lnet/selftest/framework.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lnet/selftest/framework.c
  37  *
  38  * Author: Isaac Huang <isaac@clusterfs.com>
  39  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LNET
  43
  44 #include "selftest.h"
  45
  46 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
  47
  48 static int session_timeout = 100;
  49 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  50                 "test session timeout in seconds (100 by default, 0 == never)");
  51
  52 static int rpc_timeout = 64;
  53 CFS_MODULE_PARM(rpc_timeout, "i", int, 0644,
  54                 "rpc timeout in seconds (64 by default, 0 == never)");
  55
  56 #define sfw_unpack_id(id)               \
  57 do {                                    \
  58         __swab64s(&(id).nid);           \
  59         __swab32s(&(id).pid);           \
  60 } while (0)
  61
  62 #define sfw_unpack_sid(sid)             \
  63 do {                                    \
  64         __swab64s(&(sid).ses_nid);      \
  65         __swab64s(&(sid).ses_stamp);    \
  66 } while (0)
  67
  68 #define sfw_unpack_fw_counters(fc)        \
  69 do {                                      \
  70         __swab32s(&(fc).running_ms);      \
  71         __swab32s(&(fc).active_batches);  \
  72         __swab32s(&(fc).zombie_sessions); \
  73         __swab32s(&(fc).brw_errors);      \
  74         __swab32s(&(fc).ping_errors);     \
  75 } while (0)
  76
  77 #define sfw_unpack_rpc_counters(rc)     \
  78 do {                                    \
  79         __swab32s(&(rc).errors);        \
  80         __swab32s(&(rc).rpcs_sent);     \
  81         __swab32s(&(rc).rpcs_rcvd);     \
  82         __swab32s(&(rc).rpcs_dropped);  \
  83         __swab32s(&(rc).rpcs_expired);  \
  84         __swab64s(&(rc).bulk_get);      \
  85         __swab64s(&(rc).bulk_put);      \
  86 } while (0)
  87
  88 #define sfw_unpack_lnet_counters(lc)    \
  89 do {                                    \
  90         __swab32s(&(lc).errors);        \
  91         __swab32s(&(lc).msgs_max);      \
  92         __swab32s(&(lc).msgs_alloc);    \
  93         __swab32s(&(lc).send_count);    \
  94         __swab32s(&(lc).recv_count);    \
  95         __swab32s(&(lc).drop_count);    \
  96         __swab32s(&(lc).route_count);   \
  97         __swab64s(&(lc).send_length);   \
  98         __swab64s(&(lc).recv_length);   \
  99         __swab64s(&(lc).drop_length);   \
 100         __swab64s(&(lc).route_length);  \
 101 } while (0)
 102
 103 #define sfw_test_active(t)      (cfs_atomic_read(&(t)->tsi_nactive) != 0)
 104 #define sfw_batch_active(b)     (cfs_atomic_read(&(b)->bat_nactive) != 0)
 105
 106 struct smoketest_framework {
 107         cfs_list_t         fw_zombie_rpcs;     /* RPCs to be recycled */
 108         cfs_list_t         fw_zombie_sessions; /* stopping sessions */
 109         cfs_list_t         fw_tests;           /* registered test cases */
 110         cfs_atomic_t       fw_nzombies;        /* # zombie sessions */
 111         spinlock_t         fw_lock;             /* serialise */
 112         sfw_session_t     *fw_session;          /* _the_ session */
 113         int                fw_shuttingdown;     /* shutdown in progress */
 114         srpc_server_rpc_t *fw_active_srpc;      /* running RPC */
 115 } sfw_data;
 116
 117 /* forward ref's */
 118 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 119 void sfw_destroy_session (sfw_session_t *sn);
 120
 121 static inline sfw_test_case_t *
 122 sfw_find_test_case(int id)
 123 {
 124         sfw_test_case_t *tsc;
 125
 126         LASSERT (id <= SRPC_SERVICE_MAX_ID);
 127         LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 128
 129         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 130                                        sfw_test_case_t, tsc_list) {
 131                 if (tsc->tsc_srv_service->sv_id == id)
 132                         return tsc;
 133         }
 134
 135         return NULL;
 136 }
 137
 138 static int
 139 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 140 {
 141         sfw_test_case_t *tsc;
 142
 143         if (sfw_find_test_case(service->sv_id) != NULL) {
 144                 CERROR ("Failed to register test %s (%d)\n",
 145                         service->sv_name, service->sv_id);
 146                 return -EEXIST;
 147         }
 148
 149         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 150         if (tsc == NULL)
 151                 return -ENOMEM;
 152
 153         memset(tsc, 0, sizeof(sfw_test_case_t));
 154         tsc->tsc_cli_ops     = cliops;
 155         tsc->tsc_srv_service = service;
 156
 157         cfs_list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 158         return 0;
 159 }
 160
 161 void
 162 sfw_add_session_timer (void)
 163 {
 164         sfw_session_t *sn = sfw_data.fw_session;
 165         stt_timer_t   *timer = &sn->sn_timer;
 166
 167         LASSERT (!sfw_data.fw_shuttingdown);
 168
 169         if (sn == NULL || sn->sn_timeout == 0)
 170                 return;
 171
 172         LASSERT (!sn->sn_timer_active);
 173
 174         sn->sn_timer_active = 1;
 175         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 176                                           cfs_time_current_sec());
 177         stt_add_timer(timer);
 178         return;
 179 }
 180
 181 int
 182 sfw_del_session_timer (void)
 183 {
 184         sfw_session_t *sn = sfw_data.fw_session;
 185
 186         if (sn == NULL || !sn->sn_timer_active)
 187                 return 0;
 188
 189         LASSERT (sn->sn_timeout != 0);
 190
 191         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 192                 sn->sn_timer_active = 0;
 193                 return 0;
 194         }
 195
 196 #ifndef __KERNEL__
 197         /* Racing is impossible in single-threaded userland selftest */
 198         LBUG();
 199 #endif
 200         return EBUSY; /* racing with sfw_session_expired() */
 201 }
 202
 203 /* called with sfw_data.fw_lock held */
 204 static void
 205 sfw_deactivate_session (void)
 206 {
 207         sfw_session_t *sn = sfw_data.fw_session;
 208         int            nactive = 0;
 209         sfw_batch_t   *tsb;
 210         sfw_test_case_t *tsc;
 211
 212         if (sn == NULL) return;
 213
 214         LASSERT (!sn->sn_timer_active);
 215
 216         sfw_data.fw_session = NULL;
 217         cfs_atomic_inc(&sfw_data.fw_nzombies);
 218         cfs_list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 219
 220         spin_unlock(&sfw_data.fw_lock);
 221
 222         cfs_list_for_each_entry_typed(tsc, &sfw_data.fw_tests,
 223                                       sfw_test_case_t, tsc_list) {
 224                 srpc_abort_service(tsc->tsc_srv_service);
 225         }
 226
 227         spin_lock(&sfw_data.fw_lock);
 228
 229         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 230                                        sfw_batch_t, bat_list) {
 231                 if (sfw_batch_active(tsb)) {
 232                         nactive++;
 233                         sfw_stop_batch(tsb, 1);
 234                 }
 235         }
 236
 237         if (nactive != 0)
 238                 return;   /* wait for active batches to stop */
 239
 240         cfs_list_del_init(&sn->sn_list);
 241         spin_unlock(&sfw_data.fw_lock);
 242
 243         sfw_destroy_session(sn);
 244
 245         spin_lock(&sfw_data.fw_lock);
 246 }
 247
 248 #ifndef __KERNEL__
 249
 250 int
 251 sfw_session_removed (void)
 252 {
 253         return (sfw_data.fw_session == NULL) ? 1 : 0;
 254 }
 255
 256 #endif
 257
 258 void
 259 sfw_session_expired (void *data)
 260 {
 261         sfw_session_t *sn = data;
 262
 263         spin_lock(&sfw_data.fw_lock);
 264
 265         LASSERT (sn->sn_timer_active);
 266         LASSERT (sn == sfw_data.fw_session);
 267
 268         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 269                libcfs_nid2str(sn->sn_id.ses_nid),
 270                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 271
 272         sn->sn_timer_active = 0;
 273         sfw_deactivate_session();
 274
 275         spin_unlock(&sfw_data.fw_lock);
 276 }
 277
 278 static inline void
 279 sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
 280                  unsigned features, const char *name)
 281 {
 282         stt_timer_t *timer = &sn->sn_timer;
 283
 284         memset(sn, 0, sizeof(sfw_session_t));
 285         CFS_INIT_LIST_HEAD(&sn->sn_list);
 286         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 287         cfs_atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 288         cfs_atomic_set(&sn->sn_brw_errors, 0);
 289         cfs_atomic_set(&sn->sn_ping_errors, 0);
 290         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 291
 292         sn->sn_timer_active = 0;
 293         sn->sn_id           = sid;
 294         sn->sn_features     = features;
 295         sn->sn_timeout      = session_timeout;
 296         sn->sn_started      = cfs_time_current();
 297
 298         timer->stt_data = sn;
 299         timer->stt_func = sfw_session_expired;
 300         CFS_INIT_LIST_HEAD(&timer->stt_list);
 301 }
 302
 303 /* completion handler for incoming framework RPCs */
 304 void
 305 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 306 {
 307         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 308         int                     status  = rpc->srpc_status;
 309
 310         CDEBUG (D_NET,
 311                 "Incoming framework RPC done: "
 312                 "service %s, peer %s, status %s:%d\n",
 313                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 314                 swi_state2str(rpc->srpc_wi.swi_state),
 315                 status);
 316
 317         if (rpc->srpc_bulk != NULL)
 318                 sfw_free_pages(rpc);
 319         return;
 320 }
 321
 322 void
 323 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 324 {
 325         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 326         LASSERT (cfs_list_empty(&rpc->crpc_list));
 327         LASSERT (cfs_atomic_read(&rpc->crpc_refcount) == 0);
 328 #ifndef __KERNEL__
 329         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 330 #endif
 331
 332         CDEBUG (D_NET,
 333                 "Outgoing framework RPC done: "
 334                 "service %d, peer %s, status %s:%d:%d\n",
 335                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 336                 swi_state2str(rpc->crpc_wi.swi_state),
 337                 rpc->crpc_aborted, rpc->crpc_status);
 338
 339         spin_lock(&sfw_data.fw_lock);
 340
 341         /* my callers must finish all RPCs before shutting me down */
 342         LASSERT(!sfw_data.fw_shuttingdown);
 343         cfs_list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 344
 345         spin_unlock(&sfw_data.fw_lock);
 346 }
 347
 348 sfw_batch_t *
 349 sfw_find_batch (lst_bid_t bid)
 350 {
 351         sfw_session_t *sn = sfw_data.fw_session;
 352         sfw_batch_t   *bat;
 353
 354         LASSERT (sn != NULL);
 355
 356         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 357                                        sfw_batch_t, bat_list) {
 358                 if (bat->bat_id.bat_id == bid.bat_id)
 359                         return bat;
 360         }
 361
 362         return NULL;
 363 }
 364
 365 sfw_batch_t *
 366 sfw_bid2batch (lst_bid_t bid)
 367 {
 368         sfw_session_t *sn = sfw_data.fw_session;
 369         sfw_batch_t   *bat;
 370
 371         LASSERT (sn != NULL);
 372
 373         bat = sfw_find_batch(bid);
 374         if (bat != NULL)
 375                 return bat;
 376
 377         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 378         if (bat == NULL)
 379                 return NULL;
 380
 381         bat->bat_error    = 0;
 382         bat->bat_session  = sn;
 383         bat->bat_id       = bid;
 384         cfs_atomic_set(&bat->bat_nactive, 0);
 385         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 386
 387         cfs_list_add_tail(&bat->bat_list, &sn->sn_batches);
 388         return bat;
 389 }
 390
 391 int
 392 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 393 {
 394         sfw_session_t  *sn = sfw_data.fw_session;
 395         sfw_counters_t *cnt = &reply->str_fw;
 396         sfw_batch_t    *bat;
 397         struct timeval  tv;
 398
 399         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 400
 401         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 402                 reply->str_status = EINVAL;
 403                 return 0;
 404         }
 405
 406         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 407                 reply->str_status = ESRCH;
 408                 return 0;
 409         }
 410
 411         lnet_counters_get(&reply->str_lnet);
 412         srpc_get_counters(&reply->str_rpc);
 413
 414         /* send over the msecs since the session was started
 415          - with 32 bits to send, this is ~49 days */
 416         cfs_duration_usec(cfs_time_sub(cfs_time_current(),
 417                                        sn->sn_started), &tv);
 418
 419         cnt->running_ms      = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
 420         cnt->brw_errors      = cfs_atomic_read(&sn->sn_brw_errors);
 421         cnt->ping_errors     = cfs_atomic_read(&sn->sn_ping_errors);
 422         cnt->zombie_sessions = cfs_atomic_read(&sfw_data.fw_nzombies);
 423
 424         cnt->active_batches = 0;
 425         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 426                                        sfw_batch_t, bat_list) {
 427                 if (cfs_atomic_read(&bat->bat_nactive) > 0)
 428                         cnt->active_batches++;
 429         }
 430
 431         reply->str_status = 0;
 432         return 0;
 433 }
 434
 435 int
 436 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 437 {
 438         sfw_session_t *sn = sfw_data.fw_session;
 439         srpc_msg_t    *msg = container_of(request, srpc_msg_t,
 440                                           msg_body.mksn_reqst);
 441
 442         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 443                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 444                 reply->mksn_status = EINVAL;
 445                 return 0;
 446         }
 447
 448         if (sn != NULL) {
 449                 reply->mksn_status  = 0;
 450                 reply->mksn_sid     = sn->sn_id;
 451                 reply->mksn_timeout = sn->sn_timeout;
 452
 453                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 454                         cfs_atomic_inc(&sn->sn_refcount);
 455                         return 0;
 456                 }
 457
 458                 if (!request->mksn_force) {
 459                         reply->mksn_status = EBUSY;
 460                         strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 461                         return 0;
 462                 }
 463         }
 464
 465         /* reject the request if it requires unknown features
 466          * NB: old version will always accept all features because it's not
 467          * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
 468          * harmless because it will return zero feature to console, and it's
 469          * console's responsibility to make sure all nodes in a session have
 470          * same feature mask. */
 471         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 472                 reply->mksn_status = EPROTO;
 473                 return 0;
 474         }
 475
 476         /* brand new or create by force */
 477         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 478         if (sn == NULL) {
 479                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 480                 return -ENOMEM;
 481         }
 482
 483         sfw_init_session(sn, request->mksn_sid,
 484                          msg->msg_ses_feats, &request->mksn_name[0]);
 485
 486         spin_lock(&sfw_data.fw_lock);
 487
 488         sfw_deactivate_session();
 489         LASSERT(sfw_data.fw_session == NULL);
 490         sfw_data.fw_session = sn;
 491
 492         spin_unlock(&sfw_data.fw_lock);
 493
 494         reply->mksn_status  = 0;
 495         reply->mksn_sid     = sn->sn_id;
 496         reply->mksn_timeout = sn->sn_timeout;
 497         return 0;
 498 }
 499
 500 int
 501 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 502 {
 503         sfw_session_t *sn = sfw_data.fw_session;
 504
 505         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 506
 507         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 508                 reply->rmsn_status = EINVAL;
 509                 return 0;
 510         }
 511
 512         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 513                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 514                 return 0;
 515         }
 516
 517         if (!cfs_atomic_dec_and_test(&sn->sn_refcount)) {
 518                 reply->rmsn_status = 0;
 519                 return 0;
 520         }
 521
 522         spin_lock(&sfw_data.fw_lock);
 523         sfw_deactivate_session();
 524         spin_unlock(&sfw_data.fw_lock);
 525
 526         reply->rmsn_status = 0;
 527         reply->rmsn_sid    = LST_INVALID_SID;
 528         LASSERT(sfw_data.fw_session == NULL);
 529         return 0;
 530 }
 531
 532 int
 533 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 534 {
 535         sfw_session_t *sn = sfw_data.fw_session;
 536
 537         if (sn == NULL) {
 538                 reply->dbg_status = ESRCH;
 539                 reply->dbg_sid    = LST_INVALID_SID;
 540                 return 0;
 541         }
 542
 543         reply->dbg_status  = 0;
 544         reply->dbg_sid     = sn->sn_id;
 545         reply->dbg_timeout = sn->sn_timeout;
 546         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 547
 548         return 0;
 549 }
 550
 551 void
 552 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 553 {
 554         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 555         sfw_test_instance_t *tsi = tsu->tsu_instance;
 556
 557         /* Called with hold of tsi->tsi_lock */
 558         LASSERT (cfs_list_empty(&rpc->crpc_list));
 559         cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 560 }
 561
 562 static inline int
 563 sfw_test_buffers(sfw_test_instance_t *tsi)
 564 {
 565         struct sfw_test_case    *tsc = sfw_find_test_case(tsi->tsi_service);
 566         struct srpc_service     *svc = tsc->tsc_srv_service;
 567         int                     nbuf;
 568
 569         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 570         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 571 }
 572
 573 int
 574 sfw_load_test(struct sfw_test_instance *tsi)
 575 {
 576         struct sfw_test_case    *tsc = sfw_find_test_case(tsi->tsi_service);
 577         struct srpc_service     *svc = tsc->tsc_srv_service;
 578         int                     nbuf = sfw_test_buffers(tsi);
 579         int                     rc;
 580
 581         LASSERT(tsc != NULL);
 582
 583         if (tsi->tsi_is_client) {
 584                 tsi->tsi_ops = tsc->tsc_cli_ops;
 585                 return 0;
 586         }
 587
 588         rc = srpc_service_add_buffers(svc, nbuf);
 589         if (rc != 0) {
 590                 CWARN("Failed to reserve enough buffers: "
 591                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 592                 /* NB: this error handler is not strictly correct, because
 593                  * it may release more buffers than already allocated,
 594                  * but it doesn't matter because request portal should
 595                  * be lazy portal and will grow buffers if necessary. */
 596                 srpc_service_remove_buffers(svc, nbuf);
 597                 return -ENOMEM;
 598         }
 599
 600         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 601                nbuf * (srpc_serv_is_framework(svc) ?
 602                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 603         return 0;
 604 }
 605
 606 void
 607 sfw_unload_test(struct sfw_test_instance *tsi)
 608 {
 609         struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
 610
 611         LASSERT(tsc != NULL);
 612
 613         if (tsi->tsi_is_client)
 614                 return;
 615
 616         /* shrink buffers, because request portal is lazy portal
 617          * which can grow buffers at runtime so we may leave
 618          * some buffers behind, but never mind... */
 619         srpc_service_remove_buffers(tsc->tsc_srv_service,
 620                                     sfw_test_buffers(tsi));
 621         return;
 622 }
 623
 624 void
 625 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 626 {
 627         srpc_client_rpc_t *rpc;
 628         sfw_test_unit_t   *tsu;
 629
 630         if (!tsi->tsi_is_client) goto clean;
 631
 632         tsi->tsi_ops->tso_fini(tsi);
 633
 634         LASSERT (!tsi->tsi_stopping);
 635         LASSERT (cfs_list_empty(&tsi->tsi_active_rpcs));
 636         LASSERT (!sfw_test_active(tsi));
 637
 638         while (!cfs_list_empty(&tsi->tsi_units)) {
 639                 tsu = cfs_list_entry(tsi->tsi_units.next,
 640                                      sfw_test_unit_t, tsu_list);
 641                 cfs_list_del(&tsu->tsu_list);
 642                 LIBCFS_FREE(tsu, sizeof(*tsu));
 643         }
 644
 645         while (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 646                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 647                                      srpc_client_rpc_t, crpc_list);
 648                 cfs_list_del(&rpc->crpc_list);
 649                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 650         }
 651
 652 clean:
 653         sfw_unload_test(tsi);
 654         LIBCFS_FREE(tsi, sizeof(*tsi));
 655         return;
 656 }
 657
 658 void
 659 sfw_destroy_batch (sfw_batch_t *tsb)
 660 {
 661         sfw_test_instance_t *tsi;
 662
 663         LASSERT (!sfw_batch_active(tsb));
 664         LASSERT (cfs_list_empty(&tsb->bat_list));
 665
 666         while (!cfs_list_empty(&tsb->bat_tests)) {
 667                 tsi = cfs_list_entry(tsb->bat_tests.next,
 668                                      sfw_test_instance_t, tsi_list);
 669                 cfs_list_del_init(&tsi->tsi_list);
 670                 sfw_destroy_test_instance(tsi);
 671         }
 672
 673         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 674         return;
 675 }
 676
 677 void
 678 sfw_destroy_session (sfw_session_t *sn)
 679 {
 680         sfw_batch_t *batch;
 681
 682         LASSERT (cfs_list_empty(&sn->sn_list));
 683         LASSERT (sn != sfw_data.fw_session);
 684
 685         while (!cfs_list_empty(&sn->sn_batches)) {
 686                 batch = cfs_list_entry(sn->sn_batches.next,
 687                                        sfw_batch_t, bat_list);
 688                 cfs_list_del_init(&batch->bat_list);
 689                 sfw_destroy_batch(batch);
 690         }
 691
 692         LIBCFS_FREE(sn, sizeof(*sn));
 693         cfs_atomic_dec(&sfw_data.fw_nzombies);
 694         return;
 695 }
 696
 697 void
 698 sfw_unpack_addtest_req(srpc_msg_t *msg)
 699 {
 700         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 701
 702         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 703         LASSERT (req->tsr_is_client);
 704
 705         if (msg->msg_magic == SRPC_MSG_MAGIC)
 706                 return; /* no flipping needed */
 707
 708         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 709
 710         if (req->tsr_service == SRPC_SERVICE_BRW) {
 711                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 712                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 713
 714                         __swab32s(&bulk->blk_opc);
 715                         __swab32s(&bulk->blk_npg);
 716                         __swab32s(&bulk->blk_flags);
 717
 718                 } else {
 719                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 720
 721                         __swab16s(&bulk->blk_opc);
 722                         __swab16s(&bulk->blk_flags);
 723                         __swab32s(&bulk->blk_offset);
 724                         __swab32s(&bulk->blk_len);
 725                 }
 726
 727                 return;
 728         }
 729
 730         if (req->tsr_service == SRPC_SERVICE_PING) {
 731                 test_ping_req_t *ping = &req->tsr_u.ping;
 732
 733                 __swab32s(&ping->png_size);
 734                 __swab32s(&ping->png_flags);
 735                 return;
 736         }
 737
 738         LBUG ();
 739         return;
 740 }
 741
 742 int
 743 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 744 {
 745         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 746         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 747         srpc_bulk_t         *bk = rpc->srpc_bulk;
 748         int                  ndest = req->tsr_ndest;
 749         sfw_test_unit_t     *tsu;
 750         sfw_test_instance_t *tsi;
 751         int                  i;
 752         int                  rc;
 753
 754         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 755         if (tsi == NULL) {
 756                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 757                         tsb->bat_id.bat_id);
 758                 return -ENOMEM;
 759         }
 760
 761         memset(tsi, 0, sizeof(*tsi));
 762         spin_lock_init(&tsi->tsi_lock);
 763         cfs_atomic_set(&tsi->tsi_nactive, 0);
 764         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 765         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 766         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 767
 768         tsi->tsi_stopping      = 0;
 769         tsi->tsi_batch         = tsb;
 770         tsi->tsi_loop          = req->tsr_loop;
 771         tsi->tsi_concur        = req->tsr_concur;
 772         tsi->tsi_service       = req->tsr_service;
 773         tsi->tsi_is_client     = !!(req->tsr_is_client);
 774         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 775
 776         rc = sfw_load_test(tsi);
 777         if (rc != 0) {
 778                 LIBCFS_FREE(tsi, sizeof(*tsi));
 779                 return rc;
 780         }
 781
 782         LASSERT (!sfw_batch_active(tsb));
 783
 784         if (!tsi->tsi_is_client) {
 785                 /* it's test server, just add it to tsb */
 786                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 787                 return 0;
 788         }
 789
 790         LASSERT (bk != NULL);
 791 #ifndef __KERNEL__
 792         LASSERT (bk->bk_pages != NULL);
 793 #endif
 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 #ifdef __KERNEL__
 807                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 808                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 809 #else
 810                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 811 #endif
 812                 id = dests[i % SFW_ID_PER_PAGE];
 813                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 814                         sfw_unpack_id(id);
 815
 816                 for (j = 0; j < tsi->tsi_concur; j++) {
 817                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 818                         if (tsu == NULL) {
 819                                 rc = -ENOMEM;
 820                                 CERROR ("Can't allocate tsu for %d\n",
 821                                         tsi->tsi_service);
 822                                 goto error;
 823                         }
 824
 825                         tsu->tsu_dest.nid = id.nid;
 826                         tsu->tsu_dest.pid = id.pid;
 827                         tsu->tsu_instance = tsi;
 828                         tsu->tsu_private  = NULL;
 829                         cfs_list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 830                 }
 831         }
 832
 833         rc = tsi->tsi_ops->tso_init(tsi);
 834         if (rc == 0) {
 835                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 836                 return 0;
 837         }
 838
 839 error:
 840         LASSERT (rc != 0);
 841         sfw_destroy_test_instance(tsi);
 842         return rc;
 843 }
 844
 845 static void
 846 sfw_test_unit_done (sfw_test_unit_t *tsu)
 847 {
 848         sfw_test_instance_t *tsi = tsu->tsu_instance;
 849         sfw_batch_t         *tsb = tsi->tsi_batch;
 850         sfw_session_t       *sn = tsb->bat_session;
 851
 852         LASSERT (sfw_test_active(tsi));
 853
 854         if (!cfs_atomic_dec_and_test(&tsi->tsi_nactive))
 855                 return;
 856
 857         /* the test instance is done */
 858         spin_lock(&tsi->tsi_lock);
 859
 860         tsi->tsi_stopping = 0;
 861
 862         spin_unlock(&tsi->tsi_lock);
 863
 864         spin_lock(&sfw_data.fw_lock);
 865
 866         if (!cfs_atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 867             sn == sfw_data.fw_session) {                  /* sn also active */
 868                 spin_unlock(&sfw_data.fw_lock);
 869                 return;
 870         }
 871
 872         LASSERT (!cfs_list_empty(&sn->sn_list)); /* I'm a zombie! */
 873
 874         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 875                                        sfw_batch_t, bat_list) {
 876                 if (sfw_batch_active(tsb)) {
 877                         spin_unlock(&sfw_data.fw_lock);
 878                         return;
 879                 }
 880         }
 881
 882         cfs_list_del_init(&sn->sn_list);
 883         spin_unlock(&sfw_data.fw_lock);
 884
 885         sfw_destroy_session(sn);
 886         return;
 887 }
 888
 889 void
 890 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 891 {
 892         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 893         sfw_test_instance_t *tsi = tsu->tsu_instance;
 894         int                  done = 0;
 895
 896         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 897
 898         spin_lock(&tsi->tsi_lock);
 899
 900         LASSERT (sfw_test_active(tsi));
 901         LASSERT (!cfs_list_empty(&rpc->crpc_list));
 902
 903         cfs_list_del_init(&rpc->crpc_list);
 904
 905         /* batch is stopping or loop is done or get error */
 906         if (tsi->tsi_stopping ||
 907             tsu->tsu_loop == 0 ||
 908             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 909                 done = 1;
 910
 911         /* dec ref for poster */
 912         srpc_client_rpc_decref(rpc);
 913
 914         spin_unlock(&tsi->tsi_lock);
 915
 916         if (!done) {
 917                 swi_schedule_workitem(&tsu->tsu_worker);
 918                 return;
 919         }
 920
 921         sfw_test_unit_done(tsu);
 922         return;
 923 }
 924
 925 int
 926 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
 927                     unsigned features, int nblk, int blklen,
 928                     srpc_client_rpc_t **rpcpp)
 929 {
 930         srpc_client_rpc_t   *rpc = NULL;
 931         sfw_test_instance_t *tsi = tsu->tsu_instance;
 932
 933         spin_lock(&tsi->tsi_lock);
 934
 935         LASSERT (sfw_test_active(tsi));
 936
 937         if (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 938                 /* pick request from buffer */
 939                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 940                                      srpc_client_rpc_t, crpc_list);
 941                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 942                 cfs_list_del_init(&rpc->crpc_list);
 943         }
 944
 945         spin_unlock(&tsi->tsi_lock);
 946
 947         if (rpc == NULL) {
 948                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 949                                              blklen, sfw_test_rpc_done,
 950                                              sfw_test_rpc_fini, tsu);
 951         } else {
 952                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 953                                      blklen, sfw_test_rpc_done,
 954                                      sfw_test_rpc_fini, tsu);
 955         }
 956
 957         if (rpc == NULL) {
 958                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 959                 return -ENOMEM;
 960         }
 961
 962         rpc->crpc_reqstmsg.msg_ses_feats = features;
 963         *rpcpp = rpc;
 964
 965         return 0;
 966 }
 967
 968 int
 969 sfw_run_test (swi_workitem_t *wi)
 970 {
 971         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 972         sfw_test_instance_t *tsi = tsu->tsu_instance;
 973         srpc_client_rpc_t   *rpc = NULL;
 974
 975         LASSERT (wi == &tsu->tsu_worker);
 976
 977         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 978                 LASSERT (rpc == NULL);
 979                 goto test_done;
 980         }
 981
 982         LASSERT (rpc != NULL);
 983
 984         spin_lock(&tsi->tsi_lock);
 985
 986         if (tsi->tsi_stopping) {
 987                 cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 988                 spin_unlock(&tsi->tsi_lock);
 989                 goto test_done;
 990         }
 991
 992         if (tsu->tsu_loop > 0)
 993                 tsu->tsu_loop--;
 994
 995         cfs_list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 996         spin_unlock(&tsi->tsi_lock);
 997
 998         rpc->crpc_timeout = rpc_timeout;
 999
1000         spin_lock(&rpc->crpc_lock);
1001         srpc_post_rpc(rpc);
1002         spin_unlock(&rpc->crpc_lock);
1003         return 0;
1004
1005 test_done:
1006         /*
1007          * No one can schedule me now since:
1008          * - previous RPC, if any, has done and
1009          * - no new RPC is initiated.
1010          * - my batch is still active; no one can run it again now.
1011          * Cancel pending schedules and prevent future schedule attempts:
1012          */
1013         swi_exit_workitem(wi);
1014         sfw_test_unit_done(tsu);
1015         return 1;
1016 }
1017
1018 int
1019 sfw_run_batch (sfw_batch_t *tsb)
1020 {
1021         swi_workitem_t      *wi;
1022         sfw_test_unit_t     *tsu;
1023         sfw_test_instance_t *tsi;
1024
1025         if (sfw_batch_active(tsb)) {
1026                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
1027                        tsb->bat_id.bat_id, cfs_atomic_read(&tsb->bat_nactive));
1028                 return 0;
1029         }
1030
1031         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1032                                        sfw_test_instance_t, tsi_list) {
1033                 if (!tsi->tsi_is_client) /* skip server instances */
1034                         continue;
1035
1036                 LASSERT (!tsi->tsi_stopping);
1037                 LASSERT (!sfw_test_active(tsi));
1038
1039                 cfs_atomic_inc(&tsb->bat_nactive);
1040
1041                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
1042                                                sfw_test_unit_t, tsu_list) {
1043                         cfs_atomic_inc(&tsi->tsi_nactive);
1044                         tsu->tsu_loop = tsi->tsi_loop;
1045                         wi = &tsu->tsu_worker;
1046                         swi_init_workitem(wi, tsu, sfw_run_test,
1047                                           lst_sched_test[\
1048                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1049                         swi_schedule_workitem(wi);
1050                 }
1051         }
1052
1053         return 0;
1054 }
1055
1056 int
1057 sfw_stop_batch (sfw_batch_t *tsb, int force)
1058 {
1059         sfw_test_instance_t *tsi;
1060         srpc_client_rpc_t   *rpc;
1061
1062         if (!sfw_batch_active(tsb)) {
1063                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1064                 return 0;
1065         }
1066
1067         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1068                                        sfw_test_instance_t, tsi_list) {
1069                 spin_lock(&tsi->tsi_lock);
1070
1071                 if (!tsi->tsi_is_client ||
1072                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1073                         spin_unlock(&tsi->tsi_lock);
1074                         continue;
1075                 }
1076
1077                 tsi->tsi_stopping = 1;
1078
1079                 if (!force) {
1080                         spin_unlock(&tsi->tsi_lock);
1081                         continue;
1082                 }
1083
1084                 /* abort launched rpcs in the test */
1085                 cfs_list_for_each_entry_typed(rpc, &tsi->tsi_active_rpcs,
1086                                               srpc_client_rpc_t, crpc_list) {
1087                         spin_lock(&rpc->crpc_lock);
1088
1089                         srpc_abort_rpc(rpc, -EINTR);
1090
1091                         spin_unlock(&rpc->crpc_lock);
1092                 }
1093
1094                 spin_unlock(&tsi->tsi_lock);
1095         }
1096
1097         return 0;
1098 }
1099
1100 int
1101 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1102 {
1103         sfw_test_instance_t *tsi;
1104
1105         if (testidx < 0)
1106                 return -EINVAL;
1107
1108         if (testidx == 0) {
1109                 reply->bar_active = cfs_atomic_read(&tsb->bat_nactive);
1110                 return 0;
1111         }
1112
1113         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1114                                        sfw_test_instance_t, tsi_list) {
1115                 if (testidx-- > 1)
1116                         continue;
1117
1118                 reply->bar_active = cfs_atomic_read(&tsi->tsi_nactive);
1119                 return 0;
1120         }
1121
1122         return -ENOENT;
1123 }
1124
1125 void
1126 sfw_free_pages (srpc_server_rpc_t *rpc)
1127 {
1128         srpc_free_bulk(rpc->srpc_bulk);
1129         rpc->srpc_bulk = NULL;
1130 }
1131
1132 int
1133 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1134                 int sink)
1135 {
1136         LASSERT(rpc->srpc_bulk == NULL);
1137         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1138
1139         rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1140         if (rpc->srpc_bulk == NULL)
1141                 return -ENOMEM;
1142
1143         return 0;
1144 }
1145
1146 int
1147 sfw_add_test (srpc_server_rpc_t *rpc)
1148 {
1149         sfw_session_t     *sn = sfw_data.fw_session;
1150         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1151         srpc_test_reqst_t *request;
1152         int                rc;
1153         sfw_batch_t       *bat;
1154
1155         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1156         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1157
1158         if (request->tsr_loop == 0 ||
1159             request->tsr_concur == 0 ||
1160             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1161             request->tsr_ndest > SFW_MAX_NDESTS ||
1162             (request->tsr_is_client && request->tsr_ndest == 0) ||
1163             request->tsr_concur > SFW_MAX_CONCUR ||
1164             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1165             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1166                 reply->tsr_status = EINVAL;
1167                 return 0;
1168         }
1169
1170         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1171             sfw_find_test_case(request->tsr_service) == NULL) {
1172                 reply->tsr_status = ENOENT;
1173                 return 0;
1174         }
1175
1176         bat = sfw_bid2batch(request->tsr_bid);
1177         if (bat == NULL) {
1178                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1179                         rpc->srpc_scd->scd_svc->sv_name,
1180                         libcfs_id2str(rpc->srpc_peer));
1181                 return -ENOMEM;
1182         }
1183
1184         if (sfw_batch_active(bat)) {
1185                 reply->tsr_status = EBUSY;
1186                 return 0;
1187         }
1188
1189         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1190                 /* rpc will be resumed later in sfw_bulk_ready */
1191                 int     npg = sfw_id_pages(request->tsr_ndest);
1192                 int     len;
1193
1194                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1195                         len = npg * CFS_PAGE_SIZE;
1196
1197                 } else  {
1198                         len = sizeof(lnet_process_id_packed_t) *
1199                               request->tsr_ndest;
1200                 }
1201
1202                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1203         }
1204
1205         rc = sfw_add_test_instance(bat, rpc);
1206         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1207                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1208                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1209                 request->tsr_is_client ? "client" : "server",
1210                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1211
1212         reply->tsr_status = (rc < 0) ? -rc : rc;
1213         return 0;
1214 }
1215
1216 int
1217 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1218 {
1219         sfw_session_t *sn = sfw_data.fw_session;
1220         int            rc = 0;
1221         sfw_batch_t   *bat;
1222
1223         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1224
1225         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1226                 reply->bar_status = ESRCH;
1227                 return 0;
1228         }
1229
1230         bat = sfw_find_batch(request->bar_bid);
1231         if (bat == NULL) {
1232                 reply->bar_status = ENOENT;
1233                 return 0;
1234         }
1235
1236         switch (request->bar_opc) {
1237         case SRPC_BATCH_OPC_RUN:
1238                 rc = sfw_run_batch(bat);
1239                 break;
1240
1241         case SRPC_BATCH_OPC_STOP:
1242                 rc = sfw_stop_batch(bat, request->bar_arg);
1243                 break;
1244
1245         case SRPC_BATCH_OPC_QUERY:
1246                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1247                 break;
1248
1249         default:
1250                 return -EINVAL; /* drop it */
1251         }
1252
1253         reply->bar_status = (rc < 0) ? -rc : rc;
1254         return 0;
1255 }
1256
1257 int
1258 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1259 {
1260         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1261         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1262         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1263         unsigned        features = LST_FEATS_MASK;
1264         int             rc = 0;
1265
1266         LASSERT(sfw_data.fw_active_srpc == NULL);
1267         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1268
1269         spin_lock(&sfw_data.fw_lock);
1270
1271         if (sfw_data.fw_shuttingdown) {
1272                 spin_unlock(&sfw_data.fw_lock);
1273                 return -ESHUTDOWN;
1274         }
1275
1276         /* Remove timer to avoid racing with it or expiring active session */
1277         if (sfw_del_session_timer() != 0) {
1278                 CERROR("Dropping RPC (%s) from %s: racing with expiry timer.",
1279                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1280                 spin_unlock(&sfw_data.fw_lock);
1281                 return -EAGAIN;
1282         }
1283
1284         sfw_data.fw_active_srpc = rpc;
1285         spin_unlock(&sfw_data.fw_lock);
1286
1287         sfw_unpack_message(request);
1288         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1289
1290         /* rpc module should have checked this */
1291         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1292
1293         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1294             sv->sv_id != SRPC_SERVICE_DEBUG) {
1295                 sfw_session_t *sn = sfw_data.fw_session;
1296
1297                 if (sn != NULL &&
1298                     sn->sn_features != request->msg_ses_feats) {
1299                         CNETERR("Features of framework RPC don't match "
1300                                 "features of current session: %x/%x\n",
1301                                 request->msg_ses_feats, sn->sn_features);
1302                         reply->msg_body.reply.status = EPROTO;
1303                         reply->msg_body.reply.sid    = sn->sn_id;
1304                         goto out;
1305                 }
1306
1307         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1308                 /* NB: at this point, old version will ignore features and
1309                  * create new session anyway, so console should be able
1310                  * to handle this */
1311                 reply->msg_body.reply.status = EPROTO;
1312                 goto out;
1313         }
1314
1315         switch(sv->sv_id) {
1316         default:
1317                 LBUG ();
1318         case SRPC_SERVICE_TEST:
1319                 rc = sfw_add_test(rpc);
1320                 break;
1321
1322         case SRPC_SERVICE_BATCH:
1323                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1324                                        &reply->msg_body.bat_reply);
1325                 break;
1326
1327         case SRPC_SERVICE_QUERY_STAT:
1328                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1329                                    &reply->msg_body.stat_reply);
1330                 break;
1331
1332         case SRPC_SERVICE_DEBUG:
1333                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1334                                        &reply->msg_body.dbg_reply);
1335                 break;
1336
1337         case SRPC_SERVICE_MAKE_SESSION:
1338                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1339                                       &reply->msg_body.mksn_reply);
1340                 break;
1341
1342         case SRPC_SERVICE_REMOVE_SESSION:
1343                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1344                                         &reply->msg_body.rmsn_reply);
1345                 break;
1346         }
1347
1348         if (sfw_data.fw_session != NULL)
1349                 features = sfw_data.fw_session->sn_features;
1350  out:
1351         reply->msg_ses_feats = features;
1352         rpc->srpc_done = sfw_server_rpc_done;
1353         spin_lock(&sfw_data.fw_lock);
1354
1355 #ifdef __KERNEL__
1356         if (!sfw_data.fw_shuttingdown)
1357                 sfw_add_session_timer();
1358 #else
1359         LASSERT(!sfw_data.fw_shuttingdown);
1360         sfw_add_session_timer();
1361 #endif
1362
1363         sfw_data.fw_active_srpc = NULL;
1364         spin_unlock(&sfw_data.fw_lock);
1365         return rc;
1366 }
1367
1368 int
1369 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1370 {
1371         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1372         int                     rc;
1373
1374         LASSERT(rpc->srpc_bulk != NULL);
1375         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1376         LASSERT(sfw_data.fw_active_srpc == NULL);
1377         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1378
1379         spin_lock(&sfw_data.fw_lock);
1380
1381         if (status != 0) {
1382                 CERROR("Bulk transfer failed for RPC: "
1383                        "service %s, peer %s, status %d\n",
1384                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1385                 spin_unlock(&sfw_data.fw_lock);
1386                 return -EIO;
1387         }
1388
1389         if (sfw_data.fw_shuttingdown) {
1390                 spin_unlock(&sfw_data.fw_lock);
1391                 return -ESHUTDOWN;
1392         }
1393
1394         if (sfw_del_session_timer() != 0) {
1395                 CERROR("Dropping RPC (%s) from %s: racing with expiry timer",
1396                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1397                 spin_unlock(&sfw_data.fw_lock);
1398                 return -EAGAIN;
1399         }
1400
1401         sfw_data.fw_active_srpc = rpc;
1402         spin_unlock(&sfw_data.fw_lock);
1403
1404         rc = sfw_add_test(rpc);
1405
1406         spin_lock(&sfw_data.fw_lock);
1407
1408 #ifdef __KERNEL__
1409         if (!sfw_data.fw_shuttingdown)
1410                 sfw_add_session_timer();
1411 #else
1412         LASSERT(!sfw_data.fw_shuttingdown);
1413         sfw_add_session_timer();
1414 #endif
1415
1416         sfw_data.fw_active_srpc = NULL;
1417         spin_unlock(&sfw_data.fw_lock);
1418         return rc;
1419 }
1420
1421 srpc_client_rpc_t *
1422 sfw_create_rpc(lnet_process_id_t peer, int service,
1423                unsigned features, int nbulkiov, int bulklen,
1424                void (*done)(srpc_client_rpc_t *), void *priv)
1425 {
1426         srpc_client_rpc_t *rpc = NULL;
1427
1428         spin_lock(&sfw_data.fw_lock);
1429
1430         LASSERT (!sfw_data.fw_shuttingdown);
1431         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1432
1433         if (nbulkiov == 0 && !cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1434                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1435                                      srpc_client_rpc_t, crpc_list);
1436                 cfs_list_del(&rpc->crpc_list);
1437
1438                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1439                                      done, sfw_client_rpc_fini, priv);
1440         }
1441
1442         spin_unlock(&sfw_data.fw_lock);
1443
1444         if (rpc == NULL) {
1445                 rpc = srpc_create_client_rpc(peer, service,
1446                                              nbulkiov, bulklen, done,
1447                                              nbulkiov != 0 ?  NULL :
1448                                              sfw_client_rpc_fini,
1449                                              priv);
1450         }
1451
1452         if (rpc != NULL) /* "session" is concept in framework */
1453                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1454
1455         return rpc;
1456 }
1457
1458 void
1459 sfw_unpack_message (srpc_msg_t *msg)
1460 {
1461         if (msg->msg_magic == SRPC_MSG_MAGIC)
1462                 return; /* no flipping needed */
1463
1464         /* srpc module should guarantee I wouldn't get crap */
1465         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1466
1467         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1468                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1469
1470                 __swab32s(&req->str_type);
1471                 __swab64s(&req->str_rpyid);
1472                 sfw_unpack_sid(req->str_sid);
1473                 return;
1474         }
1475
1476         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1477                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1478
1479                 __swab32s(&rep->str_status);
1480                 sfw_unpack_sid(rep->str_sid);
1481                 sfw_unpack_fw_counters(rep->str_fw);
1482                 sfw_unpack_rpc_counters(rep->str_rpc);
1483                 sfw_unpack_lnet_counters(rep->str_lnet);
1484                 return;
1485         }
1486
1487         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1488                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1489
1490                 __swab64s(&req->mksn_rpyid);
1491                 __swab32s(&req->mksn_force);
1492                 sfw_unpack_sid(req->mksn_sid);
1493                 return;
1494         }
1495
1496         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1497                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1498
1499                 __swab32s(&rep->mksn_status);
1500                 __swab32s(&rep->mksn_timeout);
1501                 sfw_unpack_sid(rep->mksn_sid);
1502                 return;
1503         }
1504
1505         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1506                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1507
1508                 __swab64s(&req->rmsn_rpyid);
1509                 sfw_unpack_sid(req->rmsn_sid);
1510                 return;
1511         }
1512
1513         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1514                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1515
1516                 __swab32s(&rep->rmsn_status);
1517                 sfw_unpack_sid(rep->rmsn_sid);
1518                 return;
1519         }
1520
1521         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1522                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1523
1524                 __swab64s(&req->dbg_rpyid);
1525                 __swab32s(&req->dbg_flags);
1526                 sfw_unpack_sid(req->dbg_sid);
1527                 return;
1528         }
1529
1530         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1531                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1532
1533                 __swab32s(&rep->dbg_nbatch);
1534                 __swab32s(&rep->dbg_timeout);
1535                 sfw_unpack_sid(rep->dbg_sid);
1536                 return;
1537         }
1538
1539         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1540                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1541
1542                 __swab32s(&req->bar_opc);
1543                 __swab64s(&req->bar_rpyid);
1544                 __swab32s(&req->bar_testidx);
1545                 __swab32s(&req->bar_arg);
1546                 sfw_unpack_sid(req->bar_sid);
1547                 __swab64s(&req->bar_bid.bat_id);
1548                 return;
1549         }
1550
1551         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1552                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1553
1554                 __swab32s(&rep->bar_status);
1555                 sfw_unpack_sid(rep->bar_sid);
1556                 return;
1557         }
1558
1559         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1560                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1561
1562                 __swab64s(&req->tsr_rpyid);
1563                 __swab64s(&req->tsr_bulkid);
1564                 __swab32s(&req->tsr_loop);
1565                 __swab32s(&req->tsr_ndest);
1566                 __swab32s(&req->tsr_concur);
1567                 __swab32s(&req->tsr_service);
1568                 sfw_unpack_sid(req->tsr_sid);
1569                 __swab64s(&req->tsr_bid.bat_id);
1570                 return;
1571         }
1572
1573         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1574                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1575
1576                 __swab32s(&rep->tsr_status);
1577                 sfw_unpack_sid(rep->tsr_sid);
1578                 return;
1579         }
1580
1581         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1582                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1583
1584                 __swab64s(&req->join_rpyid);
1585                 sfw_unpack_sid(req->join_sid);
1586                 return;
1587         }
1588
1589         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1590                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1591
1592                 __swab32s(&rep->join_status);
1593                 __swab32s(&rep->join_timeout);
1594                 sfw_unpack_sid(rep->join_sid);
1595                 return;
1596         }
1597
1598         LBUG ();
1599         return;
1600 }
1601
1602 void
1603 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1604 {
1605         LASSERT(cfs_atomic_read(&rpc->crpc_refcount) > 0);
1606         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1607
1608         spin_lock(&rpc->crpc_lock);
1609         srpc_abort_rpc(rpc, -EINTR);
1610         spin_unlock(&rpc->crpc_lock);
1611         return;
1612 }
1613
1614 void
1615 sfw_post_rpc (srpc_client_rpc_t *rpc)
1616 {
1617         spin_lock(&rpc->crpc_lock);
1618
1619         LASSERT (!rpc->crpc_closed);
1620         LASSERT (!rpc->crpc_aborted);
1621         LASSERT (cfs_list_empty(&rpc->crpc_list));
1622         LASSERT (!sfw_data.fw_shuttingdown);
1623
1624         rpc->crpc_timeout = rpc_timeout;
1625         srpc_post_rpc(rpc);
1626
1627         spin_unlock(&rpc->crpc_lock);
1628         return;
1629 }
1630
1631 static srpc_service_t sfw_services[] =
1632 {
1633         {
1634                 /* sv_id */    SRPC_SERVICE_DEBUG,
1635                 /* sv_name */  "debug",
1636                 0
1637         },
1638         {
1639                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1640                 /* sv_name */  "query stats",
1641                 0
1642         },
1643         {
1644                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1645                 /* sv_name */  "make session",
1646                 0
1647         },
1648         {
1649                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1650                 /* sv_name */  "remove session",
1651                 0
1652         },
1653         {
1654                 /* sv_id */    SRPC_SERVICE_BATCH,
1655                 /* sv_name */  "batch service",
1656                 0
1657         },
1658         {
1659                 /* sv_id */    SRPC_SERVICE_TEST,
1660                 /* sv_name */  "test service",
1661                 0
1662         },
1663         {
1664                 /* sv_id */    0,
1665                 /* sv_name */  NULL,
1666                 0
1667         }
1668 };
1669
1670 extern sfw_test_client_ops_t ping_test_client;
1671 extern srpc_service_t        ping_test_service;
1672 extern void ping_init_test_client(void);
1673 extern void ping_init_test_service(void);
1674
1675 extern sfw_test_client_ops_t brw_test_client;
1676 extern srpc_service_t        brw_test_service;
1677 extern void brw_init_test_client(void);
1678 extern void brw_init_test_service(void);
1679
1680
1681 int
1682 sfw_startup (void)
1683 {
1684         int              i;
1685         int              rc;
1686         int              error;
1687         srpc_service_t  *sv;
1688         sfw_test_case_t *tsc;
1689
1690 #ifndef __KERNEL__
1691         char *s;
1692
1693         s = getenv("SESSION_TIMEOUT");
1694         session_timeout = s != NULL ? atoi(s) : session_timeout;
1695
1696         s = getenv("RPC_TIMEOUT");
1697         rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1698 #endif
1699
1700         if (session_timeout < 0) {
1701                 CERROR ("Session timeout must be non-negative: %d\n",
1702                         session_timeout);
1703                 return -EINVAL;
1704         }
1705
1706         if (rpc_timeout < 0) {
1707                 CERROR ("RPC timeout must be non-negative: %d\n",
1708                         rpc_timeout);
1709                 return -EINVAL;
1710         }
1711
1712         if (session_timeout == 0)
1713                 CWARN ("Zero session_timeout specified "
1714                        "- test sessions never expire.\n");
1715
1716         if (rpc_timeout == 0)
1717                 CWARN ("Zero rpc_timeout specified "
1718                        "- test RPC never expire.\n");
1719
1720         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1721
1722         sfw_data.fw_session     = NULL;
1723         sfw_data.fw_active_srpc = NULL;
1724         spin_lock_init(&sfw_data.fw_lock);
1725         cfs_atomic_set(&sfw_data.fw_nzombies, 0);
1726         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1727         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1728         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1729
1730         brw_init_test_client();
1731         brw_init_test_service();
1732         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1733         LASSERT (rc == 0);
1734
1735         ping_init_test_client();
1736         ping_init_test_service();
1737         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1738         LASSERT (rc == 0);
1739
1740         error = 0;
1741         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1742                                        sfw_test_case_t, tsc_list) {
1743                 sv = tsc->tsc_srv_service;
1744
1745                 rc = srpc_add_service(sv);
1746                 LASSERT (rc != -EBUSY);
1747                 if (rc != 0) {
1748                         CWARN ("Failed to add %s service: %d\n",
1749                                sv->sv_name, rc);
1750                         error = rc;
1751                 }
1752         }
1753
1754         for (i = 0; ; i++) {
1755                 sv = &sfw_services[i];
1756                 if (sv->sv_name == NULL) break;
1757
1758                 sv->sv_bulk_ready = NULL;
1759                 sv->sv_handler    = sfw_handle_server_rpc;
1760                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1761                 if (sv->sv_id == SRPC_SERVICE_TEST)
1762                         sv->sv_bulk_ready = sfw_bulk_ready;
1763
1764                 rc = srpc_add_service(sv);
1765                 LASSERT (rc != -EBUSY);
1766                 if (rc != 0) {
1767                         CWARN ("Failed to add %s service: %d\n",
1768                                sv->sv_name, rc);
1769                         error = rc;
1770                 }
1771
1772                 /* about to sfw_shutdown, no need to add buffer */
1773                 if (error) continue;
1774
1775                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1776                 if (rc != 0) {
1777                         CWARN("Failed to reserve enough buffers: "
1778                               "service %s, %d needed: %d\n",
1779                               sv->sv_name, sv->sv_wi_total, rc);
1780                         error = -ENOMEM;
1781                 }
1782         }
1783
1784         if (error != 0)
1785                 sfw_shutdown();
1786         return error;
1787 }
1788
1789 void
1790 sfw_shutdown (void)
1791 {
1792         srpc_service_t  *sv;
1793         sfw_test_case_t *tsc;
1794         int              i;
1795
1796         spin_lock(&sfw_data.fw_lock);
1797
1798         sfw_data.fw_shuttingdown = 1;
1799 #ifdef __KERNEL__
1800         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1801                        "waiting for active RPC to finish.\n");
1802 #else
1803         LASSERT (sfw_data.fw_active_srpc == NULL);
1804 #endif
1805
1806         if (sfw_del_session_timer() != 0)
1807                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1808                                "waiting for session timer to explode.\n");
1809
1810         sfw_deactivate_session();
1811         lst_wait_until(cfs_atomic_read(&sfw_data.fw_nzombies) == 0,
1812                        sfw_data.fw_lock,
1813                        "waiting for %d zombie sessions to die.\n",
1814                        cfs_atomic_read(&sfw_data.fw_nzombies));
1815
1816         spin_unlock(&sfw_data.fw_lock);
1817
1818         for (i = 0; ; i++) {
1819                 sv = &sfw_services[i];
1820                 if (sv->sv_name == NULL)
1821                         break;
1822
1823                 srpc_shutdown_service(sv);
1824                 srpc_remove_service(sv);
1825         }
1826
1827         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1828                                        sfw_test_case_t, tsc_list) {
1829                 sv = tsc->tsc_srv_service;
1830                 srpc_shutdown_service(sv);
1831                 srpc_remove_service(sv);
1832         }
1833
1834         while (!cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1835                 srpc_client_rpc_t *rpc;
1836
1837                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1838                                      srpc_client_rpc_t, crpc_list);
1839                 cfs_list_del(&rpc->crpc_list);
1840
1841                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1842         }
1843
1844         for (i = 0; ; i++) {
1845                 sv = &sfw_services[i];
1846                 if (sv->sv_name == NULL)
1847                         break;
1848
1849                 srpc_wait_service_shutdown(sv);
1850         }
1851
1852         while (!cfs_list_empty(&sfw_data.fw_tests)) {
1853                 tsc = cfs_list_entry(sfw_data.fw_tests.next,
1854                                      sfw_test_case_t, tsc_list);
1855
1856                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1857
1858                 cfs_list_del(&tsc->tsc_list);
1859                 LIBCFS_FREE(tsc, sizeof(*tsc));
1860         }
1861
1862         return;
1863 }