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;
 577         struct srpc_service     *svc;
 578         int                     nbuf;
 579         int                     rc;
 580
 581         LASSERT(tsi != NULL);
 582         tsc = sfw_find_test_case(tsi->tsi_service);
 583         nbuf = sfw_test_buffers(tsi);
 584         LASSERT(tsc != NULL);
 585         svc = tsc->tsc_srv_service;
 586
 587         if (tsi->tsi_is_client) {
 588                 tsi->tsi_ops = tsc->tsc_cli_ops;
 589                 return 0;
 590         }
 591
 592         rc = srpc_service_add_buffers(svc, nbuf);
 593         if (rc != 0) {
 594                 CWARN("Failed to reserve enough buffers: "
 595                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 596                 /* NB: this error handler is not strictly correct, because
 597                  * it may release more buffers than already allocated,
 598                  * but it doesn't matter because request portal should
 599                  * be lazy portal and will grow buffers if necessary. */
 600                 srpc_service_remove_buffers(svc, nbuf);
 601                 return -ENOMEM;
 602         }
 603
 604         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 605                nbuf * (srpc_serv_is_framework(svc) ?
 606                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 607         return 0;
 608 }
 609
 610 void
 611 sfw_unload_test(struct sfw_test_instance *tsi)
 612 {
 613         struct sfw_test_case *tsc = sfw_find_test_case(tsi->tsi_service);
 614
 615         LASSERT(tsc != NULL);
 616
 617         if (tsi->tsi_is_client)
 618                 return;
 619
 620         /* shrink buffers, because request portal is lazy portal
 621          * which can grow buffers at runtime so we may leave
 622          * some buffers behind, but never mind... */
 623         srpc_service_remove_buffers(tsc->tsc_srv_service,
 624                                     sfw_test_buffers(tsi));
 625         return;
 626 }
 627
 628 void
 629 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 630 {
 631         srpc_client_rpc_t *rpc;
 632         sfw_test_unit_t   *tsu;
 633
 634         if (!tsi->tsi_is_client) goto clean;
 635
 636         tsi->tsi_ops->tso_fini(tsi);
 637
 638         LASSERT (!tsi->tsi_stopping);
 639         LASSERT (cfs_list_empty(&tsi->tsi_active_rpcs));
 640         LASSERT (!sfw_test_active(tsi));
 641
 642         while (!cfs_list_empty(&tsi->tsi_units)) {
 643                 tsu = cfs_list_entry(tsi->tsi_units.next,
 644                                      sfw_test_unit_t, tsu_list);
 645                 cfs_list_del(&tsu->tsu_list);
 646                 LIBCFS_FREE(tsu, sizeof(*tsu));
 647         }
 648
 649         while (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 650                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 651                                      srpc_client_rpc_t, crpc_list);
 652                 cfs_list_del(&rpc->crpc_list);
 653                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 654         }
 655
 656 clean:
 657         sfw_unload_test(tsi);
 658         LIBCFS_FREE(tsi, sizeof(*tsi));
 659         return;
 660 }
 661
 662 void
 663 sfw_destroy_batch (sfw_batch_t *tsb)
 664 {
 665         sfw_test_instance_t *tsi;
 666
 667         LASSERT (!sfw_batch_active(tsb));
 668         LASSERT (cfs_list_empty(&tsb->bat_list));
 669
 670         while (!cfs_list_empty(&tsb->bat_tests)) {
 671                 tsi = cfs_list_entry(tsb->bat_tests.next,
 672                                      sfw_test_instance_t, tsi_list);
 673                 cfs_list_del_init(&tsi->tsi_list);
 674                 sfw_destroy_test_instance(tsi);
 675         }
 676
 677         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 678         return;
 679 }
 680
 681 void
 682 sfw_destroy_session (sfw_session_t *sn)
 683 {
 684         sfw_batch_t *batch;
 685
 686         LASSERT (cfs_list_empty(&sn->sn_list));
 687         LASSERT (sn != sfw_data.fw_session);
 688
 689         while (!cfs_list_empty(&sn->sn_batches)) {
 690                 batch = cfs_list_entry(sn->sn_batches.next,
 691                                        sfw_batch_t, bat_list);
 692                 cfs_list_del_init(&batch->bat_list);
 693                 sfw_destroy_batch(batch);
 694         }
 695
 696         LIBCFS_FREE(sn, sizeof(*sn));
 697         cfs_atomic_dec(&sfw_data.fw_nzombies);
 698         return;
 699 }
 700
 701 void
 702 sfw_unpack_addtest_req(srpc_msg_t *msg)
 703 {
 704         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 705
 706         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 707         LASSERT (req->tsr_is_client);
 708
 709         if (msg->msg_magic == SRPC_MSG_MAGIC)
 710                 return; /* no flipping needed */
 711
 712         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 713
 714         if (req->tsr_service == SRPC_SERVICE_BRW) {
 715                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 716                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 717
 718                         __swab32s(&bulk->blk_opc);
 719                         __swab32s(&bulk->blk_npg);
 720                         __swab32s(&bulk->blk_flags);
 721
 722                 } else {
 723                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 724
 725                         __swab16s(&bulk->blk_opc);
 726                         __swab16s(&bulk->blk_flags);
 727                         __swab32s(&bulk->blk_offset);
 728                         __swab32s(&bulk->blk_len);
 729                 }
 730
 731                 return;
 732         }
 733
 734         if (req->tsr_service == SRPC_SERVICE_PING) {
 735                 test_ping_req_t *ping = &req->tsr_u.ping;
 736
 737                 __swab32s(&ping->png_size);
 738                 __swab32s(&ping->png_flags);
 739                 return;
 740         }
 741
 742         LBUG ();
 743         return;
 744 }
 745
 746 int
 747 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 748 {
 749         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 750         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 751         srpc_bulk_t         *bk = rpc->srpc_bulk;
 752         int                  ndest = req->tsr_ndest;
 753         sfw_test_unit_t     *tsu;
 754         sfw_test_instance_t *tsi;
 755         int                  i;
 756         int                  rc;
 757
 758         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 759         if (tsi == NULL) {
 760                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 761                         tsb->bat_id.bat_id);
 762                 return -ENOMEM;
 763         }
 764
 765         memset(tsi, 0, sizeof(*tsi));
 766         spin_lock_init(&tsi->tsi_lock);
 767         cfs_atomic_set(&tsi->tsi_nactive, 0);
 768         CFS_INIT_LIST_HEAD(&tsi->tsi_units);
 769         CFS_INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 770         CFS_INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 771
 772         tsi->tsi_stopping      = 0;
 773         tsi->tsi_batch         = tsb;
 774         tsi->tsi_loop          = req->tsr_loop;
 775         tsi->tsi_concur        = req->tsr_concur;
 776         tsi->tsi_service       = req->tsr_service;
 777         tsi->tsi_is_client     = !!(req->tsr_is_client);
 778         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 779
 780         rc = sfw_load_test(tsi);
 781         if (rc != 0) {
 782                 LIBCFS_FREE(tsi, sizeof(*tsi));
 783                 return rc;
 784         }
 785
 786         LASSERT (!sfw_batch_active(tsb));
 787
 788         if (!tsi->tsi_is_client) {
 789                 /* it's test server, just add it to tsb */
 790                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 791                 return 0;
 792         }
 793
 794         LASSERT (bk != NULL);
 795 #ifndef __KERNEL__
 796         LASSERT (bk->bk_pages != NULL);
 797 #endif
 798         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 799         LASSERT((unsigned int)bk->bk_len >=
 800                 sizeof(lnet_process_id_packed_t) * ndest);
 801
 802         sfw_unpack_addtest_req(msg);
 803         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 804
 805         for (i = 0; i < ndest; i++) {
 806                 lnet_process_id_packed_t *dests;
 807                 lnet_process_id_packed_t  id;
 808                 int                       j;
 809
 810 #ifdef __KERNEL__
 811                 dests = cfs_page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 812                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 813 #else
 814                 dests = cfs_page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 815 #endif
 816                 id = dests[i % SFW_ID_PER_PAGE];
 817                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 818                         sfw_unpack_id(id);
 819
 820                 for (j = 0; j < tsi->tsi_concur; j++) {
 821                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 822                         if (tsu == NULL) {
 823                                 rc = -ENOMEM;
 824                                 CERROR ("Can't allocate tsu for %d\n",
 825                                         tsi->tsi_service);
 826                                 goto error;
 827                         }
 828
 829                         tsu->tsu_dest.nid = id.nid;
 830                         tsu->tsu_dest.pid = id.pid;
 831                         tsu->tsu_instance = tsi;
 832                         tsu->tsu_private  = NULL;
 833                         cfs_list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 834                 }
 835         }
 836
 837         rc = tsi->tsi_ops->tso_init(tsi);
 838         if (rc == 0) {
 839                 cfs_list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 840                 return 0;
 841         }
 842
 843 error:
 844         LASSERT (rc != 0);
 845         sfw_destroy_test_instance(tsi);
 846         return rc;
 847 }
 848
 849 static void
 850 sfw_test_unit_done (sfw_test_unit_t *tsu)
 851 {
 852         sfw_test_instance_t *tsi = tsu->tsu_instance;
 853         sfw_batch_t         *tsb = tsi->tsi_batch;
 854         sfw_session_t       *sn = tsb->bat_session;
 855
 856         LASSERT (sfw_test_active(tsi));
 857
 858         if (!cfs_atomic_dec_and_test(&tsi->tsi_nactive))
 859                 return;
 860
 861         /* the test instance is done */
 862         spin_lock(&tsi->tsi_lock);
 863
 864         tsi->tsi_stopping = 0;
 865
 866         spin_unlock(&tsi->tsi_lock);
 867
 868         spin_lock(&sfw_data.fw_lock);
 869
 870         if (!cfs_atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 871             sn == sfw_data.fw_session) {                  /* sn also active */
 872                 spin_unlock(&sfw_data.fw_lock);
 873                 return;
 874         }
 875
 876         LASSERT (!cfs_list_empty(&sn->sn_list)); /* I'm a zombie! */
 877
 878         cfs_list_for_each_entry_typed (tsb, &sn->sn_batches,
 879                                        sfw_batch_t, bat_list) {
 880                 if (sfw_batch_active(tsb)) {
 881                         spin_unlock(&sfw_data.fw_lock);
 882                         return;
 883                 }
 884         }
 885
 886         cfs_list_del_init(&sn->sn_list);
 887         spin_unlock(&sfw_data.fw_lock);
 888
 889         sfw_destroy_session(sn);
 890         return;
 891 }
 892
 893 void
 894 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 895 {
 896         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 897         sfw_test_instance_t *tsi = tsu->tsu_instance;
 898         int                  done = 0;
 899
 900         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 901
 902         spin_lock(&tsi->tsi_lock);
 903
 904         LASSERT (sfw_test_active(tsi));
 905         LASSERT (!cfs_list_empty(&rpc->crpc_list));
 906
 907         cfs_list_del_init(&rpc->crpc_list);
 908
 909         /* batch is stopping or loop is done or get error */
 910         if (tsi->tsi_stopping ||
 911             tsu->tsu_loop == 0 ||
 912             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 913                 done = 1;
 914
 915         /* dec ref for poster */
 916         srpc_client_rpc_decref(rpc);
 917
 918         spin_unlock(&tsi->tsi_lock);
 919
 920         if (!done) {
 921                 swi_schedule_workitem(&tsu->tsu_worker);
 922                 return;
 923         }
 924
 925         sfw_test_unit_done(tsu);
 926         return;
 927 }
 928
 929 int
 930 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
 931                     unsigned features, int nblk, int blklen,
 932                     srpc_client_rpc_t **rpcpp)
 933 {
 934         srpc_client_rpc_t   *rpc = NULL;
 935         sfw_test_instance_t *tsi = tsu->tsu_instance;
 936
 937         spin_lock(&tsi->tsi_lock);
 938
 939         LASSERT (sfw_test_active(tsi));
 940
 941         if (!cfs_list_empty(&tsi->tsi_free_rpcs)) {
 942                 /* pick request from buffer */
 943                 rpc = cfs_list_entry(tsi->tsi_free_rpcs.next,
 944                                      srpc_client_rpc_t, crpc_list);
 945                 LASSERT (nblk == rpc->crpc_bulk.bk_niov);
 946                 cfs_list_del_init(&rpc->crpc_list);
 947         }
 948
 949         spin_unlock(&tsi->tsi_lock);
 950
 951         if (rpc == NULL) {
 952                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 953                                              blklen, sfw_test_rpc_done,
 954                                              sfw_test_rpc_fini, tsu);
 955         } else {
 956                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 957                                      blklen, sfw_test_rpc_done,
 958                                      sfw_test_rpc_fini, tsu);
 959         }
 960
 961         if (rpc == NULL) {
 962                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 963                 return -ENOMEM;
 964         }
 965
 966         rpc->crpc_reqstmsg.msg_ses_feats = features;
 967         *rpcpp = rpc;
 968
 969         return 0;
 970 }
 971
 972 int
 973 sfw_run_test (swi_workitem_t *wi)
 974 {
 975         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 976         sfw_test_instance_t *tsi = tsu->tsu_instance;
 977         srpc_client_rpc_t   *rpc = NULL;
 978
 979         LASSERT (wi == &tsu->tsu_worker);
 980
 981         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 982                 LASSERT (rpc == NULL);
 983                 goto test_done;
 984         }
 985
 986         LASSERT (rpc != NULL);
 987
 988         spin_lock(&tsi->tsi_lock);
 989
 990         if (tsi->tsi_stopping) {
 991                 cfs_list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 992                 spin_unlock(&tsi->tsi_lock);
 993                 goto test_done;
 994         }
 995
 996         if (tsu->tsu_loop > 0)
 997                 tsu->tsu_loop--;
 998
 999         cfs_list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
1000         spin_unlock(&tsi->tsi_lock);
1001
1002         rpc->crpc_timeout = rpc_timeout;
1003
1004         spin_lock(&rpc->crpc_lock);
1005         srpc_post_rpc(rpc);
1006         spin_unlock(&rpc->crpc_lock);
1007         return 0;
1008
1009 test_done:
1010         /*
1011          * No one can schedule me now since:
1012          * - previous RPC, if any, has done and
1013          * - no new RPC is initiated.
1014          * - my batch is still active; no one can run it again now.
1015          * Cancel pending schedules and prevent future schedule attempts:
1016          */
1017         swi_exit_workitem(wi);
1018         sfw_test_unit_done(tsu);
1019         return 1;
1020 }
1021
1022 int
1023 sfw_run_batch (sfw_batch_t *tsb)
1024 {
1025         swi_workitem_t      *wi;
1026         sfw_test_unit_t     *tsu;
1027         sfw_test_instance_t *tsi;
1028
1029         if (sfw_batch_active(tsb)) {
1030                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
1031                        tsb->bat_id.bat_id, cfs_atomic_read(&tsb->bat_nactive));
1032                 return 0;
1033         }
1034
1035         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1036                                        sfw_test_instance_t, tsi_list) {
1037                 if (!tsi->tsi_is_client) /* skip server instances */
1038                         continue;
1039
1040                 LASSERT (!tsi->tsi_stopping);
1041                 LASSERT (!sfw_test_active(tsi));
1042
1043                 cfs_atomic_inc(&tsb->bat_nactive);
1044
1045                 cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
1046                                                sfw_test_unit_t, tsu_list) {
1047                         cfs_atomic_inc(&tsi->tsi_nactive);
1048                         tsu->tsu_loop = tsi->tsi_loop;
1049                         wi = &tsu->tsu_worker;
1050                         swi_init_workitem(wi, tsu, sfw_run_test,
1051                                           lst_sched_test[\
1052                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1053                         swi_schedule_workitem(wi);
1054                 }
1055         }
1056
1057         return 0;
1058 }
1059
1060 int
1061 sfw_stop_batch (sfw_batch_t *tsb, int force)
1062 {
1063         sfw_test_instance_t *tsi;
1064         srpc_client_rpc_t   *rpc;
1065
1066         if (!sfw_batch_active(tsb)) {
1067                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1068                 return 0;
1069         }
1070
1071         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1072                                        sfw_test_instance_t, tsi_list) {
1073                 spin_lock(&tsi->tsi_lock);
1074
1075                 if (!tsi->tsi_is_client ||
1076                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1077                         spin_unlock(&tsi->tsi_lock);
1078                         continue;
1079                 }
1080
1081                 tsi->tsi_stopping = 1;
1082
1083                 if (!force) {
1084                         spin_unlock(&tsi->tsi_lock);
1085                         continue;
1086                 }
1087
1088                 /* abort launched rpcs in the test */
1089                 cfs_list_for_each_entry_typed(rpc, &tsi->tsi_active_rpcs,
1090                                               srpc_client_rpc_t, crpc_list) {
1091                         spin_lock(&rpc->crpc_lock);
1092
1093                         srpc_abort_rpc(rpc, -EINTR);
1094
1095                         spin_unlock(&rpc->crpc_lock);
1096                 }
1097
1098                 spin_unlock(&tsi->tsi_lock);
1099         }
1100
1101         return 0;
1102 }
1103
1104 int
1105 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1106 {
1107         sfw_test_instance_t *tsi;
1108
1109         if (testidx < 0)
1110                 return -EINVAL;
1111
1112         if (testidx == 0) {
1113                 reply->bar_active = cfs_atomic_read(&tsb->bat_nactive);
1114                 return 0;
1115         }
1116
1117         cfs_list_for_each_entry_typed (tsi, &tsb->bat_tests,
1118                                        sfw_test_instance_t, tsi_list) {
1119                 if (testidx-- > 1)
1120                         continue;
1121
1122                 reply->bar_active = cfs_atomic_read(&tsi->tsi_nactive);
1123                 return 0;
1124         }
1125
1126         return -ENOENT;
1127 }
1128
1129 void
1130 sfw_free_pages (srpc_server_rpc_t *rpc)
1131 {
1132         srpc_free_bulk(rpc->srpc_bulk);
1133         rpc->srpc_bulk = NULL;
1134 }
1135
1136 int
1137 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1138                 int sink)
1139 {
1140         LASSERT(rpc->srpc_bulk == NULL);
1141         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1142
1143         rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1144         if (rpc->srpc_bulk == NULL)
1145                 return -ENOMEM;
1146
1147         return 0;
1148 }
1149
1150 int
1151 sfw_add_test (srpc_server_rpc_t *rpc)
1152 {
1153         sfw_session_t     *sn = sfw_data.fw_session;
1154         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1155         srpc_test_reqst_t *request;
1156         int                rc;
1157         sfw_batch_t       *bat;
1158
1159         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1160         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1161
1162         if (request->tsr_loop == 0 ||
1163             request->tsr_concur == 0 ||
1164             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1165             request->tsr_ndest > SFW_MAX_NDESTS ||
1166             (request->tsr_is_client && request->tsr_ndest == 0) ||
1167             request->tsr_concur > SFW_MAX_CONCUR ||
1168             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1169             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1170                 reply->tsr_status = EINVAL;
1171                 return 0;
1172         }
1173
1174         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1175             sfw_find_test_case(request->tsr_service) == NULL) {
1176                 reply->tsr_status = ENOENT;
1177                 return 0;
1178         }
1179
1180         bat = sfw_bid2batch(request->tsr_bid);
1181         if (bat == NULL) {
1182                 CERROR ("Dropping RPC (%s) from %s under memory pressure.\n",
1183                         rpc->srpc_scd->scd_svc->sv_name,
1184                         libcfs_id2str(rpc->srpc_peer));
1185                 return -ENOMEM;
1186         }
1187
1188         if (sfw_batch_active(bat)) {
1189                 reply->tsr_status = EBUSY;
1190                 return 0;
1191         }
1192
1193         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1194                 /* rpc will be resumed later in sfw_bulk_ready */
1195                 int     npg = sfw_id_pages(request->tsr_ndest);
1196                 int     len;
1197
1198                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1199                         len = npg * CFS_PAGE_SIZE;
1200
1201                 } else  {
1202                         len = sizeof(lnet_process_id_packed_t) *
1203                               request->tsr_ndest;
1204                 }
1205
1206                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1207         }
1208
1209         rc = sfw_add_test_instance(bat, rpc);
1210         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1211                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1212                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1213                 request->tsr_is_client ? "client" : "server",
1214                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1215
1216         reply->tsr_status = (rc < 0) ? -rc : rc;
1217         return 0;
1218 }
1219
1220 int
1221 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1222 {
1223         sfw_session_t *sn = sfw_data.fw_session;
1224         int            rc = 0;
1225         sfw_batch_t   *bat;
1226
1227         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1228
1229         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1230                 reply->bar_status = ESRCH;
1231                 return 0;
1232         }
1233
1234         bat = sfw_find_batch(request->bar_bid);
1235         if (bat == NULL) {
1236                 reply->bar_status = ENOENT;
1237                 return 0;
1238         }
1239
1240         switch (request->bar_opc) {
1241         case SRPC_BATCH_OPC_RUN:
1242                 rc = sfw_run_batch(bat);
1243                 break;
1244
1245         case SRPC_BATCH_OPC_STOP:
1246                 rc = sfw_stop_batch(bat, request->bar_arg);
1247                 break;
1248
1249         case SRPC_BATCH_OPC_QUERY:
1250                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1251                 break;
1252
1253         default:
1254                 return -EINVAL; /* drop it */
1255         }
1256
1257         reply->bar_status = (rc < 0) ? -rc : rc;
1258         return 0;
1259 }
1260
1261 int
1262 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1263 {
1264         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1265         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1266         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1267         unsigned        features = LST_FEATS_MASK;
1268         int             rc = 0;
1269
1270         LASSERT(sfw_data.fw_active_srpc == NULL);
1271         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1272
1273         spin_lock(&sfw_data.fw_lock);
1274
1275         if (sfw_data.fw_shuttingdown) {
1276                 spin_unlock(&sfw_data.fw_lock);
1277                 return -ESHUTDOWN;
1278         }
1279
1280         /* Remove timer to avoid racing with it or expiring active session */
1281         if (sfw_del_session_timer() != 0) {
1282                 CERROR("Dropping RPC (%s) from %s: racing with expiry timer.",
1283                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1284                 spin_unlock(&sfw_data.fw_lock);
1285                 return -EAGAIN;
1286         }
1287
1288         sfw_data.fw_active_srpc = rpc;
1289         spin_unlock(&sfw_data.fw_lock);
1290
1291         sfw_unpack_message(request);
1292         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1293
1294         /* rpc module should have checked this */
1295         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1296
1297         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1298             sv->sv_id != SRPC_SERVICE_DEBUG) {
1299                 sfw_session_t *sn = sfw_data.fw_session;
1300
1301                 if (sn != NULL &&
1302                     sn->sn_features != request->msg_ses_feats) {
1303                         CNETERR("Features of framework RPC don't match "
1304                                 "features of current session: %x/%x\n",
1305                                 request->msg_ses_feats, sn->sn_features);
1306                         reply->msg_body.reply.status = EPROTO;
1307                         reply->msg_body.reply.sid    = sn->sn_id;
1308                         goto out;
1309                 }
1310
1311         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1312                 /* NB: at this point, old version will ignore features and
1313                  * create new session anyway, so console should be able
1314                  * to handle this */
1315                 reply->msg_body.reply.status = EPROTO;
1316                 goto out;
1317         }
1318
1319         switch(sv->sv_id) {
1320         default:
1321                 LBUG ();
1322         case SRPC_SERVICE_TEST:
1323                 rc = sfw_add_test(rpc);
1324                 break;
1325
1326         case SRPC_SERVICE_BATCH:
1327                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1328                                        &reply->msg_body.bat_reply);
1329                 break;
1330
1331         case SRPC_SERVICE_QUERY_STAT:
1332                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1333                                    &reply->msg_body.stat_reply);
1334                 break;
1335
1336         case SRPC_SERVICE_DEBUG:
1337                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1338                                        &reply->msg_body.dbg_reply);
1339                 break;
1340
1341         case SRPC_SERVICE_MAKE_SESSION:
1342                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1343                                       &reply->msg_body.mksn_reply);
1344                 break;
1345
1346         case SRPC_SERVICE_REMOVE_SESSION:
1347                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1348                                         &reply->msg_body.rmsn_reply);
1349                 break;
1350         }
1351
1352         if (sfw_data.fw_session != NULL)
1353                 features = sfw_data.fw_session->sn_features;
1354  out:
1355         reply->msg_ses_feats = features;
1356         rpc->srpc_done = sfw_server_rpc_done;
1357         spin_lock(&sfw_data.fw_lock);
1358
1359 #ifdef __KERNEL__
1360         if (!sfw_data.fw_shuttingdown)
1361                 sfw_add_session_timer();
1362 #else
1363         LASSERT(!sfw_data.fw_shuttingdown);
1364         sfw_add_session_timer();
1365 #endif
1366
1367         sfw_data.fw_active_srpc = NULL;
1368         spin_unlock(&sfw_data.fw_lock);
1369         return rc;
1370 }
1371
1372 int
1373 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1374 {
1375         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1376         int                     rc;
1377
1378         LASSERT(rpc->srpc_bulk != NULL);
1379         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1380         LASSERT(sfw_data.fw_active_srpc == NULL);
1381         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1382
1383         spin_lock(&sfw_data.fw_lock);
1384
1385         if (status != 0) {
1386                 CERROR("Bulk transfer failed for RPC: "
1387                        "service %s, peer %s, status %d\n",
1388                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1389                 spin_unlock(&sfw_data.fw_lock);
1390                 return -EIO;
1391         }
1392
1393         if (sfw_data.fw_shuttingdown) {
1394                 spin_unlock(&sfw_data.fw_lock);
1395                 return -ESHUTDOWN;
1396         }
1397
1398         if (sfw_del_session_timer() != 0) {
1399                 CERROR("Dropping RPC (%s) from %s: racing with expiry timer",
1400                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1401                 spin_unlock(&sfw_data.fw_lock);
1402                 return -EAGAIN;
1403         }
1404
1405         sfw_data.fw_active_srpc = rpc;
1406         spin_unlock(&sfw_data.fw_lock);
1407
1408         rc = sfw_add_test(rpc);
1409
1410         spin_lock(&sfw_data.fw_lock);
1411
1412 #ifdef __KERNEL__
1413         if (!sfw_data.fw_shuttingdown)
1414                 sfw_add_session_timer();
1415 #else
1416         LASSERT(!sfw_data.fw_shuttingdown);
1417         sfw_add_session_timer();
1418 #endif
1419
1420         sfw_data.fw_active_srpc = NULL;
1421         spin_unlock(&sfw_data.fw_lock);
1422         return rc;
1423 }
1424
1425 srpc_client_rpc_t *
1426 sfw_create_rpc(lnet_process_id_t peer, int service,
1427                unsigned features, int nbulkiov, int bulklen,
1428                void (*done)(srpc_client_rpc_t *), void *priv)
1429 {
1430         srpc_client_rpc_t *rpc = NULL;
1431
1432         spin_lock(&sfw_data.fw_lock);
1433
1434         LASSERT (!sfw_data.fw_shuttingdown);
1435         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1436
1437         if (nbulkiov == 0 && !cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1438                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1439                                      srpc_client_rpc_t, crpc_list);
1440                 cfs_list_del(&rpc->crpc_list);
1441
1442                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1443                                      done, sfw_client_rpc_fini, priv);
1444         }
1445
1446         spin_unlock(&sfw_data.fw_lock);
1447
1448         if (rpc == NULL) {
1449                 rpc = srpc_create_client_rpc(peer, service,
1450                                              nbulkiov, bulklen, done,
1451                                              nbulkiov != 0 ?  NULL :
1452                                              sfw_client_rpc_fini,
1453                                              priv);
1454         }
1455
1456         if (rpc != NULL) /* "session" is concept in framework */
1457                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1458
1459         return rpc;
1460 }
1461
1462 void
1463 sfw_unpack_message (srpc_msg_t *msg)
1464 {
1465         if (msg->msg_magic == SRPC_MSG_MAGIC)
1466                 return; /* no flipping needed */
1467
1468         /* srpc module should guarantee I wouldn't get crap */
1469         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1470
1471         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1472                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1473
1474                 __swab32s(&req->str_type);
1475                 __swab64s(&req->str_rpyid);
1476                 sfw_unpack_sid(req->str_sid);
1477                 return;
1478         }
1479
1480         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1481                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1482
1483                 __swab32s(&rep->str_status);
1484                 sfw_unpack_sid(rep->str_sid);
1485                 sfw_unpack_fw_counters(rep->str_fw);
1486                 sfw_unpack_rpc_counters(rep->str_rpc);
1487                 sfw_unpack_lnet_counters(rep->str_lnet);
1488                 return;
1489         }
1490
1491         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1492                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1493
1494                 __swab64s(&req->mksn_rpyid);
1495                 __swab32s(&req->mksn_force);
1496                 sfw_unpack_sid(req->mksn_sid);
1497                 return;
1498         }
1499
1500         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1501                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1502
1503                 __swab32s(&rep->mksn_status);
1504                 __swab32s(&rep->mksn_timeout);
1505                 sfw_unpack_sid(rep->mksn_sid);
1506                 return;
1507         }
1508
1509         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1510                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1511
1512                 __swab64s(&req->rmsn_rpyid);
1513                 sfw_unpack_sid(req->rmsn_sid);
1514                 return;
1515         }
1516
1517         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1518                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1519
1520                 __swab32s(&rep->rmsn_status);
1521                 sfw_unpack_sid(rep->rmsn_sid);
1522                 return;
1523         }
1524
1525         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1526                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1527
1528                 __swab64s(&req->dbg_rpyid);
1529                 __swab32s(&req->dbg_flags);
1530                 sfw_unpack_sid(req->dbg_sid);
1531                 return;
1532         }
1533
1534         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1535                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1536
1537                 __swab32s(&rep->dbg_nbatch);
1538                 __swab32s(&rep->dbg_timeout);
1539                 sfw_unpack_sid(rep->dbg_sid);
1540                 return;
1541         }
1542
1543         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1544                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1545
1546                 __swab32s(&req->bar_opc);
1547                 __swab64s(&req->bar_rpyid);
1548                 __swab32s(&req->bar_testidx);
1549                 __swab32s(&req->bar_arg);
1550                 sfw_unpack_sid(req->bar_sid);
1551                 __swab64s(&req->bar_bid.bat_id);
1552                 return;
1553         }
1554
1555         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1556                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1557
1558                 __swab32s(&rep->bar_status);
1559                 sfw_unpack_sid(rep->bar_sid);
1560                 return;
1561         }
1562
1563         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1564                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1565
1566                 __swab64s(&req->tsr_rpyid);
1567                 __swab64s(&req->tsr_bulkid);
1568                 __swab32s(&req->tsr_loop);
1569                 __swab32s(&req->tsr_ndest);
1570                 __swab32s(&req->tsr_concur);
1571                 __swab32s(&req->tsr_service);
1572                 sfw_unpack_sid(req->tsr_sid);
1573                 __swab64s(&req->tsr_bid.bat_id);
1574                 return;
1575         }
1576
1577         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1578                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1579
1580                 __swab32s(&rep->tsr_status);
1581                 sfw_unpack_sid(rep->tsr_sid);
1582                 return;
1583         }
1584
1585         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1586                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1587
1588                 __swab64s(&req->join_rpyid);
1589                 sfw_unpack_sid(req->join_sid);
1590                 return;
1591         }
1592
1593         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1594                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1595
1596                 __swab32s(&rep->join_status);
1597                 __swab32s(&rep->join_timeout);
1598                 sfw_unpack_sid(rep->join_sid);
1599                 return;
1600         }
1601
1602         LBUG ();
1603         return;
1604 }
1605
1606 void
1607 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1608 {
1609         LASSERT(cfs_atomic_read(&rpc->crpc_refcount) > 0);
1610         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1611
1612         spin_lock(&rpc->crpc_lock);
1613         srpc_abort_rpc(rpc, -EINTR);
1614         spin_unlock(&rpc->crpc_lock);
1615         return;
1616 }
1617
1618 void
1619 sfw_post_rpc (srpc_client_rpc_t *rpc)
1620 {
1621         spin_lock(&rpc->crpc_lock);
1622
1623         LASSERT (!rpc->crpc_closed);
1624         LASSERT (!rpc->crpc_aborted);
1625         LASSERT (cfs_list_empty(&rpc->crpc_list));
1626         LASSERT (!sfw_data.fw_shuttingdown);
1627
1628         rpc->crpc_timeout = rpc_timeout;
1629         srpc_post_rpc(rpc);
1630
1631         spin_unlock(&rpc->crpc_lock);
1632         return;
1633 }
1634
1635 static srpc_service_t sfw_services[] =
1636 {
1637         {
1638                 /* sv_id */    SRPC_SERVICE_DEBUG,
1639                 /* sv_name */  "debug",
1640                 0
1641         },
1642         {
1643                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1644                 /* sv_name */  "query stats",
1645                 0
1646         },
1647         {
1648                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1649                 /* sv_name */  "make session",
1650                 0
1651         },
1652         {
1653                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1654                 /* sv_name */  "remove session",
1655                 0
1656         },
1657         {
1658                 /* sv_id */    SRPC_SERVICE_BATCH,
1659                 /* sv_name */  "batch service",
1660                 0
1661         },
1662         {
1663                 /* sv_id */    SRPC_SERVICE_TEST,
1664                 /* sv_name */  "test service",
1665                 0
1666         },
1667         {
1668                 /* sv_id */    0,
1669                 /* sv_name */  NULL,
1670                 0
1671         }
1672 };
1673
1674 extern sfw_test_client_ops_t ping_test_client;
1675 extern srpc_service_t        ping_test_service;
1676 extern void ping_init_test_client(void);
1677 extern void ping_init_test_service(void);
1678
1679 extern sfw_test_client_ops_t brw_test_client;
1680 extern srpc_service_t        brw_test_service;
1681 extern void brw_init_test_client(void);
1682 extern void brw_init_test_service(void);
1683
1684
1685 int
1686 sfw_startup (void)
1687 {
1688         int              i;
1689         int              rc;
1690         int              error;
1691         srpc_service_t  *sv;
1692         sfw_test_case_t *tsc;
1693
1694 #ifndef __KERNEL__
1695         char *s;
1696
1697         s = getenv("SESSION_TIMEOUT");
1698         session_timeout = s != NULL ? atoi(s) : session_timeout;
1699
1700         s = getenv("RPC_TIMEOUT");
1701         rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1702 #endif
1703
1704         if (session_timeout < 0) {
1705                 CERROR ("Session timeout must be non-negative: %d\n",
1706                         session_timeout);
1707                 return -EINVAL;
1708         }
1709
1710         if (rpc_timeout < 0) {
1711                 CERROR ("RPC timeout must be non-negative: %d\n",
1712                         rpc_timeout);
1713                 return -EINVAL;
1714         }
1715
1716         if (session_timeout == 0)
1717                 CWARN ("Zero session_timeout specified "
1718                        "- test sessions never expire.\n");
1719
1720         if (rpc_timeout == 0)
1721                 CWARN ("Zero rpc_timeout specified "
1722                        "- test RPC never expire.\n");
1723
1724         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1725
1726         sfw_data.fw_session     = NULL;
1727         sfw_data.fw_active_srpc = NULL;
1728         spin_lock_init(&sfw_data.fw_lock);
1729         cfs_atomic_set(&sfw_data.fw_nzombies, 0);
1730         CFS_INIT_LIST_HEAD(&sfw_data.fw_tests);
1731         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1732         CFS_INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1733
1734         brw_init_test_client();
1735         brw_init_test_service();
1736         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1737         LASSERT (rc == 0);
1738
1739         ping_init_test_client();
1740         ping_init_test_service();
1741         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1742         LASSERT (rc == 0);
1743
1744         error = 0;
1745         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1746                                        sfw_test_case_t, tsc_list) {
1747                 sv = tsc->tsc_srv_service;
1748
1749                 rc = srpc_add_service(sv);
1750                 LASSERT (rc != -EBUSY);
1751                 if (rc != 0) {
1752                         CWARN ("Failed to add %s service: %d\n",
1753                                sv->sv_name, rc);
1754                         error = rc;
1755                 }
1756         }
1757
1758         for (i = 0; ; i++) {
1759                 sv = &sfw_services[i];
1760                 if (sv->sv_name == NULL) break;
1761
1762                 sv->sv_bulk_ready = NULL;
1763                 sv->sv_handler    = sfw_handle_server_rpc;
1764                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1765                 if (sv->sv_id == SRPC_SERVICE_TEST)
1766                         sv->sv_bulk_ready = sfw_bulk_ready;
1767
1768                 rc = srpc_add_service(sv);
1769                 LASSERT (rc != -EBUSY);
1770                 if (rc != 0) {
1771                         CWARN ("Failed to add %s service: %d\n",
1772                                sv->sv_name, rc);
1773                         error = rc;
1774                 }
1775
1776                 /* about to sfw_shutdown, no need to add buffer */
1777                 if (error) continue;
1778
1779                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1780                 if (rc != 0) {
1781                         CWARN("Failed to reserve enough buffers: "
1782                               "service %s, %d needed: %d\n",
1783                               sv->sv_name, sv->sv_wi_total, rc);
1784                         error = -ENOMEM;
1785                 }
1786         }
1787
1788         if (error != 0)
1789                 sfw_shutdown();
1790         return error;
1791 }
1792
1793 void
1794 sfw_shutdown (void)
1795 {
1796         srpc_service_t  *sv;
1797         sfw_test_case_t *tsc;
1798         int              i;
1799
1800         spin_lock(&sfw_data.fw_lock);
1801
1802         sfw_data.fw_shuttingdown = 1;
1803 #ifdef __KERNEL__
1804         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1805                        "waiting for active RPC to finish.\n");
1806 #else
1807         LASSERT (sfw_data.fw_active_srpc == NULL);
1808 #endif
1809
1810         if (sfw_del_session_timer() != 0)
1811                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1812                                "waiting for session timer to explode.\n");
1813
1814         sfw_deactivate_session();
1815         lst_wait_until(cfs_atomic_read(&sfw_data.fw_nzombies) == 0,
1816                        sfw_data.fw_lock,
1817                        "waiting for %d zombie sessions to die.\n",
1818                        cfs_atomic_read(&sfw_data.fw_nzombies));
1819
1820         spin_unlock(&sfw_data.fw_lock);
1821
1822         for (i = 0; ; i++) {
1823                 sv = &sfw_services[i];
1824                 if (sv->sv_name == NULL)
1825                         break;
1826
1827                 srpc_shutdown_service(sv);
1828                 srpc_remove_service(sv);
1829         }
1830
1831         cfs_list_for_each_entry_typed (tsc, &sfw_data.fw_tests,
1832                                        sfw_test_case_t, tsc_list) {
1833                 sv = tsc->tsc_srv_service;
1834                 srpc_shutdown_service(sv);
1835                 srpc_remove_service(sv);
1836         }
1837
1838         while (!cfs_list_empty(&sfw_data.fw_zombie_rpcs)) {
1839                 srpc_client_rpc_t *rpc;
1840
1841                 rpc = cfs_list_entry(sfw_data.fw_zombie_rpcs.next,
1842                                      srpc_client_rpc_t, crpc_list);
1843                 cfs_list_del(&rpc->crpc_list);
1844
1845                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1846         }
1847
1848         for (i = 0; ; i++) {
1849                 sv = &sfw_services[i];
1850                 if (sv->sv_name == NULL)
1851                         break;
1852
1853                 srpc_wait_service_shutdown(sv);
1854         }
1855
1856         while (!cfs_list_empty(&sfw_data.fw_tests)) {
1857                 tsc = cfs_list_entry(sfw_data.fw_tests.next,
1858                                      sfw_test_case_t, tsc_list);
1859
1860                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1861
1862                 cfs_list_del(&tsc->tsc_list);
1863                 LIBCFS_FREE(tsc, sizeof(*tsc));
1864         }
1865
1866         return;
1867 }