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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lnet/selftest/framework.c
  33  *
  34  * Author: Isaac Huang <isaac@clusterfs.com>
  35  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LNET
  39
  40 #include "selftest.h"
  41
  42 struct lst_sid LST_INVALID_SID = { .ses_nid = LNET_NID_ANY, .ses_stamp = -1};
  43
  44 static int session_timeout = 100;
  45 module_param(session_timeout, int, 0444);
  46 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
  47
  48 static int rpc_timeout = 64;
  49 module_param(rpc_timeout, int, 0644);
  50 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
  51
  52 #define sfw_unpack_id(id)               \
  53 do {                                    \
  54         __swab64s(&(id).nid);           \
  55         __swab32s(&(id).pid);           \
  56 } while (0)
  57
  58 #define sfw_unpack_sid(sid)             \
  59 do {                                    \
  60         __swab64s(&(sid).ses_nid);      \
  61         __swab64s(&(sid).ses_stamp);    \
  62 } while (0)
  63
  64 #define sfw_unpack_fw_counters(fc)        \
  65 do {                                      \
  66         __swab32s(&(fc).running_ms);      \
  67         __swab32s(&(fc).active_batches);  \
  68         __swab32s(&(fc).zombie_sessions); \
  69         __swab32s(&(fc).brw_errors);      \
  70         __swab32s(&(fc).ping_errors);     \
  71 } while (0)
  72
  73 #define sfw_unpack_rpc_counters(rc)     \
  74 do {                                    \
  75         __swab32s(&(rc).errors);        \
  76         __swab32s(&(rc).rpcs_sent);     \
  77         __swab32s(&(rc).rpcs_rcvd);     \
  78         __swab32s(&(rc).rpcs_dropped);  \
  79         __swab32s(&(rc).rpcs_expired);  \
  80         __swab64s(&(rc).bulk_get);      \
  81         __swab64s(&(rc).bulk_put);      \
  82 } while (0)
  83
  84 #define sfw_unpack_lnet_counters(lc)    \
  85 do {                                    \
  86         __swab32s(&(lc).errors);        \
  87         __swab32s(&(lc).msgs_max);      \
  88         __swab32s(&(lc).msgs_alloc);    \
  89         __swab32s(&(lc).send_count);    \
  90         __swab32s(&(lc).recv_count);    \
  91         __swab32s(&(lc).drop_count);    \
  92         __swab32s(&(lc).route_count);   \
  93         __swab64s(&(lc).send_length);   \
  94         __swab64s(&(lc).recv_length);   \
  95         __swab64s(&(lc).drop_length);   \
  96         __swab64s(&(lc).route_length);  \
  97 } while (0)
  98
  99 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 100 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 101
 102 static struct smoketest_framework {
 103         /* RPCs to be recycled */
 104         struct list_head        fw_zombie_rpcs;
 105         /* stopping sessions */
 106         struct list_head        fw_zombie_sessions;
 107         /* registered test cases */
 108         struct list_head        fw_tests;
 109         /* # zombie sessions */
 110         atomic_t                fw_nzombies;
 111         /* serialise */
 112         spinlock_t              fw_lock;
 113         /* _the_ session */
 114         sfw_session_t           *fw_session;
 115         /* shutdown in progress */
 116         int                     fw_shuttingdown;
 117         /* running RPC */
 118         srpc_server_rpc_t       *fw_active_srpc;
 119 } sfw_data;
 120
 121 /* forward ref's */
 122 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 123 void sfw_destroy_session (sfw_session_t *sn);
 124
 125 static inline sfw_test_case_t *
 126 sfw_find_test_case(int id)
 127 {
 128         sfw_test_case_t *tsc;
 129
 130         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 131         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 132
 133         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 134                 if (tsc->tsc_srv_service->sv_id == id)
 135                         return tsc;
 136         }
 137
 138         return NULL;
 139 }
 140
 141 static int
 142 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 143 {
 144         sfw_test_case_t *tsc;
 145
 146         if (sfw_find_test_case(service->sv_id) != NULL) {
 147                 CERROR ("Failed to register test %s (%d)\n",
 148                         service->sv_name, service->sv_id);
 149                 return -EEXIST;
 150         }
 151
 152         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 153         if (tsc == NULL)
 154                 return -ENOMEM;
 155
 156         tsc->tsc_cli_ops     = cliops;
 157         tsc->tsc_srv_service = service;
 158
 159         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 160         return 0;
 161 }
 162
 163 static void
 164 sfw_add_session_timer (void)
 165 {
 166         sfw_session_t *sn = sfw_data.fw_session;
 167         stt_timer_t   *timer = &sn->sn_timer;
 168
 169         LASSERT (!sfw_data.fw_shuttingdown);
 170
 171         if (sn == NULL || sn->sn_timeout == 0)
 172                 return;
 173
 174         LASSERT (!sn->sn_timer_active);
 175
 176         sn->sn_timer_active = 1;
 177         timer->stt_expires = ktime_get_real_seconds()+ sn->sn_timeout;
 178         stt_add_timer(timer);
 179         return;
 180 }
 181
 182 static int
 183 sfw_del_session_timer (void)
 184 {
 185         sfw_session_t *sn = sfw_data.fw_session;
 186
 187         if (sn == NULL || !sn->sn_timer_active)
 188                 return 0;
 189
 190         LASSERT (sn->sn_timeout != 0);
 191
 192         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 193                 sn->sn_timer_active = 0;
 194                 return 0;
 195         }
 196
 197         return EBUSY; /* racing with sfw_session_expired() */
 198 }
 199
 200 /* called with sfw_data.fw_lock held */
 201 static void
 202 sfw_deactivate_session (void)
 203 __must_hold(&sfw_data.fw_lock)
 204 {
 205         sfw_session_t *sn = sfw_data.fw_session;
 206         int            nactive = 0;
 207         sfw_batch_t   *tsb;
 208         sfw_test_case_t *tsc;
 209
 210         if (sn == NULL) return;
 211
 212         LASSERT(!sn->sn_timer_active);
 213
 214         sfw_data.fw_session = NULL;
 215         atomic_inc(&sfw_data.fw_nzombies);
 216         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 217
 218         spin_unlock(&sfw_data.fw_lock);
 219
 220         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 221                 srpc_abort_service(tsc->tsc_srv_service);
 222         }
 223
 224         spin_lock(&sfw_data.fw_lock);
 225
 226         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 227                 if (sfw_batch_active(tsb)) {
 228                         nactive++;
 229                         sfw_stop_batch(tsb, 1);
 230                 }
 231         }
 232
 233         if (nactive != 0)
 234                 return; /* wait for active batches to stop */
 235
 236         list_del_init(&sn->sn_list);
 237         spin_unlock(&sfw_data.fw_lock);
 238
 239         sfw_destroy_session(sn);
 240
 241         spin_lock(&sfw_data.fw_lock);
 242 }
 243
 244
 245 static void
 246 sfw_session_expired (void *data)
 247 {
 248         sfw_session_t *sn = data;
 249
 250         spin_lock(&sfw_data.fw_lock);
 251
 252         LASSERT (sn->sn_timer_active);
 253         LASSERT (sn == sfw_data.fw_session);
 254
 255         CWARN ("Session expired! sid: %s-%llu, name: %s\n",
 256                libcfs_nid2str(sn->sn_id.ses_nid),
 257                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 258
 259         sn->sn_timer_active = 0;
 260         sfw_deactivate_session();
 261
 262         spin_unlock(&sfw_data.fw_lock);
 263 }
 264
 265 static inline void
 266 sfw_init_session(sfw_session_t *sn, struct lst_sid sid,
 267                  unsigned features, const char *name)
 268 {
 269         stt_timer_t *timer = &sn->sn_timer;
 270
 271         memset(sn, 0, sizeof(sfw_session_t));
 272         INIT_LIST_HEAD(&sn->sn_list);
 273         INIT_LIST_HEAD(&sn->sn_batches);
 274         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 275         atomic_set(&sn->sn_brw_errors, 0);
 276         atomic_set(&sn->sn_ping_errors, 0);
 277         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 278
 279         sn->sn_timer_active = 0;
 280         sn->sn_id           = sid;
 281         sn->sn_features     = features;
 282         sn->sn_timeout      = session_timeout;
 283         sn->sn_started      = cfs_time_current();
 284
 285         timer->stt_data = sn;
 286         timer->stt_func = sfw_session_expired;
 287         INIT_LIST_HEAD(&timer->stt_list);
 288 }
 289
 290 /* completion handler for incoming framework RPCs */
 291 static void
 292 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 293 {
 294         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 295         int                     status  = rpc->srpc_status;
 296
 297         CDEBUG (D_NET,
 298                 "Incoming framework RPC done: "
 299                 "service %s, peer %s, status %s:%d\n",
 300                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 301                 swi_state2str(rpc->srpc_wi.swi_state),
 302                 status);
 303
 304         if (rpc->srpc_bulk != NULL)
 305                 sfw_free_pages(rpc);
 306         return;
 307 }
 308
 309 static void
 310 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 311 {
 312         LASSERT(rpc->crpc_bulk.bk_niov == 0);
 313         LASSERT(list_empty(&rpc->crpc_list));
 314         LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
 315
 316         CDEBUG(D_NET, "Outgoing framework RPC done: "
 317                "service %d, peer %s, status %s:%d:%d\n",
 318                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 319                swi_state2str(rpc->crpc_wi.swi_state),
 320                rpc->crpc_aborted, rpc->crpc_status);
 321
 322         spin_lock(&sfw_data.fw_lock);
 323
 324         /* my callers must finish all RPCs before shutting me down */
 325         LASSERT(!sfw_data.fw_shuttingdown);
 326         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 327
 328         spin_unlock(&sfw_data.fw_lock);
 329 }
 330
 331 static sfw_batch_t *
 332 sfw_find_batch(struct lst_bid bid)
 333 {
 334         sfw_session_t *sn = sfw_data.fw_session;
 335         sfw_batch_t   *bat;
 336
 337         LASSERT(sn != NULL);
 338
 339         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 340                 if (bat->bat_id.bat_id == bid.bat_id)
 341                         return bat;
 342         }
 343
 344         return NULL;
 345 }
 346
 347 static sfw_batch_t *
 348 sfw_bid2batch(struct lst_bid bid)
 349 {
 350         sfw_session_t *sn = sfw_data.fw_session;
 351         sfw_batch_t   *bat;
 352
 353         LASSERT (sn != NULL);
 354
 355         bat = sfw_find_batch(bid);
 356         if (bat != NULL)
 357                 return bat;
 358
 359         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 360         if (bat == NULL)
 361                 return NULL;
 362
 363         bat->bat_error   = 0;
 364         bat->bat_session = sn;
 365         bat->bat_id      = bid;
 366         atomic_set(&bat->bat_nactive, 0);
 367         INIT_LIST_HEAD(&bat->bat_tests);
 368
 369         list_add_tail(&bat->bat_list, &sn->sn_batches);
 370         return bat;
 371 }
 372
 373 static int
 374 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 375 {
 376         sfw_session_t  *sn = sfw_data.fw_session;
 377         struct sfw_counters *cnt = &reply->str_fw;
 378         sfw_batch_t    *bat;
 379
 380         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 381
 382         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 383                 reply->str_status = EINVAL;
 384                 return 0;
 385         }
 386
 387         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 388                 reply->str_status = ESRCH;
 389                 return 0;
 390         }
 391
 392         lnet_counters_get(&reply->str_lnet);
 393         srpc_get_counters(&reply->str_rpc);
 394
 395         /* send over the msecs since the session was started
 396          - with 32 bits to send, this is ~49 days */
 397         cnt->running_ms      = jiffies_to_msecs(jiffies - sn->sn_started);
 398         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 399         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 400         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 401
 402         cnt->active_batches = 0;
 403         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 404                 if (atomic_read(&bat->bat_nactive) > 0)
 405                         cnt->active_batches++;
 406         }
 407
 408         reply->str_status = 0;
 409         return 0;
 410 }
 411
 412 int
 413 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 414 {
 415         sfw_session_t *sn = sfw_data.fw_session;
 416         srpc_msg_t    *msg = container_of(request, srpc_msg_t,
 417                                           msg_body.mksn_reqst);
 418         int            cplen = 0;
 419
 420         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 421                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 422                 reply->mksn_status = EINVAL;
 423                 return 0;
 424         }
 425
 426         if (sn != NULL) {
 427                 reply->mksn_status  = 0;
 428                 reply->mksn_sid     = sn->sn_id;
 429                 reply->mksn_timeout = sn->sn_timeout;
 430
 431                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 432                         atomic_inc(&sn->sn_refcount);
 433                         return 0;
 434                 }
 435
 436                 if (!request->mksn_force) {
 437                         reply->mksn_status = EBUSY;
 438                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 439                                         sizeof(reply->mksn_name));
 440                         if (cplen >= sizeof(reply->mksn_name))
 441                                 return -E2BIG;
 442                         return 0;
 443                 }
 444         }
 445
 446         /* reject the request if it requires unknown features
 447          * NB: old version will always accept all features because it's not
 448          * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
 449          * harmless because it will return zero feature to console, and it's
 450          * console's responsibility to make sure all nodes in a session have
 451          * same feature mask. */
 452         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 453                 reply->mksn_status = EPROTO;
 454                 return 0;
 455         }
 456
 457         /* brand new or create by force */
 458         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 459         if (sn == NULL) {
 460                 CERROR("dropping RPC mksn under memory pressure\n");
 461                 return -ENOMEM;
 462         }
 463
 464         sfw_init_session(sn, request->mksn_sid,
 465                          msg->msg_ses_feats, &request->mksn_name[0]);
 466
 467         spin_lock(&sfw_data.fw_lock);
 468
 469         sfw_deactivate_session();
 470         LASSERT(sfw_data.fw_session == NULL);
 471         sfw_data.fw_session = sn;
 472
 473         spin_unlock(&sfw_data.fw_lock);
 474
 475         reply->mksn_status  = 0;
 476         reply->mksn_sid     = sn->sn_id;
 477         reply->mksn_timeout = sn->sn_timeout;
 478         return 0;
 479 }
 480
 481 static int
 482 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 483 {
 484         sfw_session_t *sn = sfw_data.fw_session;
 485
 486         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 487
 488         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 489                 reply->rmsn_status = EINVAL;
 490                 return 0;
 491         }
 492
 493         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 494                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 495                 return 0;
 496         }
 497
 498         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 499                 reply->rmsn_status = 0;
 500                 return 0;
 501         }
 502
 503         spin_lock(&sfw_data.fw_lock);
 504         sfw_deactivate_session();
 505         spin_unlock(&sfw_data.fw_lock);
 506
 507         reply->rmsn_status = 0;
 508         reply->rmsn_sid    = LST_INVALID_SID;
 509         LASSERT(sfw_data.fw_session == NULL);
 510         return 0;
 511 }
 512
 513 static int
 514 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 515 {
 516         sfw_session_t *sn = sfw_data.fw_session;
 517
 518         if (sn == NULL) {
 519                 reply->dbg_status = ESRCH;
 520                 reply->dbg_sid    = LST_INVALID_SID;
 521                 return 0;
 522         }
 523
 524         reply->dbg_status  = 0;
 525         reply->dbg_sid     = sn->sn_id;
 526         reply->dbg_timeout = sn->sn_timeout;
 527         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 528             >= sizeof(reply->dbg_name))
 529                 return -E2BIG;
 530
 531         return 0;
 532 }
 533
 534 static void
 535 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 536 {
 537         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 538         sfw_test_instance_t *tsi = tsu->tsu_instance;
 539
 540         /* Called with hold of tsi->tsi_lock */
 541         LASSERT(list_empty(&rpc->crpc_list));
 542         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 543 }
 544
 545 static inline int
 546 sfw_test_buffers(sfw_test_instance_t *tsi)
 547 {
 548         struct sfw_test_case    *tsc;
 549         struct srpc_service     *svc;
 550         int                     nbuf;
 551
 552         LASSERT(tsi != NULL);
 553         tsc = sfw_find_test_case(tsi->tsi_service);
 554         LASSERT(tsc != NULL);
 555         svc = tsc->tsc_srv_service;
 556         LASSERT(svc != NULL);
 557
 558         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 559         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 560 }
 561
 562 static int
 563 sfw_load_test(struct sfw_test_instance *tsi)
 564 {
 565         struct sfw_test_case    *tsc;
 566         struct srpc_service     *svc;
 567         int                     nbuf;
 568         int                     rc;
 569
 570         LASSERT(tsi != NULL);
 571         tsc = sfw_find_test_case(tsi->tsi_service);
 572         nbuf = sfw_test_buffers(tsi);
 573         LASSERT(tsc != NULL);
 574         svc = tsc->tsc_srv_service;
 575
 576         if (tsi->tsi_is_client) {
 577                 tsi->tsi_ops = tsc->tsc_cli_ops;
 578                 return 0;
 579         }
 580
 581         rc = srpc_service_add_buffers(svc, nbuf);
 582         if (rc != 0) {
 583                 CWARN("Failed to reserve enough buffers: "
 584                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 585                 /* NB: this error handler is not strictly correct, because
 586                  * it may release more buffers than already allocated,
 587                  * but it doesn't matter because request portal should
 588                  * be lazy portal and will grow buffers if necessary. */
 589                 srpc_service_remove_buffers(svc, nbuf);
 590                 return -ENOMEM;
 591         }
 592
 593         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 594                nbuf * (srpc_serv_is_framework(svc) ?
 595                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 596         return 0;
 597 }
 598
 599 static void
 600 sfw_unload_test(struct sfw_test_instance *tsi)
 601 {
 602         struct sfw_test_case *tsc;
 603
 604         LASSERT(tsi != NULL);
 605         tsc = sfw_find_test_case(tsi->tsi_service);
 606         LASSERT(tsc != NULL);
 607
 608         if (tsi->tsi_is_client)
 609                 return;
 610
 611         /* shrink buffers, because request portal is lazy portal
 612          * which can grow buffers at runtime so we may leave
 613          * some buffers behind, but never mind... */
 614         srpc_service_remove_buffers(tsc->tsc_srv_service,
 615                                     sfw_test_buffers(tsi));
 616         return;
 617 }
 618
 619 static void
 620 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 621 {
 622         srpc_client_rpc_t *rpc;
 623         sfw_test_unit_t   *tsu;
 624
 625         if (!tsi->tsi_is_client) goto clean;
 626
 627         tsi->tsi_ops->tso_fini(tsi);
 628
 629         LASSERT(!tsi->tsi_stopping);
 630         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 631         LASSERT(!sfw_test_active(tsi));
 632
 633         while (!list_empty(&tsi->tsi_units)) {
 634                 tsu = list_entry(tsi->tsi_units.next,
 635                                  sfw_test_unit_t, tsu_list);
 636                 list_del(&tsu->tsu_list);
 637                 LIBCFS_FREE(tsu, sizeof(*tsu));
 638         }
 639
 640         while (!list_empty(&tsi->tsi_free_rpcs)) {
 641                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 642                                  srpc_client_rpc_t, crpc_list);
 643                 list_del(&rpc->crpc_list);
 644                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 645         }
 646
 647 clean:
 648         sfw_unload_test(tsi);
 649         LIBCFS_FREE(tsi, sizeof(*tsi));
 650         return;
 651 }
 652
 653 static void
 654 sfw_destroy_batch (sfw_batch_t *tsb)
 655 {
 656         sfw_test_instance_t *tsi;
 657
 658         LASSERT(!sfw_batch_active(tsb));
 659         LASSERT(list_empty(&tsb->bat_list));
 660
 661         while (!list_empty(&tsb->bat_tests)) {
 662                 tsi = list_entry(tsb->bat_tests.next,
 663                                  sfw_test_instance_t, tsi_list);
 664                 list_del_init(&tsi->tsi_list);
 665                 sfw_destroy_test_instance(tsi);
 666         }
 667
 668         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 669         return;
 670 }
 671
 672 void
 673 sfw_destroy_session (sfw_session_t *sn)
 674 {
 675         sfw_batch_t *batch;
 676
 677         LASSERT(list_empty(&sn->sn_list));
 678         LASSERT(sn != sfw_data.fw_session);
 679
 680         while (!list_empty(&sn->sn_batches)) {
 681                 batch = list_entry(sn->sn_batches.next,
 682                                    sfw_batch_t, bat_list);
 683                 list_del_init(&batch->bat_list);
 684                 sfw_destroy_batch(batch);
 685         }
 686
 687         LIBCFS_FREE(sn, sizeof(*sn));
 688         atomic_dec(&sfw_data.fw_nzombies);
 689         return;
 690 }
 691
 692 static void
 693 sfw_unpack_addtest_req(srpc_msg_t *msg)
 694 {
 695         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 696
 697         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 698         LASSERT (req->tsr_is_client);
 699
 700         if (msg->msg_magic == SRPC_MSG_MAGIC)
 701                 return; /* no flipping needed */
 702
 703         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 704
 705         if (req->tsr_service == SRPC_SERVICE_BRW) {
 706                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 707                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 708
 709                         __swab32s(&bulk->blk_opc);
 710                         __swab32s(&bulk->blk_npg);
 711                         __swab32s(&bulk->blk_flags);
 712
 713                 } else {
 714                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 715
 716                         __swab16s(&bulk->blk_opc);
 717                         __swab16s(&bulk->blk_flags);
 718                         __swab32s(&bulk->blk_offset);
 719                         __swab32s(&bulk->blk_len);
 720                 }
 721
 722                 return;
 723         }
 724
 725         if (req->tsr_service == SRPC_SERVICE_PING) {
 726                 test_ping_req_t *ping = &req->tsr_u.ping;
 727
 728                 __swab32s(&ping->png_size);
 729                 __swab32s(&ping->png_flags);
 730                 return;
 731         }
 732
 733         LBUG();
 734         return;
 735 }
 736
 737 static int
 738 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 739 {
 740         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 741         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 742         srpc_bulk_t         *bk = rpc->srpc_bulk;
 743         int                  ndest = req->tsr_ndest;
 744         sfw_test_unit_t     *tsu;
 745         sfw_test_instance_t *tsi;
 746         int                  i;
 747         int                  rc;
 748
 749         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 750         if (tsi == NULL) {
 751                 CERROR ("Can't allocate test instance for batch: %llu\n",
 752                         tsb->bat_id.bat_id);
 753                 return -ENOMEM;
 754         }
 755
 756         spin_lock_init(&tsi->tsi_lock);
 757         atomic_set(&tsi->tsi_nactive, 0);
 758         INIT_LIST_HEAD(&tsi->tsi_units);
 759         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 760         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 761
 762         tsi->tsi_stopping      = 0;
 763         tsi->tsi_batch         = tsb;
 764         tsi->tsi_loop          = req->tsr_loop;
 765         tsi->tsi_concur        = req->tsr_concur;
 766         tsi->tsi_service       = req->tsr_service;
 767         tsi->tsi_is_client     = !!(req->tsr_is_client);
 768         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 769
 770         rc = sfw_load_test(tsi);
 771         if (rc != 0) {
 772                 LIBCFS_FREE(tsi, sizeof(*tsi));
 773                 return rc;
 774         }
 775
 776         LASSERT (!sfw_batch_active(tsb));
 777
 778         if (!tsi->tsi_is_client) {
 779                 /* it's test server, just add it to tsb */
 780                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 781                 return 0;
 782         }
 783
 784         LASSERT (bk != NULL);
 785         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 786         LASSERT((unsigned int)bk->bk_len >=
 787                 sizeof(struct lnet_process_id_packed) * ndest);
 788
 789         sfw_unpack_addtest_req(msg);
 790         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 791
 792         for (i = 0; i < ndest; i++) {
 793                 struct lnet_process_id_packed *dests;
 794                 struct lnet_process_id_packed  id;
 795                 int                       j;
 796
 797                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 798                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 799                 id = dests[i % SFW_ID_PER_PAGE];
 800                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 801                         sfw_unpack_id(id);
 802
 803                 for (j = 0; j < tsi->tsi_concur; j++) {
 804                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 805                         if (tsu == NULL) {
 806                                 rc = -ENOMEM;
 807                                 CERROR ("Can't allocate tsu for %d\n",
 808                                         tsi->tsi_service);
 809                                 goto error;
 810                         }
 811
 812                         tsu->tsu_dest.nid = id.nid;
 813                         tsu->tsu_dest.pid = id.pid;
 814                         tsu->tsu_instance = tsi;
 815                         tsu->tsu_private  = NULL;
 816                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 817                 }
 818         }
 819
 820         rc = tsi->tsi_ops->tso_init(tsi);
 821         if (rc == 0) {
 822                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 823                 return 0;
 824         }
 825
 826 error:
 827         LASSERT(rc != 0);
 828         sfw_destroy_test_instance(tsi);
 829         return rc;
 830 }
 831
 832 static void
 833 sfw_test_unit_done (sfw_test_unit_t *tsu)
 834 {
 835         sfw_test_instance_t *tsi = tsu->tsu_instance;
 836         sfw_batch_t         *tsb = tsi->tsi_batch;
 837         sfw_session_t       *sn = tsb->bat_session;
 838
 839         LASSERT (sfw_test_active(tsi));
 840
 841         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 842                 return;
 843
 844         /* the test instance is done */
 845         spin_lock(&tsi->tsi_lock);
 846
 847         tsi->tsi_stopping = 0;
 848
 849         spin_unlock(&tsi->tsi_lock);
 850
 851         spin_lock(&sfw_data.fw_lock);
 852
 853         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 854             sn == sfw_data.fw_session) {                  /* sn also active */
 855                 spin_unlock(&sfw_data.fw_lock);
 856                 return;
 857         }
 858
 859         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 860
 861         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 862                 if (sfw_batch_active(tsb)) {
 863                         spin_unlock(&sfw_data.fw_lock);
 864                         return;
 865                 }
 866         }
 867
 868         list_del_init(&sn->sn_list);
 869         spin_unlock(&sfw_data.fw_lock);
 870
 871         sfw_destroy_session(sn);
 872         return;
 873 }
 874
 875 static void
 876 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 877 {
 878         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 879         sfw_test_instance_t *tsi = tsu->tsu_instance;
 880         int                  done = 0;
 881
 882         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 883
 884         spin_lock(&tsi->tsi_lock);
 885
 886         LASSERT(sfw_test_active(tsi));
 887         LASSERT(!list_empty(&rpc->crpc_list));
 888
 889         list_del_init(&rpc->crpc_list);
 890
 891         /* batch is stopping or loop is done or get error */
 892         if (tsi->tsi_stopping ||
 893             tsu->tsu_loop == 0 ||
 894             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 895                 done = 1;
 896
 897         /* dec ref for poster */
 898         srpc_client_rpc_decref(rpc);
 899
 900         spin_unlock(&tsi->tsi_lock);
 901
 902         if (!done) {
 903                 swi_schedule_workitem(&tsu->tsu_worker);
 904                 return;
 905         }
 906
 907         sfw_test_unit_done(tsu);
 908         return;
 909 }
 910
 911 int
 912 sfw_create_test_rpc(sfw_test_unit_t *tsu, struct lnet_process_id peer,
 913                     unsigned features, int nblk, int blklen,
 914                     srpc_client_rpc_t **rpcpp)
 915 {
 916         srpc_client_rpc_t   *rpc = NULL;
 917         sfw_test_instance_t *tsi = tsu->tsu_instance;
 918
 919         spin_lock(&tsi->tsi_lock);
 920
 921         LASSERT (sfw_test_active(tsi));
 922
 923         if (!list_empty(&tsi->tsi_free_rpcs)) {
 924                 /* pick request from buffer */
 925                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 926                                  srpc_client_rpc_t, crpc_list);
 927                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 928                 list_del_init(&rpc->crpc_list);
 929         }
 930
 931         spin_unlock(&tsi->tsi_lock);
 932
 933         if (rpc == NULL) {
 934                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 935                                              blklen, sfw_test_rpc_done,
 936                                              sfw_test_rpc_fini, tsu);
 937         } else {
 938                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 939                                      blklen, sfw_test_rpc_done,
 940                                      sfw_test_rpc_fini, tsu);
 941         }
 942
 943         if (rpc == NULL) {
 944                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 945                 return -ENOMEM;
 946         }
 947
 948         rpc->crpc_reqstmsg.msg_ses_feats = features;
 949         *rpcpp = rpc;
 950
 951         return 0;
 952 }
 953
 954 static int
 955 sfw_run_test (swi_workitem_t *wi)
 956 {
 957         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 958         sfw_test_instance_t *tsi = tsu->tsu_instance;
 959         srpc_client_rpc_t   *rpc = NULL;
 960
 961         LASSERT (wi == &tsu->tsu_worker);
 962
 963         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 964                 LASSERT (rpc == NULL);
 965                 goto test_done;
 966         }
 967
 968         LASSERT (rpc != NULL);
 969
 970         spin_lock(&tsi->tsi_lock);
 971
 972         if (tsi->tsi_stopping) {
 973                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 974                 spin_unlock(&tsi->tsi_lock);
 975                 goto test_done;
 976         }
 977
 978         if (tsu->tsu_loop > 0)
 979                 tsu->tsu_loop--;
 980
 981         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 982         spin_unlock(&tsi->tsi_lock);
 983
 984         spin_lock(&rpc->crpc_lock);
 985         rpc->crpc_timeout = rpc_timeout;
 986         srpc_post_rpc(rpc);
 987         spin_unlock(&rpc->crpc_lock);
 988         return 0;
 989
 990 test_done:
 991         /*
 992          * No one can schedule me now since:
 993          * - previous RPC, if any, has done and
 994          * - no new RPC is initiated.
 995          * - my batch is still active; no one can run it again now.
 996          * Cancel pending schedules and prevent future schedule attempts:
 997          */
 998         swi_exit_workitem(wi);
 999         sfw_test_unit_done(tsu);
1000         return 1;
1001 }
1002
1003 static int
1004 sfw_run_batch (sfw_batch_t *tsb)
1005 {
1006         swi_workitem_t      *wi;
1007         sfw_test_unit_t     *tsu;
1008         sfw_test_instance_t *tsi;
1009
1010         if (sfw_batch_active(tsb)) {
1011                 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1012                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1013                 return 0;
1014         }
1015
1016         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1017                 if (!tsi->tsi_is_client)        /* skip server instances */
1018                         continue;
1019
1020                 LASSERT(!tsi->tsi_stopping);
1021                 LASSERT(!sfw_test_active(tsi));
1022
1023                 atomic_inc(&tsb->bat_nactive);
1024
1025                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1026                         atomic_inc(&tsi->tsi_nactive);
1027                         tsu->tsu_loop = tsi->tsi_loop;
1028                         wi = &tsu->tsu_worker;
1029                         swi_init_workitem(wi, tsu, sfw_run_test,
1030                                           lst_sched_test[\
1031                                           lnet_cpt_of_nid(tsu->tsu_dest.nid,
1032                                                           NULL)]);
1033                         swi_schedule_workitem(wi);
1034                 }
1035         }
1036
1037         return 0;
1038 }
1039
1040 int
1041 sfw_stop_batch (sfw_batch_t *tsb, int force)
1042 {
1043         sfw_test_instance_t *tsi;
1044         srpc_client_rpc_t   *rpc;
1045
1046         if (!sfw_batch_active(tsb)) {
1047                 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1048                 return 0;
1049         }
1050
1051         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1052                 spin_lock(&tsi->tsi_lock);
1053
1054                 if (!tsi->tsi_is_client ||
1055                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1056                         spin_unlock(&tsi->tsi_lock);
1057                         continue;
1058                 }
1059
1060                 tsi->tsi_stopping = 1;
1061
1062                 if (!force) {
1063                         spin_unlock(&tsi->tsi_lock);
1064                         continue;
1065                 }
1066
1067                 /* abort launched rpcs in the test */
1068                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1069                         spin_lock(&rpc->crpc_lock);
1070
1071                         srpc_abort_rpc(rpc, -EINTR);
1072
1073                         spin_unlock(&rpc->crpc_lock);
1074                 }
1075
1076                 spin_unlock(&tsi->tsi_lock);
1077         }
1078
1079         return 0;
1080 }
1081
1082 static int
1083 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1084 {
1085         sfw_test_instance_t *tsi;
1086
1087         if (testidx < 0)
1088                 return -EINVAL;
1089
1090         if (testidx == 0) {
1091                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1092                 return 0;
1093         }
1094
1095         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1096                 if (testidx-- > 1)
1097                         continue;
1098
1099                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1100                 return 0;
1101         }
1102
1103         return -ENOENT;
1104 }
1105
1106 void
1107 sfw_free_pages (srpc_server_rpc_t *rpc)
1108 {
1109         srpc_free_bulk(rpc->srpc_bulk);
1110         rpc->srpc_bulk = NULL;
1111 }
1112
1113 int
1114 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1115                 int sink)
1116 {
1117         LASSERT(rpc->srpc_bulk == NULL);
1118         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1119
1120         rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1121         if (rpc->srpc_bulk == NULL)
1122                 return -ENOMEM;
1123
1124         return 0;
1125 }
1126
1127 static int
1128 sfw_add_test (srpc_server_rpc_t *rpc)
1129 {
1130         sfw_session_t     *sn = sfw_data.fw_session;
1131         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1132         srpc_test_reqst_t *request;
1133         int                rc;
1134         sfw_batch_t       *bat;
1135
1136         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1137         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1138
1139         if (request->tsr_loop == 0 ||
1140             request->tsr_concur == 0 ||
1141             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1142             request->tsr_ndest > SFW_MAX_NDESTS ||
1143             (request->tsr_is_client && request->tsr_ndest == 0) ||
1144             request->tsr_concur > SFW_MAX_CONCUR ||
1145             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1146             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1147                 reply->tsr_status = EINVAL;
1148                 return 0;
1149         }
1150
1151         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1152             sfw_find_test_case(request->tsr_service) == NULL) {
1153                 reply->tsr_status = ENOENT;
1154                 return 0;
1155         }
1156
1157         bat = sfw_bid2batch(request->tsr_bid);
1158         if (bat == NULL) {
1159                 CERROR("dropping RPC %s from %s under memory pressure\n",
1160                         rpc->srpc_scd->scd_svc->sv_name,
1161                         libcfs_id2str(rpc->srpc_peer));
1162                 return -ENOMEM;
1163         }
1164
1165         if (sfw_batch_active(bat)) {
1166                 reply->tsr_status = EBUSY;
1167                 return 0;
1168         }
1169
1170         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1171                 /* rpc will be resumed later in sfw_bulk_ready */
1172                 int     npg = sfw_id_pages(request->tsr_ndest);
1173                 int     len;
1174
1175                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1176                         len = npg * PAGE_SIZE;
1177
1178                 } else  {
1179                         len = sizeof(struct lnet_process_id_packed) *
1180                               request->tsr_ndest;
1181                 }
1182
1183                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1184         }
1185
1186         rc = sfw_add_test_instance(bat, rpc);
1187         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1188                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1189                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1190                 request->tsr_is_client ? "client" : "server",
1191                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1192
1193         reply->tsr_status = (rc < 0) ? -rc : rc;
1194         return 0;
1195 }
1196
1197 static int
1198 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1199 {
1200         sfw_session_t *sn = sfw_data.fw_session;
1201         int            rc = 0;
1202         sfw_batch_t   *bat;
1203
1204         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1205
1206         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1207                 reply->bar_status = ESRCH;
1208                 return 0;
1209         }
1210
1211         bat = sfw_find_batch(request->bar_bid);
1212         if (bat == NULL) {
1213                 reply->bar_status = ENOENT;
1214                 return 0;
1215         }
1216
1217         switch (request->bar_opc) {
1218         case SRPC_BATCH_OPC_RUN:
1219                 rc = sfw_run_batch(bat);
1220                 break;
1221
1222         case SRPC_BATCH_OPC_STOP:
1223                 rc = sfw_stop_batch(bat, request->bar_arg);
1224                 break;
1225
1226         case SRPC_BATCH_OPC_QUERY:
1227                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1228                 break;
1229
1230         default:
1231                 return -EINVAL; /* drop it */
1232         }
1233
1234         reply->bar_status = (rc < 0) ? -rc : rc;
1235         return 0;
1236 }
1237
1238 static int
1239 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1240 {
1241         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1242         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1243         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1244         unsigned        features = LST_FEATS_MASK;
1245         int             rc = 0;
1246
1247         LASSERT(sfw_data.fw_active_srpc == NULL);
1248         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1249
1250         spin_lock(&sfw_data.fw_lock);
1251
1252         if (sfw_data.fw_shuttingdown) {
1253                 spin_unlock(&sfw_data.fw_lock);
1254                 return -ESHUTDOWN;
1255         }
1256
1257         /* Remove timer to avoid racing with it or expiring active session */
1258         if (sfw_del_session_timer() != 0) {
1259                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1260                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1261                 spin_unlock(&sfw_data.fw_lock);
1262                 return -EAGAIN;
1263         }
1264
1265         sfw_data.fw_active_srpc = rpc;
1266         spin_unlock(&sfw_data.fw_lock);
1267
1268         sfw_unpack_message(request);
1269         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1270
1271         /* rpc module should have checked this */
1272         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1273
1274         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1275             sv->sv_id != SRPC_SERVICE_DEBUG) {
1276                 sfw_session_t *sn = sfw_data.fw_session;
1277
1278                 if (sn != NULL &&
1279                     sn->sn_features != request->msg_ses_feats) {
1280                         CNETERR("Features of framework RPC don't match "
1281                                 "features of current session: %x/%x\n",
1282                                 request->msg_ses_feats, sn->sn_features);
1283                         reply->msg_body.reply.status = EPROTO;
1284                         reply->msg_body.reply.sid    = sn->sn_id;
1285                         goto out;
1286                 }
1287
1288         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1289                 /* NB: at this point, old version will ignore features and
1290                  * create new session anyway, so console should be able
1291                  * to handle this */
1292                 reply->msg_body.reply.status = EPROTO;
1293                 goto out;
1294         }
1295
1296         switch(sv->sv_id) {
1297         default:
1298                 LBUG ();
1299         case SRPC_SERVICE_TEST:
1300                 rc = sfw_add_test(rpc);
1301                 break;
1302
1303         case SRPC_SERVICE_BATCH:
1304                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1305                                        &reply->msg_body.bat_reply);
1306                 break;
1307
1308         case SRPC_SERVICE_QUERY_STAT:
1309                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1310                                    &reply->msg_body.stat_reply);
1311                 break;
1312
1313         case SRPC_SERVICE_DEBUG:
1314                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1315                                        &reply->msg_body.dbg_reply);
1316                 break;
1317
1318         case SRPC_SERVICE_MAKE_SESSION:
1319                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1320                                       &reply->msg_body.mksn_reply);
1321                 break;
1322
1323         case SRPC_SERVICE_REMOVE_SESSION:
1324                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1325                                         &reply->msg_body.rmsn_reply);
1326                 break;
1327         }
1328
1329         if (sfw_data.fw_session != NULL)
1330                 features = sfw_data.fw_session->sn_features;
1331  out:
1332         reply->msg_ses_feats = features;
1333         rpc->srpc_done = sfw_server_rpc_done;
1334         spin_lock(&sfw_data.fw_lock);
1335
1336         if (!sfw_data.fw_shuttingdown)
1337                 sfw_add_session_timer();
1338
1339         sfw_data.fw_active_srpc = NULL;
1340         spin_unlock(&sfw_data.fw_lock);
1341         return rc;
1342 }
1343
1344 static int
1345 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1346 {
1347         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1348         int                     rc;
1349
1350         LASSERT(rpc->srpc_bulk != NULL);
1351         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1352         LASSERT(sfw_data.fw_active_srpc == NULL);
1353         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1354
1355         spin_lock(&sfw_data.fw_lock);
1356
1357         if (status != 0) {
1358                 CERROR("Bulk transfer failed for RPC: "
1359                        "service %s, peer %s, status %d\n",
1360                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1361                 spin_unlock(&sfw_data.fw_lock);
1362                 return -EIO;
1363         }
1364
1365         if (sfw_data.fw_shuttingdown) {
1366                 spin_unlock(&sfw_data.fw_lock);
1367                 return -ESHUTDOWN;
1368         }
1369
1370         if (sfw_del_session_timer() != 0) {
1371                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1372                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1373                 spin_unlock(&sfw_data.fw_lock);
1374                 return -EAGAIN;
1375         }
1376
1377         sfw_data.fw_active_srpc = rpc;
1378         spin_unlock(&sfw_data.fw_lock);
1379
1380         rc = sfw_add_test(rpc);
1381
1382         spin_lock(&sfw_data.fw_lock);
1383
1384         if (!sfw_data.fw_shuttingdown)
1385                 sfw_add_session_timer();
1386
1387         sfw_data.fw_active_srpc = NULL;
1388         spin_unlock(&sfw_data.fw_lock);
1389         return rc;
1390 }
1391
1392 srpc_client_rpc_t *
1393 sfw_create_rpc(struct lnet_process_id peer, int service,
1394                unsigned features, int nbulkiov, int bulklen,
1395                void (*done)(srpc_client_rpc_t *), void *priv)
1396 {
1397         srpc_client_rpc_t *rpc = NULL;
1398
1399         spin_lock(&sfw_data.fw_lock);
1400
1401         LASSERT (!sfw_data.fw_shuttingdown);
1402         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1403
1404         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1405                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1406                                      srpc_client_rpc_t, crpc_list);
1407                 list_del(&rpc->crpc_list);
1408
1409                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1410                                      done, sfw_client_rpc_fini, priv);
1411         }
1412
1413         spin_unlock(&sfw_data.fw_lock);
1414
1415         if (rpc == NULL) {
1416                 rpc = srpc_create_client_rpc(peer, service,
1417                                              nbulkiov, bulklen, done,
1418                                              nbulkiov != 0 ?  NULL :
1419                                              sfw_client_rpc_fini,
1420                                              priv);
1421         }
1422
1423         if (rpc != NULL) /* "session" is concept in framework */
1424                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1425
1426         return rpc;
1427 }
1428
1429 void
1430 sfw_unpack_message (srpc_msg_t *msg)
1431 {
1432         if (msg->msg_magic == SRPC_MSG_MAGIC)
1433                 return; /* no flipping needed */
1434
1435         /* srpc module should guarantee I wouldn't get crap */
1436         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1437
1438         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1439                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1440
1441                 __swab32s(&req->str_type);
1442                 __swab64s(&req->str_rpyid);
1443                 sfw_unpack_sid(req->str_sid);
1444                 return;
1445         }
1446
1447         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1448                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1449
1450                 __swab32s(&rep->str_status);
1451                 sfw_unpack_sid(rep->str_sid);
1452                 sfw_unpack_fw_counters(rep->str_fw);
1453                 sfw_unpack_rpc_counters(rep->str_rpc);
1454                 sfw_unpack_lnet_counters(rep->str_lnet);
1455                 return;
1456         }
1457
1458         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1459                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1460
1461                 __swab64s(&req->mksn_rpyid);
1462                 __swab32s(&req->mksn_force);
1463                 sfw_unpack_sid(req->mksn_sid);
1464                 return;
1465         }
1466
1467         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1468                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1469
1470                 __swab32s(&rep->mksn_status);
1471                 __swab32s(&rep->mksn_timeout);
1472                 sfw_unpack_sid(rep->mksn_sid);
1473                 return;
1474         }
1475
1476         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1477                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1478
1479                 __swab64s(&req->rmsn_rpyid);
1480                 sfw_unpack_sid(req->rmsn_sid);
1481                 return;
1482         }
1483
1484         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1485                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1486
1487                 __swab32s(&rep->rmsn_status);
1488                 sfw_unpack_sid(rep->rmsn_sid);
1489                 return;
1490         }
1491
1492         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1493                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1494
1495                 __swab64s(&req->dbg_rpyid);
1496                 __swab32s(&req->dbg_flags);
1497                 sfw_unpack_sid(req->dbg_sid);
1498                 return;
1499         }
1500
1501         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1502                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1503
1504                 __swab32s(&rep->dbg_nbatch);
1505                 __swab32s(&rep->dbg_timeout);
1506                 sfw_unpack_sid(rep->dbg_sid);
1507                 return;
1508         }
1509
1510         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1511                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1512
1513                 __swab32s(&req->bar_opc);
1514                 __swab64s(&req->bar_rpyid);
1515                 __swab32s(&req->bar_testidx);
1516                 __swab32s(&req->bar_arg);
1517                 sfw_unpack_sid(req->bar_sid);
1518                 __swab64s(&req->bar_bid.bat_id);
1519                 return;
1520         }
1521
1522         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1523                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1524
1525                 __swab32s(&rep->bar_status);
1526                 sfw_unpack_sid(rep->bar_sid);
1527                 return;
1528         }
1529
1530         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1531                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1532
1533                 __swab64s(&req->tsr_rpyid);
1534                 __swab64s(&req->tsr_bulkid);
1535                 __swab32s(&req->tsr_loop);
1536                 __swab32s(&req->tsr_ndest);
1537                 __swab32s(&req->tsr_concur);
1538                 __swab32s(&req->tsr_service);
1539                 sfw_unpack_sid(req->tsr_sid);
1540                 __swab64s(&req->tsr_bid.bat_id);
1541                 return;
1542         }
1543
1544         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1545                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1546
1547                 __swab32s(&rep->tsr_status);
1548                 sfw_unpack_sid(rep->tsr_sid);
1549                 return;
1550         }
1551
1552         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1553                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1554
1555                 __swab64s(&req->join_rpyid);
1556                 sfw_unpack_sid(req->join_sid);
1557                 return;
1558         }
1559
1560         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1561                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1562
1563                 __swab32s(&rep->join_status);
1564                 __swab32s(&rep->join_timeout);
1565                 sfw_unpack_sid(rep->join_sid);
1566                 return;
1567         }
1568
1569         LBUG ();
1570         return;
1571 }
1572
1573 void
1574 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1575 {
1576         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1577         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1578
1579         spin_lock(&rpc->crpc_lock);
1580         srpc_abort_rpc(rpc, -EINTR);
1581         spin_unlock(&rpc->crpc_lock);
1582         return;
1583 }
1584
1585 void
1586 sfw_post_rpc (srpc_client_rpc_t *rpc)
1587 {
1588         spin_lock(&rpc->crpc_lock);
1589
1590         LASSERT(!rpc->crpc_closed);
1591         LASSERT(!rpc->crpc_aborted);
1592         LASSERT(list_empty(&rpc->crpc_list));
1593         LASSERT(!sfw_data.fw_shuttingdown);
1594
1595         rpc->crpc_timeout = rpc_timeout;
1596         srpc_post_rpc(rpc);
1597
1598         spin_unlock(&rpc->crpc_lock);
1599         return;
1600 }
1601
1602 static srpc_service_t sfw_services[] =
1603 {
1604         {
1605                 /* sv_id */    SRPC_SERVICE_DEBUG,
1606                 /* sv_name */  "debug",
1607                 0
1608         },
1609         {
1610                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1611                 /* sv_name */  "query stats",
1612                 0
1613         },
1614         {
1615                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1616                 /* sv_name */  "make session",
1617                 0
1618         },
1619         {
1620                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1621                 /* sv_name */  "remove session",
1622                 0
1623         },
1624         {
1625                 /* sv_id */    SRPC_SERVICE_BATCH,
1626                 /* sv_name */  "batch service",
1627                 0
1628         },
1629         {
1630                 /* sv_id */    SRPC_SERVICE_TEST,
1631                 /* sv_name */  "test service",
1632                 0
1633         },
1634         {
1635                 /* sv_id */    0,
1636                 /* sv_name */  NULL,
1637                 0
1638         }
1639 };
1640
1641 int
1642 sfw_startup (void)
1643 {
1644         int              i;
1645         int              rc;
1646         int              error;
1647         srpc_service_t  *sv;
1648         sfw_test_case_t *tsc;
1649
1650
1651         if (session_timeout < 0) {
1652                 CERROR ("Session timeout must be non-negative: %d\n",
1653                         session_timeout);
1654                 return -EINVAL;
1655         }
1656
1657         if (rpc_timeout < 0) {
1658                 CERROR ("RPC timeout must be non-negative: %d\n",
1659                         rpc_timeout);
1660                 return -EINVAL;
1661         }
1662
1663         if (session_timeout == 0)
1664                 CWARN ("Zero session_timeout specified "
1665                        "- test sessions never expire.\n");
1666
1667         if (rpc_timeout == 0)
1668                 CWARN ("Zero rpc_timeout specified "
1669                        "- test RPC never expire.\n");
1670
1671         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1672
1673         sfw_data.fw_session     = NULL;
1674         sfw_data.fw_active_srpc = NULL;
1675         spin_lock_init(&sfw_data.fw_lock);
1676         atomic_set(&sfw_data.fw_nzombies, 0);
1677         INIT_LIST_HEAD(&sfw_data.fw_tests);
1678         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1679         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1680
1681         brw_init_test_client();
1682         brw_init_test_service();
1683         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1684         LASSERT (rc == 0);
1685
1686         ping_init_test_client();
1687         ping_init_test_service();
1688         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1689         LASSERT (rc == 0);
1690
1691         error = 0;
1692         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1693                 sv = tsc->tsc_srv_service;
1694
1695                 rc = srpc_add_service(sv);
1696                 LASSERT(rc != -EBUSY);
1697                 if (rc != 0) {
1698                         CWARN("Failed to add %s service: %d\n",
1699                               sv->sv_name, rc);
1700                         error = rc;
1701                 }
1702         }
1703
1704         for (i = 0; ; i++) {
1705                 sv = &sfw_services[i];
1706                 if (sv->sv_name == NULL) break;
1707
1708                 sv->sv_bulk_ready = NULL;
1709                 sv->sv_handler    = sfw_handle_server_rpc;
1710                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1711                 if (sv->sv_id == SRPC_SERVICE_TEST)
1712                         sv->sv_bulk_ready = sfw_bulk_ready;
1713
1714                 rc = srpc_add_service(sv);
1715                 LASSERT (rc != -EBUSY);
1716                 if (rc != 0) {
1717                         CWARN ("Failed to add %s service: %d\n",
1718                                sv->sv_name, rc);
1719                         error = rc;
1720                 }
1721
1722                 /* about to sfw_shutdown, no need to add buffer */
1723                 if (error) continue;
1724
1725                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1726                 if (rc != 0) {
1727                         CWARN("Failed to reserve enough buffers: "
1728                               "service %s, %d needed: %d\n",
1729                               sv->sv_name, sv->sv_wi_total, rc);
1730                         error = -ENOMEM;
1731                 }
1732         }
1733
1734         if (error != 0)
1735                 sfw_shutdown();
1736         return error;
1737 }
1738
1739 void
1740 sfw_shutdown (void)
1741 {
1742         srpc_service_t  *sv;
1743         sfw_test_case_t *tsc;
1744         int              i;
1745
1746         spin_lock(&sfw_data.fw_lock);
1747
1748         sfw_data.fw_shuttingdown = 1;
1749         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1750                        "waiting for active RPC to finish.\n");
1751
1752         if (sfw_del_session_timer() != 0)
1753                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1754                                "waiting for session timer to explode.\n");
1755
1756         sfw_deactivate_session();
1757         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1758                        sfw_data.fw_lock,
1759                        "waiting for %d zombie sessions to die.\n",
1760                        atomic_read(&sfw_data.fw_nzombies));
1761
1762         spin_unlock(&sfw_data.fw_lock);
1763
1764         for (i = 0; ; i++) {
1765                 sv = &sfw_services[i];
1766                 if (sv->sv_name == NULL)
1767                         break;
1768
1769                 srpc_shutdown_service(sv);
1770                 srpc_remove_service(sv);
1771         }
1772
1773         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1774                 sv = tsc->tsc_srv_service;
1775                 srpc_shutdown_service(sv);
1776                 srpc_remove_service(sv);
1777         }
1778
1779         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1780                 srpc_client_rpc_t *rpc;
1781
1782                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1783                                  srpc_client_rpc_t, crpc_list);
1784                 list_del(&rpc->crpc_list);
1785
1786                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1787         }
1788
1789         for (i = 0; ; i++) {
1790                 sv = &sfw_services[i];
1791                 if (sv->sv_name == NULL)
1792                         break;
1793
1794                 srpc_wait_service_shutdown(sv);
1795         }
1796
1797         while (!list_empty(&sfw_data.fw_tests)) {
1798                 tsc = list_entry(sfw_data.fw_tests.next,
1799                                  sfw_test_case_t, tsc_list);
1800
1801                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1802
1803                 list_del(&tsc->tsc_list);
1804                 LIBCFS_FREE(tsc, sizeof(*tsc));
1805         }
1806
1807         return;
1808 }