tsr_u.ping;
 736
 737                 __swab32s(&ping->png_size);
 738                 __swab32s(&ping->png_flags);
 739                 return;
 740         }
 741
 742         LBUG();
 743         return;
 744 }
 745
 746 static int
 747 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 748 {
 749         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 750         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 751         srpc_bulk_t         *bk = rpc->srpc_bulk;
 752         int                  ndest = req->tsr_ndest;
 753         sfw_test_unit_t     *tsu;
 754         sfw_test_instance_t *tsi;
 755         int                  i;
 756         int                  rc;
 757
 758         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 759         if (tsi == NULL) {
 760                 CERROR ("Can't allocate test instance for batch: %llu\n",
 761                         tsb->bat_id.bat_id);
 762                 return -ENOMEM;
 763         }
 764
 765         spin_lock_init(&tsi->tsi_lock);
 766         atomic_set(&tsi->tsi_nactive, 0);
 767         INIT_LIST_HEAD(&tsi->tsi_units);
 768         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 769         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 770
 771         tsi->tsi_stopping      = 0;
 772         tsi->tsi_batch         = tsb;
 773         tsi->tsi_loop          = req->tsr_loop;
 774         tsi->tsi_concur        = req->tsr_concur;
 775         tsi->tsi_service       = req->tsr_service;
 776         tsi->tsi_is_client     = !!(req->tsr_is_client);
 777         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 778
 779         rc = sfw_load_test(tsi);
 780         if (rc != 0) {
 781                 LIBCFS_FREE(tsi, sizeof(*tsi));
 782                 return rc;
 783         }
 784
 785         LASSERT (!sfw_batch_active(tsb));
 786
 787         if (!tsi->tsi_is_client) {
 788                 /* it's test server, just add it to tsb */
 789                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 790                 return 0;
 791         }
 792
 793         LASSERT (bk != NULL);
 794         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 795         LASSERT((unsigned int)bk->bk_len >=
 796                 sizeof(lnet_process_id_packed_t) * ndest);
 797
 798         sfw_unpack_addtest_req(msg);
 799         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 800
 801         for (i = 0; i < ndest; i++) {
 802                 lnet_process_id_packed_t *dests;
 803                 lnet_process_id_packed_t  id;
 804                 int                       j;
 805
 806                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 807                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 808                 id = dests[i % SFW_ID_PER_PAGE];
 809                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 810                         sfw_unpack_id(id);
 811
 812                 for (j = 0; j < tsi->tsi_concur; j++) {
 813                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 814                         if (tsu == NULL) {
 815                                 rc = -ENOMEM;
 816                                 CERROR ("Can't allocate tsu for %d\n",
 817                                         tsi->tsi_service);
 818                                 goto error;
 819                         }
 820
 821                         tsu->tsu_dest.nid = id.nid;
 822                         tsu->tsu_dest.pid = id.pid;
 823                         tsu->tsu_instance = tsi;
 824                         tsu->tsu_private  = NULL;
 825                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 826                 }
 827         }
 828
 829         rc = tsi->tsi_ops->tso_init(tsi);
 830         if (rc == 0) {
 831                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 832                 return 0;
 833         }
 834
 835 error:
 836         LASSERT(rc != 0);
 837         sfw_destroy_test_instance(tsi);
 838         return rc;
 839 }
 840
 841 static void
 842 sfw_test_unit_done (sfw_test_unit_t *tsu)
 843 {
 844         sfw_test_instance_t *tsi = tsu->tsu_instance;
 845         sfw_batch_t         *tsb = tsi->tsi_batch;
 846         sfw_session_t       *sn = tsb->bat_session;
 847
 848         LASSERT (sfw_test_active(tsi));
 849
 850         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 851                 return;
 852
 853         /* the test instance is done */
 854         spin_lock(&tsi->tsi_lock);
 855
 856         tsi->tsi_stopping = 0;
 857
 858         spin_unlock(&tsi->tsi_lock);
 859
 860         spin_lock(&sfw_data.fw_lock);
 861
 862         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 863             sn == sfw_data.fw_session) {                  /* sn also active */
 864                 spin_unlock(&sfw_data.fw_lock);
 865                 return;
 866         }
 867
 868         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 869
 870         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 871                 if (sfw_batch_active(tsb)) {
 872                         spin_unlock(&sfw_data.fw_lock);
 873                         return;
 874                 }
 875         }
 876
 877         list_del_init(&sn->sn_list);
 878         spin_unlock(&sfw_data.fw_lock);
 879
 880         sfw_destroy_session(sn);
 881         return;
 882 }
 883
 884 static void
 885 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 886 {
 887         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 888         sfw_test_instance_t *tsi = tsu->tsu_instance;
 889         int                  done = 0;
 890
 891         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 892
 893         spin_lock(&tsi->tsi_lock);
 894
 895         LASSERT(sfw_test_active(tsi));
 896         LASSERT(!list_empty(&rpc->crpc_list));
 897
 898         list_del_init(&rpc->crpc_list);
 899
 900         /* batch is stopping or loop is done or get error */
 901         if (tsi->tsi_stopping ||
 902             tsu->tsu_loop == 0 ||
 903             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 904                 done = 1;
 905
 906         /* dec ref for poster */
 907         srpc_client_rpc_decref(rpc);
 908
 909         spin_unlock(&tsi->tsi_lock);
 910
 911         if (!done) {
 912                 swi_schedule_workitem(&tsu->tsu_worker);
 913                 return;
 914         }
 915
 916         sfw_test_unit_done(tsu);
 917         return;
 918 }
 919
 920 int
 921 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
 922                     unsigned features, int nblk, int blklen,
 923                     srpc_client_rpc_t **rpcpp)
 924 {
 925         srpc_client_rpc_t   *rpc = NULL;
 926         sfw_test_instance_t *tsi = tsu->tsu_instance;
 927
 928         spin_lock(&tsi->tsi_lock);
 929
 930         LASSERT (sfw_test_active(tsi));
 931
 932         if (!list_empty(&tsi->tsi_free_rpcs)) {
 933                 /* pick request from buffer */
 934                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 935                                  srpc_client_rpc_t, crpc_list);
 936                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 937                 list_del_init(&rpc->crpc_list);
 938         }
 939
 940         spin_unlock(&tsi->tsi_lock);
 941
 942         if (rpc == NULL) {
 943                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 944                                              blklen, sfw_test_rpc_done,
 945                                              sfw_test_rpc_fini, tsu);
 946         } else {
 947                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 948                                      blklen, sfw_test_rpc_done,
 949                                      sfw_test_rpc_fini, tsu);
 950         }
 951
 952         if (rpc == NULL) {
 953                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 954                 return -ENOMEM;
 955         }
 956
 957         rpc->crpc_reqstmsg.msg_ses_feats = features;
 958         *rpcpp = rpc;
 959
 960         return 0;
 961 }
 962
 963 static int
 964 sfw_run_test (swi_workitem_t *wi)
 965 {
 966         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 967         sfw_test_instance_t *tsi = tsu->tsu_instance;
 968         srpc_client_rpc_t   *rpc = NULL;
 969
 970         LASSERT (wi == &tsu->tsu_worker);
 971
 972         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 973                 LASSERT (rpc == NULL);
 974                 goto test_done;
 975         }
 976
 977         LASSERT (rpc != NULL);
 978
 979         spin_lock(&tsi->tsi_lock);
 980
 981         if (tsi->tsi_stopping) {
 982                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 983                 spin_unlock(&tsi->tsi_lock);
 984                 goto test_done;
 985         }
 986
 987         if (tsu->tsu_loop > 0)
 988                 tsu->tsu_loop--;
 989
 990         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 991         spin_unlock(&tsi->tsi_lock);
 992
 993         spin_lock(&rpc->crpc_lock);
 994         rpc->crpc_timeout = rpc_timeout;
 995         srpc_post_rpc(rpc);
 996         spin_unlock(&rpc->crpc_lock);
 997         return 0;
 998
 999 test_done:
1000         /*
1001          * No one can schedule me now since:
1002          * - previous RPC, if any, has done and
1003          * - no new RPC is initiated.
1004          * - my batch is still active; no one can run it again now.
1005          * Cancel pending schedules and prevent future schedule attempts:
1006          */
1007         swi_exit_workitem(wi);
1008         sfw_test_unit_done(tsu);
1009         return 1;
1010 }
1011
1012 static int
1013 sfw_run_batch (sfw_batch_t *tsb)
1014 {
1015         swi_workitem_t      *wi;
1016         sfw_test_unit_t     *tsu;
1017         sfw_test_instance_t *tsi;
1018
1019         if (sfw_batch_active(tsb)) {
1020                 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1021                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1022                 return 0;
1023         }
1024
1025         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1026                 if (!tsi->tsi_is_client)        /* skip server instances */
1027                         continue;
1028
1029                 LASSERT(!tsi->tsi_stopping);
1030                 LASSERT(!sfw_test_active(tsi));
1031
1032                 atomic_inc(&tsb->bat_nactive);
1033
1034                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1035                         atomic_inc(&tsi->tsi_nactive);
1036                         tsu->tsu_loop = tsi->tsi_loop;
1037                         wi = &tsu->tsu_worker;
1038                         swi_init_workitem(wi, tsu, sfw_run_test,
1039                                           lst_sched_test[\
1040                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1041                         swi_schedule_workitem(wi);
1042                 }
1043         }
1044
1045         return 0;
1046 }
1047
1048 int
1049 sfw_stop_batch (sfw_batch_t *tsb, int force)
1050 {
1051         sfw_test_instance_t *tsi;
1052         srpc_client_rpc_t   *rpc;
1053
1054         if (!sfw_batch_active(tsb)) {
1055                 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1056                 return 0;
1057         }
1058
1059         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1060                 spin_lock(&tsi->tsi_lock);
1061
1062                 if (!tsi->tsi_is_client ||
1063                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1064                         spin_unlock(&tsi->tsi_lock);
1065                         continue;
1066                 }
1067
1068                 tsi->tsi_stopping = 1;
1069
1070                 if (!force) {
1071                         spin_unlock(&tsi->tsi_lock);
1072                         continue;
1073                 }
1074
1075                 /* abort launched rpcs in the test */
1076                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1077                         spin_lock(&rpc->crpc_lock);
1078
1079                         srpc_abort_rpc(rpc, -EINTR);
1080
1081                         spin_unlock(&rpc->crpc_lock);
1082                 }
1083
1084                 spin_unlock(&tsi->tsi_lock);
1085         }
1086
1087         return 0;
1088 }
1089
1090 static int
1091 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1092 {
1093         sfw_test_instance_t *tsi;
1094
1095         if (testidx < 0)
1096                 return -EINVAL;
1097
1098         if (testidx == 0) {
1099                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1100                 return 0;
1101         }
1102
1103         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1104                 if (testidx-- > 1)
1105                         continue;
1106
1107                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1108                 return 0;
1109         }
1110
1111         return -ENOENT;
1112 }
1113
1114 void
1115 sfw_free_pages (srpc_server_rpc_t *rpc)
1116 {
1117         srpc_free_bulk(rpc->srpc_bulk);
1118         rpc->srpc_bulk = NULL;
1119 }
1120
1121 int
1122 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1123                 int sink)
1124 {
1125         LASSERT(rpc->srpc_bulk == NULL);
1126         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1127
1128         rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1129         if (rpc->srpc_bulk == NULL)
1130                 return -ENOMEM;
1131
1132         return 0;
1133 }
1134
1135 static int
1136 sfw_add_test (srpc_server_rpc_t *rpc)
1137 {
1138         sfw_session_t     *sn = sfw_data.fw_session;
1139         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1140         srpc_test_reqst_t *request;
1141         int                rc;
1142         sfw_batch_t       *bat;
1143
1144         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1145         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1146
1147         if (request->tsr_loop == 0 ||
1148             request->tsr_concur == 0 ||
1149             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1150             request->tsr_ndest > SFW_MAX_NDESTS ||
1151             (request->tsr_is_client && request->tsr_ndest == 0) ||
1152             request->tsr_concur > SFW_MAX_CONCUR ||
1153             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1154             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1155                 reply->tsr_status = EINVAL;
1156                 return 0;
1157         }
1158
1159         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1160             sfw_find_test_case(request->tsr_service) == NULL) {
1161                 reply->tsr_status = ENOENT;
1162                 return 0;
1163         }
1164
1165         bat = sfw_bid2batch(request->tsr_bid);
1166         if (bat == NULL) {
1167                 CERROR("dropping RPC %s from %s under memory pressure\n",
1168                         rpc->srpc_scd->scd_svc->sv_name,
1169                         libcfs_id2str(rpc->srpc_peer));
1170                 return -ENOMEM;
1171         }
1172
1173         if (sfw_batch_active(bat)) {
1174                 reply->tsr_status = EBUSY;
1175                 return 0;
1176         }
1177
1178         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1179                 /* rpc will be resumed later in sfw_bulk_ready */
1180                 int     npg = sfw_id_pages(request->tsr_ndest);
1181                 int     len;
1182
1183                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1184                         len = npg * PAGE_SIZE;
1185
1186                 } else  {
1187                         len = sizeof(lnet_process_id_packed_t) *
1188                               request->tsr_ndest;
1189                 }
1190
1191                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1192         }
1193
1194         rc = sfw_add_test_instance(bat, rpc);
1195         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1196                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1197                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1198                 request->tsr_is_client ? "client" : "server",
1199                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1200
1201         reply->tsr_status = (rc < 0) ? -rc : rc;
1202         return 0;
1203 }
1204
1205 static int
1206 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1207 {
1208         sfw_session_t *sn = sfw_data.fw_session;
1209         int            rc = 0;
1210         sfw_batch_t   *bat;
1211
1212         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1213
1214         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1215                 reply->bar_status = ESRCH;
1216                 return 0;
1217         }
1218
1219         bat = sfw_find_batch(request->bar_bid);
1220         if (bat == NULL) {
1221                 reply->bar_status = ENOENT;
1222                 return 0;
1223         }
1224
1225         switch (request->bar_opc) {
1226         case SRPC_BATCH_OPC_RUN:
1227                 rc = sfw_run_batch(bat);
1228                 break;
1229
1230         case SRPC_BATCH_OPC_STOP:
1231                 rc = sfw_stop_batch(bat, request->bar_arg);
1232                 break;
1233
1234         case SRPC_BATCH_OPC_QUERY:
1235                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1236                 break;
1237
1238         default:
1239                 return -EINVAL; /* drop it */
1240         }
1241
1242         reply->bar_status = (rc < 0) ? -rc : rc;
1243         return 0;
1244 }
1245
1246 static int
1247 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1248 {
1249         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1250         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1251         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1252         unsigned        features = LST_FEATS_MASK;
1253         int             rc = 0;
1254
1255         LASSERT(sfw_data.fw_active_srpc == NULL);
1256         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1257
1258         spin_lock(&sfw_data.fw_lock);
1259
1260         if (sfw_data.fw_shuttingdown) {
1261                 spin_unlock(&sfw_data.fw_lock);
1262                 return -ESHUTDOWN;
1263         }
1264
1265         /* Remove timer to avoid racing with it or expiring active session */
1266         if (sfw_del_session_timer() != 0) {
1267                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1268                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1269                 spin_unlock(&sfw_data.fw_lock);
1270                 return -EAGAIN;
1271         }
1272
1273         sfw_data.fw_active_srpc = rpc;
1274         spin_unlock(&sfw_data.fw_lock);
1275
1276         sfw_unpack_message(request);
1277         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1278
1279         /* rpc module should have checked this */
1280         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1281
1282         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1283             sv->sv_id != SRPC_SERVICE_DEBUG) {
1284                 sfw_session_t *sn = sfw_data.fw_session;
1285
1286                 if (sn != NULL &&
1287                     sn->sn_features != request->msg_ses_feats) {
1288                         CNETERR("Features of framework RPC don't match "
1289                                 "features of current session: %x/%x\n",
1290                                 request->msg_ses_feats, sn->sn_features);
1291                         reply->msg_body.reply.status = EPROTO;
1292                         reply->msg_body.reply.sid    = sn->sn_id;
1293                         goto out;
1294                 }
1295
1296         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1297                 /* NB: at this point, old version will ignore features and
1298                  * create new session anyway, so console should be able
1299                  * to handle this */
1300                 reply->msg_body.reply.status = EPROTO;
1301                 goto out;
1302         }
1303
1304         switch(sv->sv_id) {
1305         default:
1306                 LBUG ();
1307         case SRPC_SERVICE_TEST:
1308                 rc = sfw_add_test(rpc);
1309                 break;
1310
1311         case SRPC_SERVICE_BATCH:
1312                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1313                                        &reply->msg_body.bat_reply);
1314                 break;
1315
1316         case SRPC_SERVICE_QUERY_STAT:
1317                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1318                                    &reply->msg_body.stat_reply);
1319                 break;
1320
1321         case SRPC_SERVICE_DEBUG:
1322                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1323                                        &reply->msg_body.dbg_reply);
1324                 break;
1325
1326         case SRPC_SERVICE_MAKE_SESSION:
1327                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1328                                       &reply->msg_body.mksn_reply);
1329                 break;
1330
1331         case SRPC_SERVICE_REMOVE_SESSION:
1332                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1333                                         &reply->msg_body.rmsn_reply);
1334                 break;
1335         }
1336
1337         if (sfw_data.fw_session != NULL)
1338                 features = sfw_data.fw_session->sn_features;
1339  out:
1340         reply->msg_ses_feats = features;
1341         rpc->srpc_done = sfw_server_rpc_done;
1342         spin_lock(&sfw_data.fw_lock);
1343
1344         if (!sfw_data.fw_shuttingdown)
1345                 sfw_add_session_timer();
1346
1347         sfw_data.fw_active_srpc = NULL;
1348         spin_unlock(&sfw_data.fw_lock);
1349         return rc;
1350 }
1351
1352 static int
1353 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1354 {
1355         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1356         int                     rc;
1357
1358         LASSERT(rpc->srpc_bulk != NULL);
1359         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1360         LASSERT(sfw_data.fw_active_srpc == NULL);
1361         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1362
1363         spin_lock(&sfw_data.fw_lock);
1364
1365         if (status != 0) {
1366                 CERROR("Bulk transfer failed for RPC: "
1367                        "service %s, peer %s, status %d\n",
1368                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1369                 spin_unlock(&sfw_data.fw_lock);
1370                 return -EIO;
1371         }
1372
1373         if (sfw_data.fw_shuttingdown) {
1374                 spin_unlock(&sfw_data.fw_lock);
1375                 return -ESHUTDOWN;
1376         }
1377
1378         if (sfw_del_session_timer() != 0) {
1379                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1380                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1381                 spin_unlock(&sfw_data.fw_lock);
1382                 return -EAGAIN;
1383         }
1384
1385         sfw_data.fw_active_srpc = rpc;
1386         spin_unlock(&sfw_data.fw_lock);
1387
1388         rc = sfw_add_test(rpc);
1389
1390         spin_lock(&sfw_data.fw_lock);
1391
1392         if (!sfw_data.fw_shuttingdown)
1393                 sfw_add_session_timer();
1394
1395         sfw_data.fw_active_srpc = NULL;
1396         spin_unlock(&sfw_data.fw_lock);
1397         return rc;
1398 }
1399
1400 srpc_client_rpc_t *
1401 sfw_create_rpc(lnet_process_id_t peer, int service,
1402                unsigned features, int nbulkiov, int bulklen,
1403                void (*done)(srpc_client_rpc_t *), void *priv)
1404 {
1405         srpc_client_rpc_t *rpc = NULL;
1406
1407         spin_lock(&sfw_data.fw_lock);
1408
1409         LASSERT (!sfw_data.fw_shuttingdown);
1410         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1411
1412         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1413                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1414                                      srpc_client_rpc_t, crpc_list);
1415                 list_del(&rpc->crpc_list);
1416
1417                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1418                                      done, sfw_client_rpc_fini, priv);
1419         }
1420
1421         spin_unlock(&sfw_data.fw_lock);
1422
1423         if (rpc == NULL) {
1424                 rpc = srpc_create_client_rpc(peer, service,
1425                                              nbulkiov, bulklen, done,
1426                                              nbulkiov != 0 ?  NULL :
1427                                              sfw_client_rpc_fini,
1428                                              priv);
1429         }
1430
1431         if (rpc != NULL) /* "session" is concept in framework */
1432                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1433
1434         return rpc;
1435 }
1436
1437 void
1438 sfw_unpack_message (srpc_msg_t *msg)
1439 {
1440         if (msg->msg_magic == SRPC_MSG_MAGIC)
1441                 return; /* no flipping needed */
1442
1443         /* srpc module should guarantee I wouldn't get crap */
1444         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1445
1446         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1447                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1448
1449                 __swab32s(&req->str_type);
1450                 __swab64s(&req->str_rpyid);
1451                 sfw_unpack_sid(req->str_sid);
1452                 return;
1453         }
1454
1455         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1456                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1457
1458                 __swab32s(&rep->str_status);
1459                 sfw_unpack_sid(rep->str_sid);
1460                 sfw_unpack_fw_counters(rep->str_fw);
1461                 sfw_unpack_rpc_counters(rep->str_rpc);
1462                 sfw_unpack_lnet_counters(rep->str_lnet);
1463                 return;
1464         }
1465
1466         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1467                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1468
1469                 __swab64s(&req->mksn_rpyid);
1470                 __swab32s(&req->mksn_force);
1471                 sfw_unpack_sid(req->mksn_sid);
1472                 return;
1473         }
1474
1475         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1476                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1477
1478                 __swab32s(&rep->mksn_status);
1479                 __swab32s(&rep->mksn_timeout);
1480                 sfw_unpack_sid(rep->mksn_sid);
1481                 return;
1482         }
1483
1484         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1485                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1486
1487                 __swab64s(&req->rmsn_rpyid);
1488                 sfw_unpack_sid(req->rmsn_sid);
1489                 return;
1490         }
1491
1492         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1493                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1494
1495                 __swab32s(&rep->rmsn_status);
1496                 sfw_unpack_sid(rep->rmsn_sid);
1497                 return;
1498         }
1499
1500         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1501                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1502
1503                 __swab64s(&req->dbg_rpyid);
1504                 __swab32s(&req->dbg_flags);
1505                 sfw_unpack_sid(req->dbg_sid);
1506                 return;
1507         }
1508
1509         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1510                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1511
1512                 __swab32s(&rep->dbg_nbatch);
1513                 __swab32s(&rep->dbg_timeout);
1514                 sfw_unpack_sid(rep->dbg_sid);
1515                 return;
1516         }
1517
1518         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1519                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1520
1521                 __swab32s(&req->bar_opc);
1522                 __swab64s(&req->bar_rpyid);
1523                 __swab32s(&req->bar_testidx);
1524                 __swab32s(&req->bar_arg);
1525                 sfw_unpack_sid(req->bar_sid);
1526                 __swab64s(&req->bar_bid.bat_id);
1527                 return;
1528         }
1529
1530         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1531                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1532
1533                 __swab32s(&rep->bar_status);
1534                 sfw_unpack_sid(rep->bar_sid);
1535                 return;
1536         }
1537
1538         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1539                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1540
1541                 __swab64s(&req->tsr_rpyid);
1542                 __swab64s(&req->tsr_bulkid);
1543                 __swab32s(&req->tsr_loop);
1544                 __swab32s(&req->tsr_ndest);
1545                 __swab32s(&req->tsr_concur);
1546                 __swab32s(&req->tsr_service);
1547                 sfw_unpack_sid(req->tsr_sid);
1548                 __swab64s(&req->tsr_bid.bat_id);
1549                 return;
1550         }
1551
1552         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1553                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1554
1555                 __swab32s(&rep->tsr_status);
1556                 sfw_unpack_sid(rep->tsr_sid);
1557                 return;
1558         }
1559
1560         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1561                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1562
1563                 __swab64s(&req->join_rpyid);
1564                 sfw_unpack_sid(req->join_sid);
1565                 return;
1566         }
1567
1568         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1569                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1570
1571                 __swab32s(&rep->join_status);
1572                 __swab32s(&rep->join_timeout);
1573                 sfw_unpack_sid(rep->join_sid);
1574                 return;
1575         }
1576
1577         LBUG ();
1578         return;
1579 }
1580
1581 void
1582 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1583 {
1584         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1585         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1586
1587         spin_lock(&rpc->crpc_lock);
1588         srpc_abort_rpc(rpc, -EINTR);
1589         spin_unlock(&rpc->crpc_lock);
1590         return;
1591 }
1592
1593 void
1594 sfw_post_rpc (srpc_client_rpc_t *rpc)
1595 {
1596         spin_lock(&rpc->crpc_lock);
1597
1598         LASSERT(!rpc->crpc_closed);
1599         LASSERT(!rpc->crpc_aborted);
1600         LASSERT(list_empty(&rpc->crpc_list));
1601         LASSERT(!sfw_data.fw_shuttingdown);
1602
1603         rpc->crpc_timeout = rpc_timeout;
1604         srpc_post_rpc(rpc);
1605
1606         spin_unlock(&rpc->crpc_lock);
1607         return;
1608 }
1609
1610 static srpc_service_t sfw_services[] =
1611 {
1612         {
1613                 /* sv_id */    SRPC_SERVICE_DEBUG,
1614                 /* sv_name */  "debug",
1615                 0
1616         },
1617         {
1618                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1619                 /* sv_name */  "query stats",
1620                 0
1621         },
1622         {
1623                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1624                 /* sv_name */  "make session",
1625                 0
1626         },
1627         {
1628                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1629                 /* sv_name */  "remove session",
1630                 0
1631         },
1632         {
1633                 /* sv_id */    SRPC_SERVICE_BATCH,
1634                 /* sv_name */  "batch service",
1635                 0
1636         },
1637         {
1638                 /* sv_id */    SRPC_SERVICE_TEST,
1639                 /* sv_name */  "test service",
1640                 0
1641         },
1642         {
1643                 /* sv_id */    0,
1644                 /* sv_name */  NULL,
1645                 0
1646         }
1647 };
1648
1649 int
1650 sfw_startup (void)
1651 {
1652         int              i;
1653         int              rc;
1654         int              error;
1655         srpc_service_t  *sv;
1656         sfw_test_case_t *tsc;
1657
1658
1659         if (session_timeout < 0) {
1660                 CERROR ("Session timeout must be non-negative: %d\n",
1661                         session_timeout);
1662                 return -EINVAL;
1663         }
1664
1665         if (rpc_timeout < 0) {
1666                 CERROR ("RPC timeout must be non-negative: %d\n",
1667                         rpc_timeout);
1668                 return -EINVAL;
1669         }
1670
1671         if (session_timeout == 0)
1672                 CWARN ("Zero session_timeout specified "
1673                        "- test sessions never expire.\n");
1674
1675         if (rpc_timeout == 0)
1676                 CWARN ("Zero rpc_timeout specified "
1677                        "- test RPC never expire.\n");
1678
1679         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1680
1681         sfw_data.fw_session     = NULL;
1682         sfw_data.fw_active_srpc = NULL;
1683         spin_lock_init(&sfw_data.fw_lock);
1684         atomic_set(&sfw_data.fw_nzombies, 0);
1685         INIT_LIST_HEAD(&sfw_data.fw_tests);
1686         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1687         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1688
1689         brw_init_test_client();
1690         brw_init_test_service();
1691         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1692         LASSERT (rc == 0);
1693
1694         ping_init_test_client();
1695         ping_init_test_service();
1696         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1697         LASSERT (rc == 0);
1698
1699         error = 0;
1700         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1701                 sv = tsc->tsc_srv_service;
1702
1703                 rc = srpc_add_service(sv);
1704                 LASSERT(rc != -EBUSY);
1705                 if (rc != 0) {
1706                         CWARN("Failed to add %s service: %d\n",
1707                               sv->sv_name, rc);
1708                         error = rc;
1709                 }
1710         }
1711
1712         for (i = 0; ; i++) {
1713                 sv = &sfw_services[i];
1714                 if (sv->sv_name == NULL) break;
1715
1716                 sv->sv_bulk_ready = NULL;
1717                 sv->sv_handler    = sfw_handle_server_rpc;
1718                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1719                 if (sv->sv_id == SRPC_SERVICE_TEST)
1720                         sv->sv_bulk_ready = sfw_bulk_ready;
1721
1722                 rc = srpc_add_service(sv);
1723                 LASSERT (rc != -EBUSY);
1724                 if (rc != 0) {
1725                         CWARN ("Failed to add %s service: %d\n",
1726                                sv->sv_name, rc);
1727                         error = rc;
1728                 }
1729
1730                 /* about to sfw_shutdown, no need to add buffer */
1731                 if (error) continue;
1732
1733                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1734                 if (rc != 0) {
1735                         CWARN("Failed to reserve enough buffers: "
1736                               "service %s, %d needed: %d\n",
1737                               sv->sv_name, sv->sv_wi_total, rc);
1738                         error = -ENOMEM;
1739                 }
1740         }
1741
1742         if (error != 0)
1743                 sfw_shutdown();
1744         return error;
1745 }
1746
1747 void
1748 sfw_shutdown (void)
1749 {
1750         srpc_service_t  *sv;
1751         sfw_test_case_t *tsc;
1752         int              i;
1753
1754         spin_lock(&sfw_data.fw_lock);
1755
1756         sfw_data.fw_shuttingdown = 1;
1757         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1758                        "waiting for active RPC to finish.\n");
1759
1760         if (sfw_del_session_timer() != 0)
1761                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1762                                "waiting for session timer to explode.\n");
1763
1764         sfw_deactivate_session();
1765         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1766                        sfw_data.fw_lock,
1767                        "waiting for %d zombie sessions to die.\n",
1768                        atomic_read(&sfw_data.fw_nzombies));
1769
1770         spin_unlock(&sfw_data.fw_lock);
1771
1772         for (i = 0; ; i++) {
1773                 sv = &sfw_services[i];
1774                 if (sv->sv_name == NULL)
1775                         break;
1776
1777                 srpc_shutdown_service(sv);
1778                 srpc_remove_service(sv);
1779         }
1780
1781         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1782                 sv = tsc->tsc_srv_service;
1783                 srpc_shutdown_service(sv);
1784                 srpc_remove_service(sv);
1785         }
1786
1787         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1788                 srpc_client_rpc_t *rpc;
1789
1790                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1791                                  srpc_client_rpc_t, crpc_list);
1792                 list_del(&rpc->crpc_list);
1793
1794                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1795         }
1796
1797         for (i = 0; ; i++) {
1798                 sv = &sfw_services[i];
1799                 if (sv->sv_name == NULL)
1800                         break;
1801
1802                 srpc_wait_service_shutdown(sv);
1803         }
1804
1805         while (!list_empty(&sfw_data.fw_tests)) {
1806                 tsc = list_entry(sfw_data.fw_tests.next,
1807                                  sfw_test_case_t, tsc_list);
1808
1809                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1810
1811                 list_del(&tsc->tsc_list);
1812                 LIBCFS_FREE(tsc, sizeof(*tsc));
1813         }
1814
1815         return;
1816 }