lnet/selftest/framework.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  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 int brw_inject_errors = 0;
  49 CFS_MODULE_PARM(brw_inject_errors, "i", int, 0644,
  50                 "# data errors to inject randomly, zero by default");
  51
  52 static int session_timeout = 100;
  53 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  54                 "test session timeout in seconds (100 by default, 0 == never)");
  55
  56 #define SFW_TEST_CONCURRENCY     128
  57 #define SFW_TEST_RPC_TIMEOUT     64
  58 #define SFW_CLIENT_RPC_TIMEOUT   64  /* in seconds */
  59 #define SFW_EXTRA_TEST_BUFFERS   8 /* tolerate buggy peers with extra buffers */
  60
  61 #define sfw_test_buffers(tsi)    ((tsi)->tsi_loop + SFW_EXTRA_TEST_BUFFERS)
  62
  63 #define sfw_unpack_id(id)               \
  64 do {                                    \
  65         __swab64s(&(id).nid);           \
  66         __swab32s(&(id).pid);           \
  67 } while (0)
  68
  69 #define sfw_unpack_sid(sid)             \
  70 do {                                    \
  71         __swab64s(&(sid).ses_nid);      \
  72         __swab64s(&(sid).ses_stamp);    \
  73 } while (0)
  74
  75 #define sfw_unpack_fw_counters(fc)        \
  76 do {                                      \
  77         __swab32s(&(fc).brw_errors);      \
  78         __swab32s(&(fc).ping_errors);     \
  79         __swab32s(&(fc).active_tests);    \
  80         __swab32s(&(fc).active_batches);  \
  81         __swab32s(&(fc).zombie_sessions); \
  82 } while (0)
  83
  84 #define sfw_unpack_rpc_counters(rc)     \
  85 do {                                    \
  86         __swab32s(&(rc).errors);        \
  87         __swab32s(&(rc).rpcs_sent);     \
  88         __swab32s(&(rc).rpcs_rcvd);     \
  89         __swab32s(&(rc).rpcs_dropped);  \
  90         __swab32s(&(rc).rpcs_expired);  \
  91         __swab64s(&(rc).bulk_get);      \
  92         __swab64s(&(rc).bulk_put);      \
  93 } while (0)
  94
  95 #define sfw_unpack_lnet_counters(lc)    \
  96 do {                                    \
  97         __swab32s(&(lc).errors);        \
  98         __swab32s(&(lc).msgs_max);      \
  99         __swab32s(&(lc).msgs_alloc);    \
 100         __swab32s(&(lc).send_count);    \
 101         __swab32s(&(lc).recv_count);    \
 102         __swab32s(&(lc).drop_count);    \
 103         __swab32s(&(lc).route_count);   \
 104         __swab64s(&(lc).send_length);   \
 105         __swab64s(&(lc).recv_length);   \
 106         __swab64s(&(lc).drop_length);   \
 107         __swab64s(&(lc).route_length);  \
 108 } while (0)
 109
 110 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 111 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 112
 113 struct smoketest_framework {
 114         struct list_head   fw_zombie_rpcs;     /* RPCs to be recycled */
 115         struct list_head   fw_zombie_sessions; /* stopping sessions */
 116         struct list_head   fw_tests;           /* registered test cases */
 117         atomic_t           fw_nzombies;        /* # zombie sessions */
 118         spinlock_t         fw_lock;            /* serialise */
 119         sfw_session_t     *fw_session;         /* _the_ session */
 120         int                fw_shuttingdown;    /* shutdown in progress */
 121         srpc_server_rpc_t *fw_active_srpc;     /* running RPC */
 122 } sfw_data;
 123
 124 /* forward ref's */
 125 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 126 void sfw_destroy_session (sfw_session_t *sn);
 127
 128 static inline sfw_test_case_t *
 129 sfw_find_test_case(int id)
 130 {
 131         sfw_test_case_t *tsc;
 132
 133         LASSERT (id <= SRPC_SERVICE_MAX_ID);
 134         LASSERT (id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 135
 136         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
 137                                        sfw_test_case_t, tsc_list) {
 138                 if (tsc->tsc_srv_service->sv_id == id)
 139                         return tsc;
 140         }
 141
 142         return NULL;
 143 }
 144
 145 static int
 146 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 147 {
 148         sfw_test_case_t *tsc;
 149
 150         if (sfw_find_test_case(service->sv_id) != NULL) {
 151                 CERROR ("Failed to register test %s (%d)\n",
 152                         service->sv_name, service->sv_id);
 153                 return -EEXIST;
 154         }
 155
 156         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 157         if (tsc == NULL)
 158                 return -ENOMEM;
 159
 160         memset(tsc, 0, sizeof(sfw_test_case_t));
 161         tsc->tsc_cli_ops     = cliops;
 162         tsc->tsc_srv_service = service;
 163
 164         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 165         return 0;
 166 }
 167
 168 void
 169 sfw_add_session_timer (void)
 170 {
 171         sfw_session_t *sn = sfw_data.fw_session;
 172         stt_timer_t   *timer = &sn->sn_timer;
 173
 174         LASSERT (!sfw_data.fw_shuttingdown);
 175
 176         if (sn == NULL || sn->sn_timeout == 0)
 177                 return;
 178
 179         LASSERT (!sn->sn_timer_active);
 180
 181         sn->sn_timer_active = 1;
 182         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 183                                           cfs_time_current_sec());
 184         stt_add_timer(timer);
 185         return;
 186 }
 187
 188 int
 189 sfw_del_session_timer (void)
 190 {
 191         sfw_session_t *sn = sfw_data.fw_session;
 192
 193         if (sn == NULL || !sn->sn_timer_active)
 194                 return 0;
 195
 196         LASSERT (sn->sn_timeout != 0);
 197
 198         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 199                 sn->sn_timer_active = 0;
 200                 return 0;
 201         }
 202
 203 #ifndef __KERNEL__
 204         /* Racing is impossible in single-threaded userland selftest */
 205         LBUG();
 206 #endif
 207         return EBUSY; /* racing with sfw_session_expired() */
 208 }
 209
 210 /* called with sfw_data.fw_lock held */
 211 static void
 212 sfw_deactivate_session (void)
 213 {
 214         sfw_session_t *sn = sfw_data.fw_session;
 215         int            nactive = 0;
 216         sfw_batch_t   *tsb;
 217
 218         if (sn == NULL) return;
 219
 220         LASSERT (!sn->sn_timer_active);
 221
 222         sfw_data.fw_session = NULL;
 223         atomic_inc(&sfw_data.fw_nzombies);
 224         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 225
 226         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 227                                        sfw_batch_t, bat_list) {
 228                 if (sfw_batch_active(tsb)) {
 229                         nactive++;
 230                         sfw_stop_batch(tsb, 1);
 231                 }
 232         }
 233
 234         if (nactive != 0)
 235                 return;   /* wait for active batches to stop */
 236
 237         list_del_init(&sn->sn_list);
 238         spin_unlock(&sfw_data.fw_lock);
 239
 240         sfw_destroy_session(sn);
 241
 242         spin_lock(&sfw_data.fw_lock);
 243         return;
 244 }
 245
 246 #ifndef __KERNEL__
 247
 248 int
 249 sfw_session_removed (void)
 250 {
 251         return (sfw_data.fw_session == NULL) ? 1 : 0;
 252 }
 253
 254 #endif
 255
 256 void
 257 sfw_session_expired (void *data)
 258 {
 259         sfw_session_t *sn = data;
 260
 261         spin_lock(&sfw_data.fw_lock);
 262
 263         LASSERT (sn->sn_timer_active);
 264         LASSERT (sn == sfw_data.fw_session);
 265
 266         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 267                libcfs_nid2str(sn->sn_id.ses_nid),
 268                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 269
 270         sn->sn_timer_active = 0;
 271         sfw_deactivate_session();
 272
 273         spin_unlock(&sfw_data.fw_lock);
 274         return;
 275 }
 276
 277 static inline void
 278 sfw_init_session (sfw_session_t *sn, lst_sid_t sid, const char *name)
 279 {
 280         stt_timer_t *timer = &sn->sn_timer;
 281
 282         memset(sn, 0, sizeof(sfw_session_t));
 283         CFS_INIT_LIST_HEAD(&sn->sn_list);
 284         CFS_INIT_LIST_HEAD(&sn->sn_batches);
 285         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 286         atomic_set(&sn->sn_brw_errors, 0);
 287         atomic_set(&sn->sn_ping_errors, 0);
 288         strncpy(&sn->sn_name[0], name, LST_NAME_SIZE);
 289
 290         sn->sn_timer_active = 0;
 291         sn->sn_id           = sid;
 292         sn->sn_timeout      = session_timeout;
 293
 294         timer->stt_data = sn;
 295         timer->stt_func = sfw_session_expired;
 296         CFS_INIT_LIST_HEAD(&timer->stt_list);
 297 }
 298
 299 /* completion handler for incoming framework RPCs */
 300 void
 301 sfw_server_rpc_done (srpc_server_rpc_t *rpc)
 302 {
 303         srpc_service_t *sv = rpc->srpc_service;
 304         int             status = rpc->srpc_status;
 305
 306         CDEBUG (D_NET,
 307                 "Incoming framework RPC done: "
 308                 "service %s, peer %s, status %s:%d\n",
 309                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 310                 swi_state2str(rpc->srpc_wi.wi_state),
 311                 status);
 312
 313         if (rpc->srpc_bulk != NULL)
 314                 sfw_free_pages(rpc);
 315         return;
 316 }
 317
 318 void
 319 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 320 {
 321         LASSERT (rpc->crpc_bulk.bk_niov == 0);
 322         LASSERT (list_empty(&rpc->crpc_list));
 323         LASSERT (atomic_read(&rpc->crpc_refcount) == 0);
 324 #ifndef __KERNEL__
 325         LASSERT (rpc->crpc_bulk.bk_pages == NULL);
 326 #endif
 327
 328         CDEBUG (D_NET,
 329                 "Outgoing framework RPC done: "
 330                 "service %d, peer %s, status %s:%d:%d\n",
 331                 rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 332                 swi_state2str(rpc->crpc_wi.wi_state),
 333                 rpc->crpc_aborted, rpc->crpc_status);
 334
 335         spin_lock(&sfw_data.fw_lock);
 336
 337         /* my callers must finish all RPCs before shutting me down */
 338         LASSERT (!sfw_data.fw_shuttingdown);
 339         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 340
 341         spin_unlock(&sfw_data.fw_lock);
 342         return;
 343 }
 344
 345 sfw_batch_t *
 346 sfw_find_batch (lst_bid_t bid)
 347 {
 348         sfw_session_t *sn = sfw_data.fw_session;
 349         sfw_batch_t   *bat;
 350
 351         LASSERT (sn != NULL);
 352
 353         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 354                                        sfw_batch_t, bat_list) {
 355                 if (bat->bat_id.bat_id == bid.bat_id)
 356                         return bat;
 357         }
 358
 359         return NULL;
 360 }
 361
 362 sfw_batch_t *
 363 sfw_bid2batch (lst_bid_t bid)
 364 {
 365         sfw_session_t *sn = sfw_data.fw_session;
 366         sfw_batch_t   *bat;
 367
 368         LASSERT (sn != NULL);
 369
 370         bat = sfw_find_batch(bid);
 371         if (bat != NULL)
 372                 return bat;
 373
 374         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 375         if (bat == NULL)
 376                 return NULL;
 377
 378         bat->bat_error    = 0;
 379         bat->bat_session  = sn;
 380         bat->bat_id       = bid;
 381         atomic_set(&bat->bat_nactive, 0);
 382         CFS_INIT_LIST_HEAD(&bat->bat_tests);
 383
 384         list_add_tail(&bat->bat_list, &sn->sn_batches);
 385         return bat;
 386 }
 387
 388 int
 389 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 390 {
 391         sfw_session_t  *sn = sfw_data.fw_session;
 392         sfw_counters_t *cnt = &reply->str_fw;
 393         sfw_batch_t    *bat;
 394
 395         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 396
 397         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 398                 reply->str_status = EINVAL;
 399                 return 0;
 400         }
 401
 402         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 403                 reply->str_status = ESRCH;
 404                 return 0;
 405         }
 406
 407         LNET_LOCK();
 408         reply->str_lnet = the_lnet.ln_counters;
 409         LNET_UNLOCK();
 410
 411         srpc_get_counters(&reply->str_rpc);
 412
 413         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 414         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 415         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 416
 417         cnt->active_tests = cnt->active_batches = 0;
 418         cfs_list_for_each_entry_typed (bat, &sn->sn_batches,
 419                                        sfw_batch_t, bat_list) {
 420                 int n = atomic_read(&bat->bat_nactive);
 421
 422                 if (n > 0) {
 423                         cnt->active_batches++;
 424                         cnt->active_tests += n;
 425                 }
 426         }
 427
 428         reply->str_status = 0;
 429         return 0;
 430 }
 431
 432 int
 433 sfw_make_session (srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 434 {
 435         sfw_session_t *sn = sfw_data.fw_session;
 436
 437         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 438                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 439                 reply->mksn_status = EINVAL;
 440                 return 0;
 441         }
 442
 443         if (sn != NULL) {
 444                 reply->mksn_status  = 0;
 445                 reply->mksn_sid     = sn->sn_id;
 446                 reply->mksn_timeout = sn->sn_timeout;
 447
 448                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 449                         atomic_inc(&sn->sn_refcount);
 450                         return 0;
 451                 }
 452
 453                 if (!request->mksn_force) {
 454                         reply->mksn_status = EBUSY;
 455                         strncpy(&reply->mksn_name[0], &sn->sn_name[0], LST_NAME_SIZE);
 456                         return 0;
 457                 }
 458         }
 459
 460         /* brand new or create by force */
 461         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 462         if (sn == NULL) {
 463                 CERROR ("Dropping RPC (mksn) under memory pressure.\n");
 464                 return -ENOMEM;
 465         }
 466
 467         sfw_init_session(sn, request->mksn_sid, &request->mksn_name[0]);
 468
 469         spin_lock(&sfw_data.fw_lock);
 470
 471         sfw_deactivate_session();
 472         LASSERT (sfw_data.fw_session == NULL);
 473         sfw_data.fw_session = sn;
 474
 475         spin_unlock(&sfw_data.fw_lock);
 476
 477         reply->mksn_status  = 0;
 478         reply->mksn_sid     = sn->sn_id;
 479         reply->mksn_timeout = sn->sn_timeout;
 480         return 0;
 481 }
 482
 483 int
 484 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 485 {
 486         sfw_session_t *sn = sfw_data.fw_session;
 487
 488         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 489
 490         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 491                 reply->rmsn_status = EINVAL;
 492                 return 0;
 493         }
 494
 495         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 496                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 497                 return 0;
 498         }
 499
 500         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 501                 reply->rmsn_status = 0;
 502                 return 0;
 503         }
 504
 505         spin_lock(&sfw_data.fw_lock);
 506         sfw_deactivate_session();
 507         spin_unlock(&sfw_data.fw_lock);
 508
 509         reply->rmsn_status = 0;
 510         reply->rmsn_sid    = LST_INVALID_SID;
 511         LASSERT (sfw_data.fw_session == NULL);
 512         return 0;
 513 }
 514
 515 int
 516 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 517 {
 518         sfw_session_t *sn = sfw_data.fw_session;
 519
 520         if (sn == NULL) {
 521                 reply->dbg_status = ESRCH;
 522                 reply->dbg_sid    = LST_INVALID_SID;
 523                 return 0;
 524         }
 525
 526         reply->dbg_status  = 0;
 527         reply->dbg_sid     = sn->sn_id;
 528         reply->dbg_timeout = sn->sn_timeout;
 529         strncpy(reply->dbg_name, &sn->sn_name[0], LST_NAME_SIZE);
 530
 531         return 0;
 532 }
 533
 534 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 int
 546 sfw_load_test (sfw_test_instance_t *tsi)
 547 {
 548         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 549         int              nrequired = sfw_test_buffers(tsi);
 550         int              nposted;
 551
 552         LASSERT (tsc != NULL);
 553
 554         if (tsi->tsi_is_client) {
 555                 tsi->tsi_ops = tsc->tsc_cli_ops;
 556                 return 0;
 557         }
 558
 559         nposted = srpc_service_add_buffers(tsc->tsc_srv_service, nrequired);
 560         if (nposted != nrequired) {
 561                 CWARN ("Failed to reserve enough buffers: "
 562                        "service %s, %d needed, %d reserved\n",
 563                        tsc->tsc_srv_service->sv_name, nrequired, nposted);
 564                 srpc_service_remove_buffers(tsc->tsc_srv_service, nposted);
 565                 return -ENOMEM;
 566         }
 567
 568         CDEBUG (D_NET, "Reserved %d buffers for test %s\n",
 569                 nposted, tsc->tsc_srv_service->sv_name);
 570         return 0;
 571 }
 572
 573 void
 574 sfw_unload_test (sfw_test_instance_t *tsi)
 575 {
 576         sfw_test_case_t *tsc = sfw_find_test_case(tsi->tsi_service);
 577
 578         LASSERT (tsc != NULL);
 579
 580         if (!tsi->tsi_is_client)
 581                 srpc_service_remove_buffers(tsc->tsc_srv_service,
 582                                             sfw_test_buffers(tsi));
 583         return;
 584 }
 585
 586 void
 587 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 588 {
 589         srpc_client_rpc_t *rpc;
 590         sfw_test_unit_t   *tsu;
 591
 592         if (!tsi->tsi_is_client) goto clean;
 593
 594         tsi->tsi_ops->tso_fini(tsi);
 595
 596         LASSERT (!tsi->tsi_stopping);
 597         LASSERT (list_empty(&tsi->tsi_active_rpcs));
 598         LASSERT (!sfw_test_active(tsi));
 599
 600         while (!list_empty(&tsi->tsi_units)) {
 601                 tsu = list_entry(tsi->tsi_units.next,
 602                                  sfw_test_unit_t, tsu_list);
 603                 list_del(&tsu->tsu_list);
 604                 LIBCFS_FREE(tsu, sizeof(*tsu));
 605         }
 606
 607         while (!list_empty(&tsi->tsi_free_rpcs)) {
 608                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 609                                  srpc_client_rpc_t, crpc_list);
 610                 list_del(&rpc->crpc_list);
 611                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 612         }
 613
 614 clean:
 615         sfw_unload_test(tsi);
 616         LIBCFS_FREE(tsi, sizeof(*tsi));
 617         return;
 618 }
 619
 620 void
 621 sfw_destroy_batch (sfw_batch_t *tsb)
 622 {
 623         sfw_test_instance_t *tsi;
 624
 625         LASSERT (!sfw_batch_active(tsb));
 626         LASSERT (list_empty(&tsb->bat_list));
 627
 628         while (!list_empty(&tsb->bat_tests)) {
 629                 tsi = list_entry(tsb->bat_tests.next,
 630                                  sfw_test_instance_t, tsi_list);
 631                 list_del_init(&tsi->tsi_list);
 632                 sfw_destroy_test_instance(tsi);
 633         }
 634
 635         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 636         return;
 637 }
 638
 639 void
 640 sfw_destroy_session (sfw_session_t *sn)
 641 {
 642         sfw_batch_t *batch;
 643
 644         LASSERT (list_empty(&sn->sn_list));
 645         LASSERT (sn != sfw_data.fw_session);
 646
 647         while (!list_empty(&sn->sn_batches)) {
 648                 batch = list_entry(sn->sn_batches.next,
 649                                    sfw_batch_t, bat_list);
 650                 list_del_init(&batch->bat_list);
 651                 sfw_destroy_batch(batch);
 652         }
 653
 654         LIBCFS_FREE(sn, sizeof(*sn));
 655         atomic_dec(&sfw_data.fw_nzombies);
 656         return;
 657 }
 658
 659 void
 660 sfw_unpack_test_req (srpc_msg_t *msg)
 661 {
 662         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 663
 664         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 665         LASSERT (req->tsr_is_client);
 666
 667         if (msg->msg_magic == SRPC_MSG_MAGIC)
 668                 return; /* no flipping needed */
 669
 670         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 671
 672         if (req->tsr_service == SRPC_SERVICE_BRW) {
 673                 test_bulk_req_t *bulk = &req->tsr_u.bulk;
 674
 675                 __swab32s(&bulk->blk_opc);
 676                 __swab32s(&bulk->blk_npg);
 677                 __swab32s(&bulk->blk_flags);
 678                 return;
 679         }
 680
 681         if (req->tsr_service == SRPC_SERVICE_PING) {
 682                 test_ping_req_t *ping = &req->tsr_u.ping;
 683
 684                 __swab32s(&ping->png_size);
 685                 __swab32s(&ping->png_flags);
 686                 return;
 687         }
 688
 689         LBUG ();
 690         return;
 691 }
 692
 693 int
 694 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 695 {
 696         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 697         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 698         srpc_bulk_t         *bk = rpc->srpc_bulk;
 699         int                  ndest = req->tsr_ndest;
 700         sfw_test_unit_t     *tsu;
 701         sfw_test_instance_t *tsi;
 702         int                  i;
 703         int                  rc;
 704
 705         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 706         if (tsi == NULL) {
 707                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 708                         tsb->bat_id.bat_id);
 709                 return -ENOMEM;
 710         }
 711
 712         memset(tsi, 0, sizeof(*tsi));
 713         spin_lock_init(&tsi->tsi_lock);
 714         atomic_set(&tsi->tsi_nactive, 0);
 715         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 716         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 717         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 718
 719         tsi->tsi_stopping      = 0;
 720         tsi->tsi_batch         = tsb;
 721         tsi->tsi_loop          = req->tsr_loop;
 722         tsi->tsi_concur        = req->tsr_concur;
 723         tsi->tsi_service       = req->tsr_service;
 724         tsi->tsi_is_client     = !!(req->tsr_is_client);
 725         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 726
 727         rc = sfw_load_test(tsi);
 728         if (rc != 0) {
 729                 LIBCFS_FREE(tsi, sizeof(*tsi));
 730                 return rc;
 731         }
 732
 733         LASSERT (!sfw_batch_active(tsb));
 734
 735         if (!tsi->tsi_is_client) {
 736                 /* it's test server, just add it to tsb */
 737                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 738                 return 0;
 739         }
 740
 741         LASSERT (bk != NULL);
 742 #ifndef __KERNEL__
 743         LASSERT (bk->bk_pages != NULL);
 744 #endif
 745         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 746         LASSERT ((unsigned int)bk->bk_len >= sizeof(lnet_process_id_t) * ndest);
 747
 748         sfw_unpack_test_req(msg);
 749         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 750
 751         for (i = 0; i < ndest; i++) {
 752                 lnet_process_id_t *dests;
 753                 lnet_process_id_t  id;
 754                 int                j;
 755
 756 #ifdef __KERNEL__
 757                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 758                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 759 #else
 760                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 761 #endif
 762                 id = dests[i % SFW_ID_PER_PAGE];
 763                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 764                         sfw_unpack_id(id);
 765
 766                 for (j = 0; j < tsi->tsi_concur; j++) {
 767                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 768                         if (tsu == NULL) {
 769                                 rc = -ENOMEM;
 770                                 CERROR ("Can't allocate tsu for %d\n",
 771                                         tsi->tsi_service);
 772                                 goto error;
 773                         }
 774
 775                         tsu->tsu_dest     = id;
 776                         tsu->tsu_instance = tsi;
 777                         tsu->tsu_private  = NULL;
 778                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 779                 }
 780         }
 781
 782         rc = tsi->tsi_ops->tso_init(tsi);
 783         if (rc == 0) {
 784                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 785                 return 0;
 786         }
 787
 788 error:
 789         LASSERT (rc != 0);
 790         sfw_destroy_test_instance(tsi);
 791         return rc;
 792 }
 793
 794 static void
 795 sfw_test_unit_done (sfw_test_unit_t *tsu)
 796 {
 797         sfw_test_instance_t *tsi = tsu->tsu_instance;
 798         sfw_batch_t         *tsb = tsi->tsi_batch;
 799         sfw_session_t       *sn = tsb->bat_session;
 800
 801         LASSERT (sfw_test_active(tsi));
 802
 803         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 804                 return;
 805
 806         /* the test instance is done */
 807         spin_lock(&tsi->tsi_lock);
 808
 809         tsi->tsi_stopping = 0;
 810
 811         spin_unlock(&tsi->tsi_lock);
 812
 813         spin_lock(&sfw_data.fw_lock);
 814
 815         if (!atomic_dec_and_test(&tsb->bat_nactive) || /* tsb still active */
 816             sn == sfw_data.fw_session) {               /* sn also active */
 817                 spin_unlock(&sfw_data.fw_lock);
 818                 return;
 819         }
 820
 821         LASSERT (!list_empty(&sn->sn_list)); /* I'm a zombie! */
 822
 823         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 824                                        sfw_batch_t, bat_list) {
 825                 if (sfw_batch_active(tsb)) {
 826                         spin_unlock(&sfw_data.fw_lock);
 827                         return;
 828                 }
 829         }
 830
 831         list_del_init(&sn->sn_list);
 832         spin_unlock(&sfw_data.fw_lock);
 833
 834         sfw_destroy_session(sn);
 835         return;
 836 }
 837
 838 void
 839 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 840 {
 841         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 842         sfw_test_instance_t *tsi = tsu->tsu_instance;
 843         int                  done = 0;
 844
 845         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 846
 847         spin_lock(&tsi->tsi_lock);
 848
 849         LASSERT (sfw_test_active(tsi));
 850         LASSERT (!list_empty(&rpc->crpc_list));
 851
 852         list_del_init(&rpc->crpc_list);
 853
 854         /* batch is stopping or loop is done or get error */
 855         if (tsi->tsi_stopping ||
 856             tsu->tsu_loop == 0 ||
 857             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 858                 done = 1;
 859
 860         /* dec ref for poster */
 861         srpc_client_rpc_decref(rpc);
 862
 863         spin_unlock(&tsi->tsi_lock);
 864
 865         if (!done) {
 866                 swi_schedule_workitem(&tsu->tsu_worker);
 867                 return;
 868         }
 869
 870         sfw_test_unit_done(tsu);
 871         return;
 872 }
 873
 874 int
 875 sfw_create_test_rpc (sfw_test_unit_t *tsu, lnet_process_id_t peer,
 876                      int nblk, int blklen, srpc_client_rpc_t **rpcpp)
 877 {
 878         srpc_client_rpc_t   *rpc = NULL;
 879         sfw_test_instance_t *tsi = tsu->tsu_instance;
 880
 881         spin_lock(&tsi->tsi_lock);
 882
 883         LASSERT (sfw_test_active(tsi));
 884
 885         if (!list_empty(&tsi->tsi_free_rpcs)) {
 886                 /* pick request from buffer */
 887                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 888                                  srpc_client_rpc_t, crpc_list);
 889                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 890                 list_del_init(&rpc->crpc_list);
 891
 892                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 893                                      blklen, sfw_test_rpc_done,
 894                                      sfw_test_rpc_fini, tsu);
 895         }
 896
 897         spin_unlock(&tsi->tsi_lock);
 898
 899         if (rpc == NULL)
 900                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 901                                              blklen, sfw_test_rpc_done,
 902                                              sfw_test_rpc_fini, tsu);
 903         if (rpc == NULL) {
 904                 CERROR ("Can't create rpc for test %d\n", tsi->tsi_service);
 905                 return -ENOMEM;
 906         }
 907
 908         *rpcpp = rpc;
 909         return 0;
 910 }
 911
 912 int
 913 sfw_run_test (swi_workitem_t *wi)
 914 {
 915         sfw_test_unit_t     *tsu = wi->wi_data;
 916         sfw_test_instance_t *tsi = tsu->tsu_instance;
 917         srpc_client_rpc_t   *rpc = NULL;
 918
 919         LASSERT (wi == &tsu->tsu_worker);
 920
 921         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 922                 LASSERT (rpc == NULL);
 923                 goto test_done;
 924         }
 925
 926         LASSERT (rpc != NULL);
 927
 928         spin_lock(&tsi->tsi_lock);
 929
 930         if (tsi->tsi_stopping) {
 931                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 932                 spin_unlock(&tsi->tsi_lock);
 933                 goto test_done;
 934         }
 935
 936         if (tsu->tsu_loop > 0)
 937                 tsu->tsu_loop--;
 938
 939         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 940         spin_unlock(&tsi->tsi_lock);
 941
 942         rpc->crpc_timeout = SFW_TEST_RPC_TIMEOUT;
 943
 944         spin_lock(&rpc->crpc_lock);
 945         srpc_post_rpc(rpc);
 946         spin_unlock(&rpc->crpc_lock);
 947         return 0;
 948
 949 test_done:
 950         /*
 951          * No one can schedule me now since:
 952          * - previous RPC, if any, has done and
 953          * - no new RPC is initiated.
 954          * - my batch is still active; no one can run it again now.
 955          * Cancel pending schedules and prevent future schedule attempts:
 956          */
 957         swi_kill_workitem(wi);
 958         sfw_test_unit_done(tsu);
 959         return 1;
 960 }
 961
 962 int
 963 sfw_run_batch (sfw_batch_t *tsb)
 964 {
 965         swi_workitem_t      *wi;
 966         sfw_test_unit_t     *tsu;
 967         sfw_test_instance_t *tsi;
 968
 969         if (sfw_batch_active(tsb)) {
 970                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
 971                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
 972                 return 0;
 973         }
 974
 975         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
 976                                        sfw_test_instance_t, tsi_list) {
 977                 if (!tsi->tsi_is_client) /* skip server instances */
 978                         continue;
 979
 980                 LASSERT (!tsi->tsi_stopping);
 981                 LASSERT (!sfw_test_active(tsi));
 982
 983                 atomic_inc(&tsb->bat_nactive);
 984
 985                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
 986                                                sfw_test_unit_t, tsu_list) {
 987                         atomic_inc(&tsi->tsi_nactive);
 988                         tsu->tsu_loop = tsi->tsi_loop;
 989                         wi = &tsu->tsu_worker;
 990                         swi_init_workitem(wi, tsu, sfw_run_test);
 991                         swi_schedule_workitem(wi);
 992                 }
 993         }
 994
 995         return 0;
 996 }
 997
 998 int
 999 sfw_stop_batch (sfw_batch_t *tsb, int force)
1000 {
1001         sfw_test_instance_t *tsi;
1002         srpc_client_rpc_t   *rpc;
1003
1004         if (!sfw_batch_active(tsb)) {
1005                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1006                 return 0;
1007         }
1008
1009         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1010                                        sfw_test_instance_t, tsi_list) {
1011                 spin_lock(&tsi->tsi_lock);
1012
1013                 if (!tsi->tsi_is_client ||
1014                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1015                         spin_unlock(&tsi->tsi_lock);
1016                         continue;
1017                 }
1018
1019                 tsi->tsi_stopping = 1;
1020
1021                 if (!force) {
1022                         spin_unlock(&tsi->tsi_lock);
1023                         continue;
1024                 }
1025
1026                 /* abort launched rpcs in the test */
1027                 cfs_list_for_each_entry_typed (rpc, &tsi->tsi_active_rpcs,
1028                                                srpc_client_rpc_t, crpc_list) {
1029                         spin_lock(&rpc->crpc_lock);
1030
1031                         srpc_abort_rpc(rpc, -EINTR);
1032
1033                         spin_unlock(&rpc->crpc_lock);
1034                 }
1035
1036                 spin_unlock(&tsi->tsi_lock);
1037         }
1038
1039         return 0;
1040 }
1041
1042 int
1043 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1044 {
1045         sfw_test_instance_t *tsi;
1046
1047         if (testidx < 0)
1048                 return -EINVAL;
1049
1050         if (testidx == 0) {
1051                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1052                 return 0;
1053         }
1054
1055         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1056                                        sfw_test_instance_t, tsi_list) {
1057                 if (testidx-- > 1)
1058                         continue;
1059
1060                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1061                 return 0;
1062         }
1063
1064         return -ENOENT;
1065 }
1066
1067 void
1068 sfw_free_pages (srpc_server_rpc_t *rpc)
1069 {
1070         srpc_free_bulk(rpc->srpc_bulk);
1071         rpc->srpc_bulk = NULL;
1072 }
1073
1074 int
1075 sfw_alloc_pages (srpc_server_rpc_t *rpc, int npages, int sink)
1076 {
1077         LASSERT (rpc->srpc_bulk == NULL);
1078         LASSERT (npages > 0 && npages <= LNET_MAX_IOV);
1079
1080         rpc->srpc_bulk = srpc_alloc_bulk(npages, sink);
1081         if (rpc->srpc_bulk == NULL) return -ENOMEM;
1082
1083         return 0;
1084 }
1085
1086 int
1087 sfw_add_test (srpc_server_rpc_t *rpc)
1088 {
1089         sfw_session_t     *sn = sfw_data.fw_session;
1090         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1091         srpc_test_reqst_t *request;
1092         int                rc;
1093         sfw_batch_t       *bat;
1094
1095         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1096         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1097
1098         if (request->tsr_loop == 0 ||
1099             request->tsr_concur == 0 ||
1100             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1101             request->tsr_ndest > SFW_MAX_NDESTS ||
1102             (request->tsr_is_client && request->tsr_ndest == 0) ||
1103             request->tsr_concur > SFW_MAX_CONCUR ||
1104             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1105             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1106                 reply->tsr_status = EINVAL;
1107                 return 0;
1108         }
1109
1110         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1111             sfw_find_test_case(request->tsr_service) == NULL) {
1112                 reply->tsr_status = ENOENT;
1113                 return 0;
1114         }
1115
1116         bat = sfw_bid2batch(request->tsr_bid);
1117         if (bat == NULL) {
1118                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1119                         rpc->srpc_service->sv_name,
1120                         libcfs_id2str(rpc->srpc_peer));
1121                 return -ENOMEM;
1122         }
1123
1124         if (sfw_batch_active(bat)) {
1125                 reply->tsr_status = EBUSY;
1126                 return 0;
1127         }
1128
1129         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1130                 /* rpc will be resumed later in sfw_bulk_ready */
1131                 return sfw_alloc_pages(rpc,
1132                                        sfw_id_pages(request->tsr_ndest), 1);
1133         }
1134
1135         rc = sfw_add_test_instance(bat, rpc);
1136         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1137                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1138                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1139                 request->tsr_is_client ? "client" : "server",
1140                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1141
1142         reply->tsr_status = (rc < 0) ? -rc : rc;
1143         return 0;
1144 }
1145
1146 int
1147 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1148 {
1149         sfw_session_t *sn = sfw_data.fw_session;
1150         int            rc = 0;
1151         sfw_batch_t   *bat;
1152
1153         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1154
1155         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1156                 reply->bar_status = ESRCH;
1157                 return 0;
1158         }
1159
1160         bat = sfw_find_batch(request->bar_bid);
1161         if (bat == NULL) {
1162                 reply->bar_status = ENOENT;
1163                 return 0;
1164         }
1165
1166         switch (request->bar_opc) {
1167         case SRPC_BATCH_OPC_RUN:
1168                 rc = sfw_run_batch(bat);
1169                 break;
1170
1171         case SRPC_BATCH_OPC_STOP:
1172                 rc = sfw_stop_batch(bat, request->bar_arg);
1173                 break;
1174
1175         case SRPC_BATCH_OPC_QUERY:
1176                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1177                 break;
1178
1179         default:
1180                 return -EINVAL; /* drop it */
1181         }
1182
1183         reply->bar_status = (rc < 0) ? -rc : rc;
1184         return 0;
1185 }
1186
1187 int
1188 sfw_handle_server_rpc (srpc_server_rpc_t *rpc)
1189 {
1190         srpc_service_t *sv = rpc->srpc_service;
1191         srpc_msg_t     *reply = &rpc->srpc_replymsg;
1192         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1193         int             rc = 0;
1194
1195         LASSERT (sfw_data.fw_active_srpc == NULL);
1196         LASSERT (sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1197
1198         spin_lock(&sfw_data.fw_lock);
1199
1200         if (sfw_data.fw_shuttingdown) {
1201                 spin_unlock(&sfw_data.fw_lock);
1202                 return -ESHUTDOWN;
1203         }
1204
1205         /* Remove timer to avoid racing with it or expiring active session */
1206         if (sfw_del_session_timer() != 0) {
1207                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer.",
1208                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1209                 spin_unlock(&sfw_data.fw_lock);
1210                 return -EAGAIN;
1211         }
1212
1213         sfw_data.fw_active_srpc = rpc;
1214         spin_unlock(&sfw_data.fw_lock);
1215
1216         sfw_unpack_message(request);
1217         LASSERT (request->msg_type == srpc_service2request(sv->sv_id));
1218
1219         switch(sv->sv_id) {
1220         default:
1221                 LBUG ();
1222         case SRPC_SERVICE_TEST:
1223                 rc = sfw_add_test(rpc);
1224                 break;
1225
1226         case SRPC_SERVICE_BATCH:
1227                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1228                                        &reply->msg_body.bat_reply);
1229                 break;
1230
1231         case SRPC_SERVICE_QUERY_STAT:
1232                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1233                                    &reply->msg_body.stat_reply);
1234                 break;
1235
1236         case SRPC_SERVICE_DEBUG:
1237                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1238                                        &reply->msg_body.dbg_reply);
1239                 break;
1240
1241         case SRPC_SERVICE_MAKE_SESSION:
1242                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1243                                       &reply->msg_body.mksn_reply);
1244                 break;
1245
1246         case SRPC_SERVICE_REMOVE_SESSION:
1247                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1248                                         &reply->msg_body.rmsn_reply);
1249                 break;
1250         }
1251
1252         rpc->srpc_done = sfw_server_rpc_done;
1253         spin_lock(&sfw_data.fw_lock);
1254
1255 #ifdef __KERNEL__
1256         if (!sfw_data.fw_shuttingdown)
1257                 sfw_add_session_timer();
1258 #else
1259         LASSERT (!sfw_data.fw_shuttingdown);
1260         sfw_add_session_timer();
1261 #endif
1262
1263         sfw_data.fw_active_srpc = NULL;
1264         spin_unlock(&sfw_data.fw_lock);
1265         return rc;
1266 }
1267
1268 int
1269 sfw_bulk_ready (srpc_server_rpc_t *rpc, int status)
1270 {
1271         srpc_service_t *sv = rpc->srpc_service;
1272         int             rc;
1273
1274         LASSERT (rpc->srpc_bulk != NULL);
1275         LASSERT (sv->sv_id == SRPC_SERVICE_TEST);
1276         LASSERT (sfw_data.fw_active_srpc == NULL);
1277         LASSERT (rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1278
1279         spin_lock(&sfw_data.fw_lock);
1280
1281         if (status != 0) {
1282                 CERROR ("Bulk transfer failed for RPC: "
1283                         "service %s, peer %s, status %d\n",
1284                         sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1285                 spin_unlock(&sfw_data.fw_lock);
1286                 return -EIO;
1287         }
1288
1289         if (sfw_data.fw_shuttingdown) {
1290                 spin_unlock(&sfw_data.fw_lock);
1291                 return -ESHUTDOWN;
1292         }
1293
1294         if (sfw_del_session_timer() != 0) {
1295                 CERROR ("Dropping RPC (%s) from %s: racing with expiry timer",
1296                         sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1297                 spin_unlock(&sfw_data.fw_lock);
1298                 return -EAGAIN;
1299         }
1300
1301         sfw_data.fw_active_srpc = rpc;
1302         spin_unlock(&sfw_data.fw_lock);
1303
1304         rc = sfw_add_test(rpc);
1305
1306         spin_lock(&sfw_data.fw_lock);
1307
1308 #ifdef __KERNEL__
1309         if (!sfw_data.fw_shuttingdown)
1310                 sfw_add_session_timer();
1311 #else
1312         LASSERT (!sfw_data.fw_shuttingdown);
1313         sfw_add_session_timer();
1314 #endif
1315
1316         sfw_data.fw_active_srpc = NULL;
1317         spin_unlock(&sfw_data.fw_lock);
1318         return rc;
1319 }
1320
1321 srpc_client_rpc_t *
1322 sfw_create_rpc (lnet_process_id_t peer, int service,
1323                 int nbulkiov, int bulklen,
1324                 void (*done) (srpc_client_rpc_t *), void *priv)
1325 {
1326         srpc_client_rpc_t *rpc;
1327
1328         spin_lock(&sfw_data.fw_lock);
1329
1330         LASSERT (!sfw_data.fw_shuttingdown);
1331         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1332
1333         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1334                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1335                                  srpc_client_rpc_t, crpc_list);
1336                 list_del(&rpc->crpc_list);
1337                 spin_unlock(&sfw_data.fw_lock);
1338
1339                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1340                                      done, sfw_client_rpc_fini, priv);
1341                 return rpc;
1342         }
1343
1344         spin_unlock(&sfw_data.fw_lock);
1345
1346         rpc = srpc_create_client_rpc(peer, service, nbulkiov, bulklen, done,
1347                                      nbulkiov != 0 ? NULL : sfw_client_rpc_fini,
1348                                      priv);
1349         return rpc;
1350 }
1351
1352 void
1353 sfw_unpack_message (srpc_msg_t *msg)
1354 {
1355         if (msg->msg_magic == SRPC_MSG_MAGIC)
1356                 return; /* no flipping needed */
1357
1358         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1359
1360         __swab32s(&msg->msg_type);
1361
1362         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1363                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1364
1365                 __swab32s(&req->str_type);
1366                 __swab64s(&req->str_rpyid);
1367                 sfw_unpack_sid(req->str_sid);
1368                 return;
1369         }
1370
1371         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1372                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1373
1374                 __swab32s(&rep->str_status);
1375                 sfw_unpack_sid(rep->str_sid);
1376                 sfw_unpack_fw_counters(rep->str_fw);
1377                 sfw_unpack_rpc_counters(rep->str_rpc);
1378                 sfw_unpack_lnet_counters(rep->str_lnet);
1379                 return;
1380         }
1381
1382         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1383                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1384
1385                 __swab64s(&req->mksn_rpyid);
1386                 __swab32s(&req->mksn_force);
1387                 sfw_unpack_sid(req->mksn_sid);
1388                 return;
1389         }
1390
1391         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1392                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1393
1394                 __swab32s(&rep->mksn_status);
1395                 __swab32s(&rep->mksn_timeout);
1396                 sfw_unpack_sid(rep->mksn_sid);
1397                 return;
1398         }
1399
1400         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1401                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1402
1403                 __swab64s(&req->rmsn_rpyid);
1404                 sfw_unpack_sid(req->rmsn_sid);
1405                 return;
1406         }
1407
1408         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1409                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1410
1411                 __swab32s(&rep->rmsn_status);
1412                 sfw_unpack_sid(rep->rmsn_sid);
1413                 return;
1414         }
1415
1416         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1417                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1418
1419                 __swab64s(&req->dbg_rpyid);
1420                 __swab32s(&req->dbg_flags);
1421                 sfw_unpack_sid(req->dbg_sid);
1422                 return;
1423         }
1424
1425         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1426                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1427
1428                 __swab32s(&rep->dbg_nbatch);
1429                 __swab32s(&rep->dbg_timeout);
1430                 sfw_unpack_sid(rep->dbg_sid);
1431                 return;
1432         }
1433
1434         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1435                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1436
1437                 __swab32s(&req->bar_opc);
1438                 __swab64s(&req->bar_rpyid);
1439                 __swab32s(&req->bar_testidx);
1440                 __swab32s(&req->bar_arg);
1441                 sfw_unpack_sid(req->bar_sid);
1442                 __swab64s(&req->bar_bid.bat_id);
1443                 return;
1444         }
1445
1446         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1447                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1448
1449                 __swab32s(&rep->bar_status);
1450                 sfw_unpack_sid(rep->bar_sid);
1451                 return;
1452         }
1453
1454         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1455                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1456
1457                 __swab64s(&req->tsr_rpyid);
1458                 __swab64s(&req->tsr_bulkid);
1459                 __swab32s(&req->tsr_loop);
1460                 __swab32s(&req->tsr_ndest);
1461                 __swab32s(&req->tsr_concur);
1462                 __swab32s(&req->tsr_service);
1463                 sfw_unpack_sid(req->tsr_sid);
1464                 __swab64s(&req->tsr_bid.bat_id);
1465                 return;
1466         }
1467
1468         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1469                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1470
1471                 __swab32s(&rep->tsr_status);
1472                 sfw_unpack_sid(rep->tsr_sid);
1473                 return;
1474         }
1475
1476         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1477                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1478
1479                 __swab64s(&req->join_rpyid);
1480                 sfw_unpack_sid(req->join_sid);
1481                 return;
1482         }
1483
1484         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1485                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1486
1487                 __swab32s(&rep->join_status);
1488                 __swab32s(&rep->join_timeout);
1489                 sfw_unpack_sid(rep->join_sid);
1490                 return;
1491         }
1492
1493         LBUG ();
1494         return;
1495 }
1496
1497 void
1498 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1499 {
1500         LASSERT (atomic_read(&rpc->crpc_refcount) > 0);
1501         LASSERT (rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1502
1503         spin_lock(&rpc->crpc_lock);
1504         srpc_abort_rpc(rpc, -EINTR);
1505         spin_unlock(&rpc->crpc_lock);
1506         return;
1507 }
1508
1509 void
1510 sfw_post_rpc (srpc_client_rpc_t *rpc)
1511 {
1512         spin_lock(&rpc->crpc_lock);
1513
1514         LASSERT (!rpc->crpc_closed);
1515         LASSERT (!rpc->crpc_aborted);
1516         LASSERT (list_empty(&rpc->crpc_list));
1517         LASSERT (!sfw_data.fw_shuttingdown);
1518
1519         rpc->crpc_timeout = SFW_CLIENT_RPC_TIMEOUT;
1520         srpc_post_rpc(rpc);
1521
1522         spin_unlock(&rpc->crpc_lock);
1523         return;
1524 }
1525
1526 static srpc_service_t sfw_services[] =
1527 {
1528         {
1529                 /* sv_id */    SRPC_SERVICE_DEBUG,
1530                 /* sv_name */  "debug",
1531                 0
1532         },
1533         {
1534                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1535                 /* sv_name */  "query stats",
1536                 0
1537         },
1538         {
1539                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1540                 /* sv_name */  "make session",
1541                 0
1542         },
1543         {
1544                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1545                 /* sv_name */  "remove session",
1546                 0
1547         },
1548         {
1549                 /* sv_id */    SRPC_SERVICE_BATCH,
1550                 /* sv_name */  "batch service",
1551                 0
1552         },
1553         {
1554                 /* sv_id */    SRPC_SERVICE_TEST,
1555                 /* sv_name */  "test service",
1556                 0
1557         },
1558         {
1559                 /* sv_id */    0,
1560                 /* sv_name */  NULL,
1561                 0
1562         }
1563 };
1564
1565 extern sfw_test_client_ops_t ping_test_client;
1566 extern srpc_service_t        ping_test_service;
1567 extern void ping_init_test_client(void);
1568 extern void ping_init_test_service(void);
1569
1570 extern sfw_test_client_ops_t brw_test_client;
1571 extern srpc_service_t        brw_test_service;
1572 extern void brw_init_test_client(void);
1573 extern void brw_init_test_service(void);
1574
1575
1576 int
1577 sfw_startup (void)
1578 {
1579         int              i;
1580         int              rc;
1581         int              error;
1582         srpc_service_t  *sv;
1583         sfw_test_case_t *tsc;
1584
1585 #ifndef __KERNEL__
1586         char *s;
1587
1588         s = getenv("SESSION_TIMEOUT");
1589         session_timeout = s != NULL ? atoi(s) : session_timeout;
1590
1591         s = getenv("BRW_INJECT_ERRORS");
1592         brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
1593 #endif
1594
1595         if (session_timeout < 0) {
1596                 CERROR ("Session timeout must be non-negative: %d\n",
1597                         session_timeout);
1598                 return -EINVAL;
1599         }
1600
1601         if (session_timeout == 0)
1602                 CWARN ("Zero session_timeout specified "
1603                        "- test sessions never expire.\n");
1604
1605         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1606
1607         sfw_data.fw_session     = NULL;
1608         sfw_data.fw_active_srpc = NULL;
1609         spin_lock_init(&sfw_data.fw_lock);
1610         atomic_set(&sfw_data.fw_nzombies, 0);
1611         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1612         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1613         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1614
1615         brw_init_test_client();
1616         brw_init_test_service();
1617         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1618         LASSERT (rc == 0);
1619
1620         ping_init_test_client();
1621         ping_init_test_service();
1622         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1623         LASSERT (rc == 0);
1624
1625         error = 0;
1626         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1627                                        sfw_test_case_t, tsc_list) {
1628                 sv = tsc->tsc_srv_service;
1629                 sv->sv_concur = SFW_TEST_CONCURRENCY;
1630
1631                 rc = srpc_add_service(sv);
1632                 LASSERT (rc != -EBUSY);
1633                 if (rc != 0) {
1634                         CWARN ("Failed to add %s service: %d\n",
1635                                sv->sv_name, rc);
1636                         error = rc;
1637                 }
1638         }
1639
1640         for (i = 0; ; i++) {
1641                 sv = &sfw_services[i];
1642                 if (sv->sv_name == NULL) break;
1643
1644                 sv->sv_bulk_ready = NULL;
1645                 sv->sv_handler    = sfw_handle_server_rpc;
1646                 sv->sv_concur     = SFW_SERVICE_CONCURRENCY;
1647                 if (sv->sv_id == SRPC_SERVICE_TEST)
1648                         sv->sv_bulk_ready = sfw_bulk_ready;
1649
1650                 rc = srpc_add_service(sv);
1651                 LASSERT (rc != -EBUSY);
1652                 if (rc != 0) {
1653                         CWARN ("Failed to add %s service: %d\n",
1654                                sv->sv_name, rc);
1655                         error = rc;
1656                 }
1657
1658                 /* about to sfw_shutdown, no need to add buffer */
1659                 if (error) continue;
1660
1661                 rc = srpc_service_add_buffers(sv, SFW_POST_BUFFERS);
1662                 if (rc != SFW_POST_BUFFERS) {
1663                         CWARN ("Failed to reserve enough buffers: "
1664                                "service %s, %d needed, %d reserved\n",
1665                                sv->sv_name, SFW_POST_BUFFERS, rc);
1666                         error = -ENOMEM;
1667                 }
1668         }
1669
1670         if (error != 0)
1671                 sfw_shutdown();
1672         return error;
1673 }
1674
1675 void
1676 sfw_shutdown (void)
1677 {
1678         srpc_service_t  *sv;
1679         sfw_test_case_t *tsc;
1680         int              i;
1681
1682         spin_lock(&sfw_data.fw_lock);
1683
1684         sfw_data.fw_shuttingdown = 1;
1685 #ifdef __KERNEL__
1686         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1687                        "waiting for active RPC to finish.\n");
1688 #else
1689         LASSERT (sfw_data.fw_active_srpc == NULL);
1690 #endif
1691
1692         if (sfw_del_session_timer() != 0)
1693                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1694                                "waiting for session timer to explode.\n");
1695
1696         sfw_deactivate_session();
1697         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1698                        sfw_data.fw_lock,
1699                        "waiting for %d zombie sessions to die.\n",
1700                        atomic_read(&sfw_data.fw_nzombies));
1701
1702         spin_unlock(&sfw_data.fw_lock);
1703
1704         for (i = 0; ; i++) {
1705                 sv = &sfw_services[i];
1706                 if (sv->sv_name == NULL)
1707                         break;
1708
1709                 srpc_shutdown_service(sv);
1710                 srpc_remove_service(sv);
1711         }
1712
1713         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1714                                        sfw_test_case_t, tsc_list) {
1715                 sv = tsc->tsc_srv_service;
1716                 srpc_shutdown_service(sv);
1717                 srpc_remove_service(sv);
1718         }
1719
1720         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1721                 srpc_client_rpc_t *rpc;
1722
1723                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1724                                  srpc_client_rpc_t, crpc_list);
1725                 list_del(&rpc->crpc_list);
1726
1727                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1728         }
1729
1730         for (i = 0; ; i++) {
1731                 sv = &sfw_services[i];
1732                 if (sv->sv_name == NULL)
1733                         break;
1734
1735                 srpc_wait_service_shutdown(sv);
1736         }
1737
1738         while (!list_empty(&sfw_data.fw_tests)) {
1739                 tsc = list_entry(sfw_data.fw_tests.next,
1740                                  sfw_test_case_t, tsc_list);
1741
1742                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1743
1744                 list_del(&tsc->tsc_list);
1745                 LIBCFS_FREE(tsc, sizeof(*tsc));
1746         }
1747
1748         return;
1749 }