lnet/selftest/framework.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lnet/selftest/framework.c
  32  *
  33  * Author: Isaac Huang <isaac@clusterfs.com>
  34  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LNET
  38
  39 #include "selftest.h"
  40
  41 struct lst_session_id LST_INVALID_SID = { .ses_nid = LNET_ANY_NID, .ses_stamp = -1};
  42
  43 static int session_timeout = 100;
  44 module_param(session_timeout, int, 0444);
  45 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
  46
  47 static int rpc_timeout = 64;
  48 module_param(rpc_timeout, int, 0644);
  49 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
  50
  51 #define sfw_unpack_id(id)               \
  52 do {                                    \
  53         __swab64s(&(id).nid);           \
  54         __swab32s(&(id).pid);           \
  55 } while (0)
  56
  57 #define sfw_unpack_sid(sid)             \
  58 do {                                    \
  59         __swab64s(&(sid).ses_nid);      \
  60         __swab64s(&(sid).ses_stamp);    \
  61 } while (0)
  62
  63 #define sfw_unpack_fw_counters(fc)        \
  64 do {                                      \
  65         __swab32s(&(fc).running_ms);      \
  66         __swab32s(&(fc).active_batches);  \
  67         __swab32s(&(fc).zombie_sessions); \
  68         __swab32s(&(fc).brw_errors);      \
  69         __swab32s(&(fc).ping_errors);     \
  70 } while (0)
  71
  72 #define sfw_unpack_rpc_counters(rc)     \
  73 do {                                    \
  74         __swab32s(&(rc).errors);        \
  75         __swab32s(&(rc).rpcs_sent);     \
  76         __swab32s(&(rc).rpcs_rcvd);     \
  77         __swab32s(&(rc).rpcs_dropped);  \
  78         __swab32s(&(rc).rpcs_expired);  \
  79         __swab64s(&(rc).bulk_get);      \
  80         __swab64s(&(rc).bulk_put);      \
  81 } while (0)
  82
  83 #define sfw_unpack_lnet_counters(lc)    \
  84 do {                                    \
  85         __swab32s(&(lc).lcc_errors);        \
  86         __swab32s(&(lc).lcc_msgs_max);      \
  87         __swab32s(&(lc).lcc_msgs_alloc);    \
  88         __swab32s(&(lc).lcc_send_count);    \
  89         __swab32s(&(lc).lcc_recv_count);    \
  90         __swab32s(&(lc).lcc_drop_count);    \
  91         __swab32s(&(lc).lcc_route_count);   \
  92         __swab64s(&(lc).lcc_send_length);   \
  93         __swab64s(&(lc).lcc_recv_length);   \
  94         __swab64s(&(lc).lcc_drop_length);   \
  95         __swab64s(&(lc).lcc_route_length);  \
  96 } while (0)
  97
  98 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
  99 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 100
 101 static struct smoketest_framework {
 102         /* RPCs to be recycled */
 103         struct list_head        fw_zombie_rpcs;
 104         /* stopping sessions */
 105         struct list_head        fw_zombie_sessions;
 106         /* registered test cases */
 107         struct list_head        fw_tests;
 108         /* # zombie sessions */
 109         atomic_t                fw_nzombies;
 110         /* serialise */
 111         spinlock_t              fw_lock;
 112         /* _the_ session */
 113         struct sfw_session      *fw_session;
 114         /* shutdown in progress */
 115         int                     fw_shuttingdown;
 116         /* running RPC */
 117         struct srpc_server_rpc  *fw_active_srpc;
 118 } sfw_data;
 119
 120 static struct srpc_service sfw_services[] = {
 121         { .sv_id = SRPC_SERVICE_DEBUG,          .sv_name = "debug", },
 122         { .sv_id = SRPC_SERVICE_QUERY_STAT,     .sv_name = "query stats", },
 123         { .sv_id = SRPC_SERVICE_MAKE_SESSION,   .sv_name = "make session", },
 124         { .sv_id = SRPC_SERVICE_REMOVE_SESSION, .sv_name = "remove session", },
 125         { .sv_id = SRPC_SERVICE_BATCH,          .sv_name = "batch service", },
 126         { .sv_id = SRPC_SERVICE_TEST,           .sv_name = "test service", },
 127         { .sv_id = 0, }
 128 };
 129
 130 /* forward ref's */
 131 static int sfw_stop_batch(struct sfw_batch *tsb, int force);
 132 static void sfw_destroy_session(struct sfw_session *sn);
 133
 134 static inline struct sfw_test_case *
 135 sfw_find_test_case(int id)
 136 {
 137         struct sfw_test_case *tsc;
 138
 139         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 140         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 141
 142         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 143                 if (tsc->tsc_srv_service->sv_id == id)
 144                         return tsc;
 145         }
 146
 147         return NULL;
 148 }
 149
 150 static int
 151 sfw_register_test(struct srpc_service *service,
 152                   struct sfw_test_client_ops *cliops)
 153 {
 154         struct sfw_test_case *tsc;
 155
 156         if (sfw_find_test_case(service->sv_id) != NULL) {
 157                 CERROR("Failed to register test %s (%d): rc = %d\n",
 158                        service->sv_name, service->sv_id, -EEXIST);
 159                 return -EEXIST;
 160         }
 161
 162         LIBCFS_ALLOC(tsc, sizeof(*tsc));
 163         if (tsc == NULL)
 164                 return -ENOMEM;
 165
 166         tsc->tsc_cli_ops     = cliops;
 167         tsc->tsc_srv_service = service;
 168
 169         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 170         return 0;
 171 }
 172
 173 static void
 174 sfw_add_session_timer(void)
 175 {
 176         struct sfw_session *sn = sfw_data.fw_session;
 177         struct stt_timer *timer = &sn->sn_timer;
 178
 179         LASSERT(!sfw_data.fw_shuttingdown);
 180
 181         if (sn == NULL || sn->sn_timeout == 0)
 182                 return;
 183
 184         LASSERT(!sn->sn_timer_active);
 185
 186         sn->sn_timer_active = 1;
 187         timer->stt_expires = ktime_get_real_seconds() + sn->sn_timeout;
 188         stt_add_timer(timer);
 189 }
 190
 191 static int
 192 sfw_del_session_timer(void)
 193 {
 194         struct sfw_session *sn = sfw_data.fw_session;
 195
 196         if (sn == NULL || !sn->sn_timer_active)
 197                 return 0;
 198
 199         LASSERT(sn->sn_timeout != 0);
 200
 201         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 202                 sn->sn_timer_active = 0;
 203                 return 0;
 204         }
 205
 206         return -EBUSY; /* racing with sfw_session_expired() */
 207 }
 208
 209 /* called with sfw_data.fw_lock held */
 210 static void
 211 sfw_deactivate_session(void)
 212 __must_hold(&sfw_data.fw_lock)
 213 {
 214         struct sfw_session *sn = sfw_data.fw_session;
 215         int            nactive = 0;
 216         struct sfw_batch *tsb;
 217         struct sfw_test_case *tsc;
 218
 219         if (sn == NULL)
 220                 return;
 221
 222         LASSERT(!sn->sn_timer_active);
 223
 224         sfw_data.fw_session = NULL;
 225         atomic_inc(&sfw_data.fw_nzombies);
 226         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 227
 228         spin_unlock(&sfw_data.fw_lock);
 229
 230         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 231                 srpc_abort_service(tsc->tsc_srv_service);
 232         }
 233
 234         spin_lock(&sfw_data.fw_lock);
 235
 236         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 237                 if (sfw_batch_active(tsb)) {
 238                         nactive++;
 239                         sfw_stop_batch(tsb, 1);
 240                 }
 241         }
 242
 243         if (nactive != 0)
 244                 return; /* wait for active batches to stop */
 245
 246         list_del_init(&sn->sn_list);
 247         spin_unlock(&sfw_data.fw_lock);
 248
 249         sfw_destroy_session(sn);
 250
 251         spin_lock(&sfw_data.fw_lock);
 252 }
 253
 254 static void
 255 sfw_session_expired(void *data)
 256 {
 257         struct sfw_session *sn = data;
 258
 259         spin_lock(&sfw_data.fw_lock);
 260
 261         LASSERT(sn->sn_timer_active);
 262         LASSERT(sn == sfw_data.fw_session);
 263
 264         CWARN("Session expired! sid: %s-%llu, name: %s\n",
 265               libcfs_nidstr(&sn->sn_id.ses_nid),
 266               sn->sn_id.ses_stamp, &sn->sn_name[0]);
 267
 268         sn->sn_timer_active = 0;
 269         sfw_deactivate_session();
 270
 271         spin_unlock(&sfw_data.fw_lock);
 272 }
 273
 274 static inline void
 275 sfw_init_session(struct sfw_session *sn, struct lst_sid sid,
 276                  unsigned features, const char *name)
 277 {
 278         struct stt_timer *timer = &sn->sn_timer;
 279
 280         memset(sn, 0, sizeof(struct sfw_session));
 281         INIT_LIST_HEAD(&sn->sn_list);
 282         INIT_LIST_HEAD(&sn->sn_batches);
 283         refcount_set(&sn->sn_refcount, 1); /* +1 for caller */
 284         atomic_set(&sn->sn_brw_errors, 0);
 285         atomic_set(&sn->sn_ping_errors, 0);
 286         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 287
 288         sn->sn_timer_active = 0;
 289         sn->sn_id.ses_stamp = sid.ses_stamp;
 290         lnet_nid4_to_nid(sid.ses_nid, &sn->sn_id.ses_nid);
 291         sn->sn_features = features;
 292         sn->sn_timeout = session_timeout;
 293         sn->sn_started = ktime_get();
 294
 295         timer->stt_data = sn;
 296         timer->stt_func = sfw_session_expired;
 297         INIT_LIST_HEAD(&timer->stt_list);
 298 }
 299
 300 /* completion handler for incoming framework RPCs */
 301 static void
 302 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 303 {
 304         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 305         int                     status  = rpc->srpc_status;
 306
 307         CDEBUG(D_NET,
 308                "Incoming framework RPC done: service %s, peer %s, status %s:%d\n",
 309                sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 310                swi_state2str(rpc->srpc_wi.swi_state), status);
 311
 312         if (rpc->srpc_bulk != NULL)
 313                 sfw_free_pages(rpc);
 314 }
 315
 316 static void
 317 sfw_client_rpc_fini(struct srpc_client_rpc *rpc)
 318 {
 319         LASSERT(rpc->crpc_bulk.bk_niov == 0);
 320         LASSERT(list_empty(&rpc->crpc_list));
 321         LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
 322
 323         CDEBUG(D_NET, "Outgoing framework RPC done: "
 324                "service %d, peer %s, status %s:%d:%d\n",
 325                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 326                swi_state2str(rpc->crpc_wi.swi_state),
 327                rpc->crpc_aborted, rpc->crpc_status);
 328
 329         spin_lock(&sfw_data.fw_lock);
 330
 331         /* my callers must finish all RPCs before shutting me down */
 332         LASSERT(!sfw_data.fw_shuttingdown);
 333         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 334
 335         spin_unlock(&sfw_data.fw_lock);
 336 }
 337
 338 static struct sfw_batch *
 339 sfw_find_batch(struct lst_bid bid)
 340 {
 341         struct sfw_session *sn = sfw_data.fw_session;
 342         struct sfw_batch *bat;
 343
 344         LASSERT(sn != NULL);
 345
 346         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 347                 if (bat->bat_id.bat_id == bid.bat_id)
 348                         return bat;
 349         }
 350
 351         return NULL;
 352 }
 353
 354 static struct sfw_batch *
 355 sfw_bid2batch(struct lst_bid bid)
 356 {
 357         struct sfw_session *sn = sfw_data.fw_session;
 358         struct sfw_batch *bat;
 359
 360         LASSERT(sn != NULL);
 361
 362         bat = sfw_find_batch(bid);
 363         if (bat != NULL)
 364                 return bat;
 365
 366         LIBCFS_ALLOC(bat, sizeof(*bat));
 367         if (bat == NULL)
 368                 return NULL;
 369
 370         bat->bat_error   = 0;
 371         bat->bat_session = sn;
 372         bat->bat_id      = bid;
 373         atomic_set(&bat->bat_nactive, 0);
 374         INIT_LIST_HEAD(&bat->bat_tests);
 375
 376         list_add_tail(&bat->bat_list, &sn->sn_batches);
 377         return bat;
 378 }
 379
 380 static struct lst_sid get_old_sid(struct sfw_session *sn)
 381 {
 382         struct lst_sid sid = { .ses_nid = LNET_NID_ANY, .ses_stamp = -1 };
 383
 384         if (sn) {
 385                 sid.ses_stamp = sn->sn_id.ses_stamp;
 386                 sid.ses_nid = lnet_nid_to_nid4(&sn->sn_id.ses_nid);
 387         }
 388
 389         return sid;
 390 }
 391
 392 static int
 393 sfw_get_stats(struct srpc_stat_reqst *request, struct srpc_stat_reply *reply)
 394 {
 395         struct sfw_session *sn = sfw_data.fw_session;
 396         struct sfw_counters *cnt = &reply->str_fw;
 397         struct sfw_batch *bat;
 398
 399         reply->str_sid = get_old_sid(sn);
 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_common(&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         cnt->running_ms = ktime_ms_delta(ktime_get(), sn->sn_started);
 417         cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
 418         cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
 419         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 420
 421         cnt->active_batches = 0;
 422         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 423                 if (atomic_read(&bat->bat_nactive) > 0)
 424                         cnt->active_batches++;
 425         }
 426
 427         reply->str_status = 0;
 428         return 0;
 429 }
 430
 431 int
 432 sfw_make_session(struct srpc_mksn_reqst *request, struct srpc_mksn_reply *reply)
 433 {
 434         struct sfw_session *sn = sfw_data.fw_session;
 435         struct srpc_msg *msg = container_of(request, struct srpc_msg,
 436                                             msg_body.mksn_reqst);
 437         int cplen = 0;
 438
 439         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 440                 reply->mksn_sid = get_old_sid(sn);
 441                 reply->mksn_status = EINVAL;
 442                 return 0;
 443         }
 444
 445         if (sn != NULL) {
 446                 reply->mksn_status  = 0;
 447                 reply->mksn_sid = get_old_sid(sn);
 448                 reply->mksn_timeout = sn->sn_timeout;
 449
 450                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 451                         refcount_inc(&sn->sn_refcount);
 452                         return 0;
 453                 }
 454
 455                 if (!request->mksn_force) {
 456                         reply->mksn_status = EBUSY;
 457                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 458                                         sizeof(reply->mksn_name));
 459                         if (cplen >= sizeof(reply->mksn_name))
 460                                 return -E2BIG;
 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 struct srpc_msg::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(*sn));
 478         if (sn == NULL) {
 479                 CERROR("dropping RPC mksn under memory pressure: rc = %d\n",
 480                        -ENOMEM);
 481                 return -ENOMEM;
 482         }
 483
 484         sfw_init_session(sn, request->mksn_sid,
 485                          msg->msg_ses_feats, &request->mksn_name[0]);
 486
 487         spin_lock(&sfw_data.fw_lock);
 488
 489         sfw_deactivate_session();
 490         LASSERT(sfw_data.fw_session == NULL);
 491         sfw_data.fw_session = sn;
 492
 493         spin_unlock(&sfw_data.fw_lock);
 494
 495         reply->mksn_status  = 0;
 496         reply->mksn_sid = get_old_sid(sn);
 497         reply->mksn_timeout = sn->sn_timeout;
 498         return 0;
 499 }
 500
 501 static int
 502 sfw_remove_session(struct srpc_rmsn_reqst *request,
 503                    struct srpc_rmsn_reply *reply)
 504 {
 505         struct sfw_session *sn = sfw_data.fw_session;
 506
 507         reply->rmsn_sid = get_old_sid(sn);
 508
 509         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 510                 reply->rmsn_status = EINVAL;
 511                 return 0;
 512         }
 513
 514         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 515                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 516                 return 0;
 517         }
 518
 519         if (!refcount_dec_and_test(&sn->sn_refcount)) {
 520                 reply->rmsn_status = 0;
 521                 return 0;
 522         }
 523
 524         spin_lock(&sfw_data.fw_lock);
 525         sfw_deactivate_session();
 526         spin_unlock(&sfw_data.fw_lock);
 527
 528         reply->rmsn_status = 0;
 529         reply->rmsn_sid = get_old_sid(NULL);
 530         LASSERT(sfw_data.fw_session == NULL);
 531         return 0;
 532 }
 533
 534 static int
 535 sfw_debug_session(struct srpc_debug_reqst *request,
 536                   struct srpc_debug_reply *reply)
 537 {
 538         struct sfw_session *sn = sfw_data.fw_session;
 539
 540         if (sn == NULL) {
 541                 reply->dbg_status = ESRCH;
 542                 reply->dbg_sid = get_old_sid(NULL);
 543                 return 0;
 544         }
 545
 546         reply->dbg_status  = 0;
 547         reply->dbg_sid = get_old_sid(sn);
 548         reply->dbg_timeout = sn->sn_timeout;
 549         if (strlcpy(reply->dbg_name, &sn->sn_name[0],
 550                     sizeof(reply->dbg_name)) >= sizeof(reply->dbg_name))
 551                 return -E2BIG;
 552
 553         return 0;
 554 }
 555
 556 static void
 557 sfw_test_rpc_fini(struct srpc_client_rpc *rpc)
 558 {
 559         struct sfw_test_unit *tsu = rpc->crpc_priv;
 560         struct sfw_test_instance *tsi = tsu->tsu_instance;
 561
 562         /* Called with hold of tsi->tsi_lock */
 563         LASSERT(list_empty(&rpc->crpc_list));
 564         rpc->crpc_wi.swi_state = SWI_STATE_DONE;
 565         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 566 }
 567
 568 static inline int
 569 sfw_test_buffers(struct sfw_test_instance *tsi)
 570 {
 571         struct sfw_test_case    *tsc;
 572         struct srpc_service     *svc;
 573         int                     nbuf;
 574
 575         LASSERT(tsi != NULL);
 576         tsc = sfw_find_test_case(tsi->tsi_service);
 577         LASSERT(tsc != NULL);
 578         svc = tsc->tsc_srv_service;
 579         LASSERT(svc != NULL);
 580
 581         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 582         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 583 }
 584
 585 static int
 586 sfw_load_test(struct sfw_test_instance *tsi)
 587 {
 588         struct sfw_test_case    *tsc;
 589         struct srpc_service     *svc;
 590         int                     nbuf;
 591         int                     rc;
 592
 593         LASSERT(tsi != NULL);
 594         tsc = sfw_find_test_case(tsi->tsi_service);
 595         nbuf = sfw_test_buffers(tsi);
 596         LASSERT(tsc != NULL);
 597         svc = tsc->tsc_srv_service;
 598
 599         if (tsi->tsi_is_client) {
 600                 tsi->tsi_ops = tsc->tsc_cli_ops;
 601                 return 0;
 602         }
 603
 604         rc = srpc_service_add_buffers(svc, nbuf);
 605         if (rc != 0) {
 606                 CWARN("Failed to reserve enough buffers: "
 607                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 608                 /* NB: this error handler is not strictly correct, because
 609                  * it may release more buffers than already allocated,
 610                  * but it doesn't matter because request portal should
 611                  * be lazy portal and will grow buffers if necessary. */
 612                 srpc_service_remove_buffers(svc, nbuf);
 613                 return -ENOMEM;
 614         }
 615
 616         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 617                nbuf * (srpc_serv_is_framework(svc) ?
 618                        1 : cfs_cpt_number(cfs_cpt_tab)), svc->sv_name);
 619         return 0;
 620 }
 621
 622 static void
 623 sfw_unload_test(struct sfw_test_instance *tsi)
 624 {
 625         struct sfw_test_case *tsc;
 626
 627         LASSERT(tsi != NULL);
 628         tsc = sfw_find_test_case(tsi->tsi_service);
 629         LASSERT(tsc != NULL);
 630
 631         if (tsi->tsi_is_client)
 632                 return;
 633
 634         /* shrink buffers, because request portal is lazy portal
 635          * which can grow buffers at runtime so we may leave
 636          * some buffers behind, but never mind... */
 637         srpc_service_remove_buffers(tsc->tsc_srv_service,
 638                                     sfw_test_buffers(tsi));
 639 }
 640
 641 static void
 642 sfw_destroy_test_instance(struct sfw_test_instance *tsi)
 643 {
 644         struct srpc_client_rpc *rpc;
 645         struct sfw_test_unit *tsu;
 646
 647         if (!tsi->tsi_is_client)
 648                 goto clean;
 649
 650         tsi->tsi_ops->tso_fini(tsi);
 651
 652         LASSERT(!tsi->tsi_stopping);
 653         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 654         LASSERT(!sfw_test_active(tsi));
 655
 656         while (!list_empty(&tsi->tsi_units)) {
 657                 tsu = list_first_entry(&tsi->tsi_units,
 658                                        struct sfw_test_unit, tsu_list);
 659                 list_del(&tsu->tsu_list);
 660                 LIBCFS_FREE(tsu, sizeof(*tsu));
 661         }
 662
 663         while (!list_empty(&tsi->tsi_free_rpcs)) {
 664                 rpc = list_first_entry(&tsi->tsi_free_rpcs,
 665                                        struct srpc_client_rpc, crpc_list);
 666                 list_del(&rpc->crpc_list);
 667                 swi_cancel_workitem(&rpc->crpc_wi);
 668                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 669         }
 670
 671 clean:
 672         sfw_unload_test(tsi);
 673         LIBCFS_FREE(tsi, sizeof(*tsi));
 674 }
 675
 676 static void
 677 sfw_destroy_batch(struct sfw_batch *tsb)
 678 {
 679         struct sfw_test_instance *tsi;
 680
 681         LASSERT(!sfw_batch_active(tsb));
 682         LASSERT(list_empty(&tsb->bat_list));
 683
 684         while (!list_empty(&tsb->bat_tests)) {
 685                 tsi = list_first_entry(&tsb->bat_tests,
 686                                        struct sfw_test_instance, tsi_list);
 687                 list_del_init(&tsi->tsi_list);
 688                 sfw_destroy_test_instance(tsi);
 689         }
 690
 691         LIBCFS_FREE(tsb, sizeof(*tsb));
 692 }
 693
 694 static void
 695 sfw_destroy_session(struct sfw_session *sn)
 696 {
 697         struct sfw_batch *batch;
 698
 699         LASSERT(list_empty(&sn->sn_list));
 700         LASSERT(sn != sfw_data.fw_session);
 701
 702         while (!list_empty(&sn->sn_batches)) {
 703                 batch = list_first_entry(&sn->sn_batches,
 704                                          struct sfw_batch, bat_list);
 705                 list_del_init(&batch->bat_list);
 706                 sfw_destroy_batch(batch);
 707         }
 708
 709         LIBCFS_FREE(sn, sizeof(*sn));
 710         atomic_dec(&sfw_data.fw_nzombies);
 711 }
 712
 713 static void
 714 sfw_unpack_addtest_req(struct srpc_msg *msg)
 715 {
 716         struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
 717
 718         LASSERT(msg->msg_type == SRPC_MSG_TEST_REQST);
 719         LASSERT(req->tsr_is_client);
 720
 721         if (msg->msg_magic == SRPC_MSG_MAGIC)
 722                 return; /* no flipping needed */
 723
 724         LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 725
 726         if (req->tsr_service == SRPC_SERVICE_BRW) {
 727                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 728                         struct test_bulk_req *bulk = &req->tsr_u.bulk_v0;
 729
 730                         __swab32s(&bulk->blk_opc);
 731                         __swab32s(&bulk->blk_npg);
 732                         __swab32s(&bulk->blk_flags);
 733
 734                 } else {
 735                         struct test_bulk_req_v1 *bulk = &req->tsr_u.bulk_v1;
 736
 737                         __swab16s(&bulk->blk_opc);
 738                         __swab16s(&bulk->blk_flags);
 739                         __swab32s(&bulk->blk_offset);
 740                         __swab32s(&bulk->blk_len);
 741                 }
 742
 743                 return;
 744         }
 745
 746         if (req->tsr_service == SRPC_SERVICE_PING) {
 747                 struct test_ping_req *ping = &req->tsr_u.ping;
 748
 749                 __swab32s(&ping->png_size);
 750                 __swab32s(&ping->png_flags);
 751                 return;
 752         }
 753
 754         LBUG();
 755 }
 756
 757 static int
 758 sfw_add_test_instance(struct sfw_batch *tsb, struct srpc_server_rpc *rpc)
 759 {
 760         struct srpc_msg *msg = &rpc->srpc_reqstbuf->buf_msg;
 761         struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
 762         struct srpc_bulk *bk = rpc->srpc_bulk;
 763         int ndest = req->tsr_ndest;
 764         struct sfw_test_unit *tsu;
 765         struct sfw_test_instance *tsi;
 766         int i;
 767         int rc;
 768
 769         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 770         if (tsi == NULL) {
 771                 CERROR("Can't allocate test instance for batch: %llu: rc = %d\n",
 772                        tsb->bat_id.bat_id, -ENOMEM);
 773                 return -ENOMEM;
 774         }
 775
 776         spin_lock_init(&tsi->tsi_lock);
 777         atomic_set(&tsi->tsi_nactive, 0);
 778         INIT_LIST_HEAD(&tsi->tsi_units);
 779         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 780         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 781
 782         tsi->tsi_stopping      = 0;
 783         tsi->tsi_batch         = tsb;
 784         tsi->tsi_loop          = req->tsr_loop;
 785         tsi->tsi_concur        = req->tsr_concur;
 786         tsi->tsi_service       = req->tsr_service;
 787         tsi->tsi_is_client     = !!(req->tsr_is_client);
 788         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 789
 790         rc = sfw_load_test(tsi);
 791         if (rc != 0) {
 792                 LIBCFS_FREE(tsi, sizeof(*tsi));
 793                 return rc;
 794         }
 795
 796         LASSERT(!sfw_batch_active(tsb));
 797
 798         if (!tsi->tsi_is_client) {
 799                 /* it's test server, just add it to tsb */
 800                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 801                 return 0;
 802         }
 803
 804         LASSERT(bk != NULL);
 805         LASSERT(bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 806         LASSERT((unsigned int)bk->bk_len >=
 807                 sizeof(struct lnet_process_id_packed) * ndest);
 808
 809         sfw_unpack_addtest_req(msg);
 810         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 811
 812         for (i = 0; i < ndest; i++) {
 813                 struct lnet_process_id_packed *dests;
 814                 struct lnet_process_id_packed  id;
 815                 int j;
 816
 817                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].bv_page);
 818                 LASSERT(dests != NULL);  /* my pages are within KVM always */
 819                 id = dests[i % SFW_ID_PER_PAGE];
 820                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 821                         sfw_unpack_id(id);
 822
 823                 for (j = 0; j < tsi->tsi_concur; j++) {
 824                         LIBCFS_ALLOC(tsu, sizeof(*tsu));
 825                         if (tsu == NULL) {
 826                                 rc = -ENOMEM;
 827                                 CERROR("Can't allocate tsu for %d: rc = %d\n",
 828                                        tsi->tsi_service, rc);
 829                                 goto error;
 830                         }
 831
 832                         tsu->tsu_dest.nid = id.nid;
 833                         tsu->tsu_dest.pid = id.pid;
 834                         tsu->tsu_instance = tsi;
 835                         tsu->tsu_private  = NULL;
 836                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 837                 }
 838         }
 839
 840         rc = tsi->tsi_ops->tso_init(tsi);
 841         if (rc == 0) {
 842                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 843                 return 0;
 844         }
 845
 846 error:
 847         LASSERT(rc != 0);
 848         sfw_destroy_test_instance(tsi);
 849         return rc;
 850 }
 851
 852 static void
 853 sfw_test_unit_done(struct sfw_test_unit *tsu)
 854 {
 855         struct sfw_test_instance *tsi = tsu->tsu_instance;
 856         struct sfw_batch *tsb = tsi->tsi_batch;
 857         struct sfw_session *sn = tsb->bat_session;
 858
 859         LASSERT(sfw_test_active(tsi));
 860
 861         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 862                 return;
 863
 864         /* the test instance is done */
 865         spin_lock(&tsi->tsi_lock);
 866
 867         tsi->tsi_stopping = 0;
 868
 869         spin_unlock(&tsi->tsi_lock);
 870
 871         spin_lock(&sfw_data.fw_lock);
 872
 873         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 874             sn == sfw_data.fw_session) {                  /* sn also active */
 875                 spin_unlock(&sfw_data.fw_lock);
 876                 return;
 877         }
 878
 879         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 880
 881         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 882                 if (sfw_batch_active(tsb)) {
 883                         spin_unlock(&sfw_data.fw_lock);
 884                         return;
 885                 }
 886         }
 887
 888         list_del_init(&sn->sn_list);
 889         spin_unlock(&sfw_data.fw_lock);
 890
 891         sfw_destroy_session(sn);
 892 }
 893
 894 static void
 895 sfw_test_rpc_done(struct srpc_client_rpc *rpc)
 896 {
 897         struct sfw_test_unit *tsu = rpc->crpc_priv;
 898         struct sfw_test_instance *tsi = tsu->tsu_instance;
 899         int done = 0;
 900
 901         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 902
 903         spin_lock(&tsi->tsi_lock);
 904
 905         LASSERT(sfw_test_active(tsi));
 906         LASSERT(!list_empty(&rpc->crpc_list));
 907
 908         list_del_init(&rpc->crpc_list);
 909
 910         /* batch is stopping or loop is done or get error */
 911         if (tsi->tsi_stopping || 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 }
 927
 928 int
 929 sfw_create_test_rpc(struct sfw_test_unit *tsu, struct lnet_process_id peer,
 930                     unsigned features, int nblk, int blklen,
 931                     struct srpc_client_rpc **rpcpp)
 932 {
 933         struct srpc_client_rpc *rpc = NULL;
 934         struct sfw_test_instance *tsi = tsu->tsu_instance;
 935
 936         spin_lock(&tsi->tsi_lock);
 937
 938         LASSERT(sfw_test_active(tsi));
 939
 940         if (!list_empty(&tsi->tsi_free_rpcs)) {
 941                 /* pick request from buffer */
 942                 rpc = list_first_entry(&tsi->tsi_free_rpcs,
 943                                        struct srpc_client_rpc, crpc_list);
 944                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 945                 list_del_init(&rpc->crpc_list);
 946         }
 947
 948         spin_unlock(&tsi->tsi_lock);
 949
 950         if (rpc == NULL) {
 951                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 952                                              blklen, sfw_test_rpc_done,
 953                                              sfw_test_rpc_fini, tsu);
 954         } else {
 955                 swi_cancel_workitem(&rpc->crpc_wi);
 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: rc = %d\n",
 963                        tsi->tsi_service, -ENOMEM);
 964                 return -ENOMEM;
 965         }
 966
 967         rpc->crpc_reqstmsg.msg_ses_feats = features;
 968         *rpcpp = rpc;
 969
 970         return 0;
 971 }
 972
 973 static void
 974 sfw_run_test(struct swi_workitem *wi)
 975 {
 976         struct sfw_test_unit *tsu = container_of(wi, struct sfw_test_unit,
 977                                                  tsu_worker);
 978         struct sfw_test_instance *tsi = tsu->tsu_instance;
 979         struct srpc_client_rpc *rpc = NULL;
 980
 981         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 982                 LASSERT(rpc == NULL);
 983                 wi->swi_state = SWI_STATE_DONE;
 984                 goto test_done;
 985         }
 986
 987         LASSERT(rpc != NULL);
 988
 989         spin_lock(&tsi->tsi_lock);
 990         if (wi->swi_state == SWI_STATE_DONE) {
 991                 spin_unlock(&tsi->tsi_lock);
 992                 return;
 993         }
 994
 995         if (tsi->tsi_stopping) {
 996                 wi->swi_state = SWI_STATE_DONE;
 997                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 998                 spin_unlock(&tsi->tsi_lock);
 999                 goto test_done;
1000         }
1001
1002         if (tsu->tsu_loop > 0)
1003                 tsu->tsu_loop--;
1004
1005         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
1006         wi->swi_state = SWI_STATE_RUNNING;
1007         spin_unlock(&tsi->tsi_lock);
1008
1009         spin_lock(&rpc->crpc_lock);
1010         rpc->crpc_timeout = rpc_timeout;
1011         srpc_post_rpc(rpc);
1012         spin_unlock(&rpc->crpc_lock);
1013         return;
1014
1015 test_done:
1016         /*
1017          * No one can schedule me now since:
1018          * - previous RPC, if any, has done and
1019          * - no new RPC is initiated.
1020          * - my batch is still active; no one can run it again now.
1021          * Cancel pending schedules and prevent future schedule attempts:
1022          */
1023         sfw_test_unit_done(tsu);
1024 }
1025
1026 static int
1027 sfw_run_batch(struct sfw_batch *tsb)
1028 {
1029         struct swi_workitem *wi;
1030         struct sfw_test_unit *tsu;
1031         struct sfw_test_instance *tsi;
1032
1033         if (sfw_batch_active(tsb)) {
1034                 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1035                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1036                 return 0;
1037         }
1038
1039         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1040                 if (!tsi->tsi_is_client)        /* skip server instances */
1041                         continue;
1042
1043                 LASSERT(!tsi->tsi_stopping);
1044                 LASSERT(!sfw_test_active(tsi));
1045
1046                 atomic_inc(&tsb->bat_nactive);
1047
1048                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1049                         atomic_inc(&tsi->tsi_nactive);
1050                         tsu->tsu_loop = tsi->tsi_loop;
1051                         wi = &tsu->tsu_worker;
1052                         swi_init_workitem(wi, sfw_run_test,
1053                                           lst_test_wq[lnet_cpt_of_nid(
1054                                                               tsu->tsu_dest.nid,
1055                                                               NULL)]);
1056                         swi_schedule_workitem(wi);
1057                 }
1058         }
1059
1060         return 0;
1061 }
1062
1063 static int
1064 sfw_stop_batch(struct sfw_batch *tsb, int force)
1065 {
1066         struct sfw_test_instance *tsi;
1067         struct srpc_client_rpc *rpc;
1068
1069         if (!sfw_batch_active(tsb)) {
1070                 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1071                 return 0;
1072         }
1073
1074         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1075                 spin_lock(&tsi->tsi_lock);
1076
1077                 if (!tsi->tsi_is_client ||
1078                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1079                         spin_unlock(&tsi->tsi_lock);
1080                         continue;
1081                 }
1082
1083                 tsi->tsi_stopping = 1;
1084
1085                 if (!force) {
1086                         spin_unlock(&tsi->tsi_lock);
1087                         continue;
1088                 }
1089
1090                 /* abort launched rpcs in the test */
1091                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1092                         spin_lock(&rpc->crpc_lock);
1093
1094                         srpc_abort_rpc(rpc, -EINTR);
1095
1096                         spin_unlock(&rpc->crpc_lock);
1097                 }
1098
1099                 spin_unlock(&tsi->tsi_lock);
1100         }
1101
1102         return 0;
1103 }
1104
1105 static int
1106 sfw_query_batch(struct sfw_batch *tsb, int testidx,
1107                 struct srpc_batch_reply *reply)
1108 {
1109         struct sfw_test_instance *tsi;
1110
1111         if (testidx < 0)
1112                 return -EINVAL;
1113
1114         if (testidx == 0) {
1115                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1116                 return 0;
1117         }
1118
1119         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1120                 if (testidx-- > 1)
1121                         continue;
1122
1123                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1124                 return 0;
1125         }
1126
1127         return -ENOENT;
1128 }
1129
1130 void
1131 sfw_free_pages(struct srpc_server_rpc *rpc)
1132 {
1133         srpc_free_bulk(rpc->srpc_bulk);
1134         rpc->srpc_bulk = NULL;
1135 }
1136
1137 int
1138 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1139                 int sink)
1140 {
1141         LASSERT(rpc->srpc_bulk == NULL);
1142         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1143
1144         rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1145         if (rpc->srpc_bulk == NULL)
1146                 return -ENOMEM;
1147
1148         return 0;
1149 }
1150
1151 static int
1152 sfw_add_test(struct srpc_server_rpc *rpc)
1153 {
1154         struct sfw_session *sn = sfw_data.fw_session;
1155         struct srpc_test_reply *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1156         struct srpc_test_reqst *request;
1157         int rc;
1158         struct sfw_batch *bat;
1159
1160         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1161         reply->tsr_sid = get_old_sid(sn);
1162
1163         if (request->tsr_loop == 0 || 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: rc = %d\n",
1183                         rpc->srpc_scd->scd_svc->sv_name,
1184                         libcfs_id2str(rpc->srpc_peer), -ENOMEM);
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 * PAGE_SIZE;
1200
1201                 } else  {
1202                         len = sizeof(struct lnet_process_id_packed) *
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 static int
1221 sfw_control_batch(struct srpc_batch_reqst *request,
1222                   struct srpc_batch_reply *reply)
1223 {
1224         struct sfw_session *sn = sfw_data.fw_session;
1225         struct sfw_batch *bat;
1226         int rc = 0;
1227
1228         reply->bar_sid = get_old_sid(sn);
1229
1230         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1231                 reply->bar_status = ESRCH;
1232                 return 0;
1233         }
1234
1235         bat = sfw_find_batch(request->bar_bid);
1236         if (bat == NULL) {
1237                 reply->bar_status = ENOENT;
1238                 return 0;
1239         }
1240
1241         switch (request->bar_opc) {
1242         case SRPC_BATCH_OPC_RUN:
1243                 rc = sfw_run_batch(bat);
1244                 break;
1245
1246         case SRPC_BATCH_OPC_STOP:
1247                 rc = sfw_stop_batch(bat, request->bar_arg);
1248                 break;
1249
1250         case SRPC_BATCH_OPC_QUERY:
1251                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1252                 break;
1253
1254         default:
1255                 return -EINVAL; /* drop it */
1256         }
1257
1258         reply->bar_status = (rc < 0) ? -rc : rc;
1259         return 0;
1260 }
1261
1262 static int
1263 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1264 {
1265         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1266         struct srpc_msg     *reply      = &rpc->srpc_replymsg;
1267         struct srpc_msg     *request    = &rpc->srpc_reqstbuf->buf_msg;
1268         unsigned        features = LST_FEATS_MASK;
1269         int             rc = 0;
1270
1271         LASSERT(sfw_data.fw_active_srpc == NULL);
1272         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1273
1274         spin_lock(&sfw_data.fw_lock);
1275
1276         if (sfw_data.fw_shuttingdown) {
1277                 spin_unlock(&sfw_data.fw_lock);
1278                 return -ESHUTDOWN;
1279         }
1280
1281         /* Remove timer to avoid racing with it or expiring active session */
1282         if (sfw_del_session_timer() != 0) {
1283                 CERROR("dropping RPC %s from %s: racing with expiry timer: rc = %d\n",
1284                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), -EAGAIN);
1285                 spin_unlock(&sfw_data.fw_lock);
1286                 return -EAGAIN;
1287         }
1288
1289         sfw_data.fw_active_srpc = rpc;
1290         spin_unlock(&sfw_data.fw_lock);
1291
1292         sfw_unpack_message(request);
1293         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1294
1295         /* rpc module should have checked this */
1296         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1297
1298         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1299             sv->sv_id != SRPC_SERVICE_DEBUG) {
1300                 struct sfw_session *sn = sfw_data.fw_session;
1301
1302                 if (sn != NULL &&
1303                     sn->sn_features != request->msg_ses_feats) {
1304                         CNETERR("Features of framework RPC don't match 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.ses_stamp = sn->sn_id.ses_stamp;
1308                         reply->msg_body.reply.sid.ses_nid =
1309                                 lnet_nid_to_nid4(&sn->sn_id.ses_nid);
1310                         goto out;
1311                 }
1312
1313         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1314                 /* NB: at this point, old version will ignore features and
1315                  * create new session anyway, so console should be able
1316                  * to handle this */
1317                 reply->msg_body.reply.status = EPROTO;
1318                 goto out;
1319         }
1320
1321         switch (sv->sv_id) {
1322         default:
1323                 LBUG();
1324         case SRPC_SERVICE_TEST:
1325                 rc = sfw_add_test(rpc);
1326                 break;
1327
1328         case SRPC_SERVICE_BATCH:
1329                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1330                                        &reply->msg_body.bat_reply);
1331                 break;
1332
1333         case SRPC_SERVICE_QUERY_STAT:
1334                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1335                                    &reply->msg_body.stat_reply);
1336                 break;
1337
1338         case SRPC_SERVICE_DEBUG:
1339                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1340                                        &reply->msg_body.dbg_reply);
1341                 break;
1342
1343         case SRPC_SERVICE_MAKE_SESSION:
1344                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1345                                       &reply->msg_body.mksn_reply);
1346                 break;
1347
1348         case SRPC_SERVICE_REMOVE_SESSION:
1349                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1350                                         &reply->msg_body.rmsn_reply);
1351                 break;
1352         }
1353
1354         if (sfw_data.fw_session != NULL)
1355                 features = sfw_data.fw_session->sn_features;
1356  out:
1357         reply->msg_ses_feats = features;
1358         rpc->srpc_done = sfw_server_rpc_done;
1359         spin_lock(&sfw_data.fw_lock);
1360
1361         if (!sfw_data.fw_shuttingdown)
1362                 sfw_add_session_timer();
1363
1364         sfw_data.fw_active_srpc = NULL;
1365         spin_unlock(&sfw_data.fw_lock);
1366         return rc;
1367 }
1368
1369 static int
1370 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1371 {
1372         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1373         int                     rc;
1374
1375         LASSERT(rpc->srpc_bulk != NULL);
1376         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1377         LASSERT(sfw_data.fw_active_srpc == NULL);
1378         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1379
1380         spin_lock(&sfw_data.fw_lock);
1381
1382         if (status != 0) {
1383                 CERROR("Bulk transfer failed for RPC: service %s, peer %s, status %d: rc = %d\n",
1384                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status, -EIO);
1385                 spin_unlock(&sfw_data.fw_lock);
1386                 return -EIO;
1387         }
1388
1389         if (sfw_data.fw_shuttingdown) {
1390                 spin_unlock(&sfw_data.fw_lock);
1391                 return -ESHUTDOWN;
1392         }
1393
1394         if (sfw_del_session_timer() != 0) {
1395                 CERROR("dropping RPC %s from %s: racing with expiry timer: rc = %d\n",
1396                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), -EAGAIN);
1397                 spin_unlock(&sfw_data.fw_lock);
1398                 return -EAGAIN;
1399         }
1400
1401         sfw_data.fw_active_srpc = rpc;
1402         spin_unlock(&sfw_data.fw_lock);
1403
1404         rc = sfw_add_test(rpc);
1405
1406         spin_lock(&sfw_data.fw_lock);
1407
1408         if (!sfw_data.fw_shuttingdown)
1409                 sfw_add_session_timer();
1410
1411         sfw_data.fw_active_srpc = NULL;
1412         spin_unlock(&sfw_data.fw_lock);
1413         return rc;
1414 }
1415
1416 struct srpc_client_rpc *
1417 sfw_create_rpc(struct lnet_process_id peer, int service,
1418                unsigned features, int nbulkiov, int bulklen,
1419                void (*done)(struct srpc_client_rpc *), void *priv)
1420 {
1421         struct srpc_client_rpc *rpc = NULL;
1422
1423         spin_lock(&sfw_data.fw_lock);
1424
1425         LASSERT(!sfw_data.fw_shuttingdown);
1426         LASSERT(service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1427
1428         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1429                 rpc = list_first_entry(&sfw_data.fw_zombie_rpcs,
1430                                        struct srpc_client_rpc, crpc_list);
1431                 list_del(&rpc->crpc_list);
1432         }
1433         spin_unlock(&sfw_data.fw_lock);
1434
1435         if (rpc) {
1436                 /* Ensure that rpc is done */
1437                 swi_cancel_workitem(&rpc->crpc_wi);
1438                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1439                                      done, sfw_client_rpc_fini, priv);
1440         } else {
1441                 rpc = srpc_create_client_rpc(peer, service,
1442                                              nbulkiov, bulklen, done,
1443                                              nbulkiov != 0 ?  NULL :
1444                                              sfw_client_rpc_fini,
1445                                              priv);
1446         }
1447
1448         if (rpc != NULL) /* "session" is concept in framework */
1449                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1450
1451         return rpc;
1452 }
1453
1454 void
1455 sfw_unpack_message(struct srpc_msg *msg)
1456 {
1457         if (msg->msg_magic == SRPC_MSG_MAGIC)
1458                 return; /* no flipping needed */
1459
1460         /* srpc module should guarantee I wouldn't get crap */
1461         LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1462
1463         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1464                 struct srpc_stat_reqst *req = &msg->msg_body.stat_reqst;
1465
1466                 __swab32s(&req->str_type);
1467                 __swab64s(&req->str_rpyid);
1468                 sfw_unpack_sid(req->str_sid);
1469                 return;
1470         }
1471
1472         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1473                 struct srpc_stat_reply *rep = &msg->msg_body.stat_reply;
1474
1475                 __swab32s(&rep->str_status);
1476                 sfw_unpack_sid(rep->str_sid);
1477                 sfw_unpack_fw_counters(rep->str_fw);
1478                 sfw_unpack_rpc_counters(rep->str_rpc);
1479                 sfw_unpack_lnet_counters(rep->str_lnet);
1480                 return;
1481         }
1482
1483         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1484                 struct srpc_mksn_reqst *req = &msg->msg_body.mksn_reqst;
1485
1486                 __swab64s(&req->mksn_rpyid);
1487                 __swab32s(&req->mksn_force);
1488                 sfw_unpack_sid(req->mksn_sid);
1489                 return;
1490         }
1491
1492         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1493                 struct srpc_mksn_reply *rep = &msg->msg_body.mksn_reply;
1494
1495                 __swab32s(&rep->mksn_status);
1496                 __swab32s(&rep->mksn_timeout);
1497                 sfw_unpack_sid(rep->mksn_sid);
1498                 return;
1499         }
1500
1501         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1502                 struct srpc_rmsn_reqst *req = &msg->msg_body.rmsn_reqst;
1503
1504                 __swab64s(&req->rmsn_rpyid);
1505                 sfw_unpack_sid(req->rmsn_sid);
1506                 return;
1507         }
1508
1509         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1510                 struct srpc_rmsn_reply *rep = &msg->msg_body.rmsn_reply;
1511
1512                 __swab32s(&rep->rmsn_status);
1513                 sfw_unpack_sid(rep->rmsn_sid);
1514                 return;
1515         }
1516
1517         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1518                 struct srpc_debug_reqst *req = &msg->msg_body.dbg_reqst;
1519
1520                 __swab64s(&req->dbg_rpyid);
1521                 __swab32s(&req->dbg_flags);
1522                 sfw_unpack_sid(req->dbg_sid);
1523                 return;
1524         }
1525
1526         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1527                 struct srpc_debug_reply *rep = &msg->msg_body.dbg_reply;
1528
1529                 __swab32s(&rep->dbg_nbatch);
1530                 __swab32s(&rep->dbg_timeout);
1531                 sfw_unpack_sid(rep->dbg_sid);
1532                 return;
1533         }
1534
1535         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1536                 struct srpc_batch_reqst *req = &msg->msg_body.bat_reqst;
1537
1538                 __swab32s(&req->bar_opc);
1539                 __swab64s(&req->bar_rpyid);
1540                 __swab32s(&req->bar_testidx);
1541                 __swab32s(&req->bar_arg);
1542                 sfw_unpack_sid(req->bar_sid);
1543                 __swab64s(&req->bar_bid.bat_id);
1544                 return;
1545         }
1546
1547         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1548                 struct srpc_batch_reply *rep = &msg->msg_body.bat_reply;
1549
1550                 __swab32s(&rep->bar_status);
1551                 sfw_unpack_sid(rep->bar_sid);
1552                 return;
1553         }
1554
1555         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1556                 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
1557
1558                 __swab64s(&req->tsr_rpyid);
1559                 __swab64s(&req->tsr_bulkid);
1560                 __swab32s(&req->tsr_loop);
1561                 __swab32s(&req->tsr_ndest);
1562                 __swab32s(&req->tsr_concur);
1563                 __swab32s(&req->tsr_service);
1564                 sfw_unpack_sid(req->tsr_sid);
1565                 __swab64s(&req->tsr_bid.bat_id);
1566                 return;
1567         }
1568
1569         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1570                 struct srpc_test_reply *rep = &msg->msg_body.tes_reply;
1571
1572                 __swab32s(&rep->tsr_status);
1573                 sfw_unpack_sid(rep->tsr_sid);
1574                 return;
1575         }
1576
1577         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1578                 struct srpc_join_reqst *req = &msg->msg_body.join_reqst;
1579
1580                 __swab64s(&req->join_rpyid);
1581                 sfw_unpack_sid(req->join_sid);
1582                 return;
1583         }
1584
1585         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1586                 struct srpc_join_reply *rep = &msg->msg_body.join_reply;
1587
1588                 __swab32s(&rep->join_status);
1589                 __swab32s(&rep->join_timeout);
1590                 sfw_unpack_sid(rep->join_sid);
1591                 return;
1592         }
1593
1594         LBUG();
1595 }
1596
1597 void
1598 sfw_abort_rpc(struct srpc_client_rpc *rpc)
1599 {
1600         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1601         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1602
1603         spin_lock(&rpc->crpc_lock);
1604         srpc_abort_rpc(rpc, -EINTR);
1605         spin_unlock(&rpc->crpc_lock);
1606 }
1607
1608 void
1609 sfw_post_rpc(struct srpc_client_rpc *rpc)
1610 {
1611         spin_lock(&rpc->crpc_lock);
1612
1613         LASSERT(!rpc->crpc_closed);
1614         LASSERT(!rpc->crpc_aborted);
1615         LASSERT(list_empty(&rpc->crpc_list));
1616         LASSERT(!sfw_data.fw_shuttingdown);
1617
1618         rpc->crpc_timeout = rpc_timeout;
1619         srpc_post_rpc(rpc);
1620
1621         spin_unlock(&rpc->crpc_lock);
1622 }
1623
1624 #if 0
1625 static struct srpc_service sfw_services[] = {
1626         { .sv_id = SRPC_SERVICE_DEBUG,          .sv_name = "debug", },
1627         { .sv_id = SRPC_SERVICE_QUERY_STAT,     .sv_name = "query stats", },
1628         { .sv_id = SRPC_SERVICE_MAKE_SESSION,   .sv_name = "make session", },
1629         { .sv_id = SRPC_SERVICE_REMOVE_SESSION, .sv_name = "remove session", },
1630         { .sv_id = SRPC_SERVICE_BATCH,          .sv_name = "batch service", },
1631         { .sv_id = SRPC_SERVICE_TEST,           .sv_name = "test service", },
1632         { .sv_id = 0, } };
1633 #endif
1634
1635 int
1636 sfw_startup(void)
1637 {
1638         struct srpc_service *sv;
1639         struct sfw_test_case *tsc;
1640         int i, rc, error;
1641
1642         if (session_timeout < 0) {
1643                 CERROR("Session timeout must be non-negative: %d: rc = %d\n",
1644                        session_timeout, -EINVAL);
1645                 return -EINVAL;
1646         }
1647
1648         if (rpc_timeout < 0) {
1649                 CERROR("RPC timeout must be non-negative: %d: rc = %d\n",
1650                        rpc_timeout, -EINVAL);
1651                 return -EINVAL;
1652         }
1653
1654         if (session_timeout == 0)
1655                 CWARN("Zero session_timeout specified - test sessions never expire.\n");
1656
1657         if (rpc_timeout == 0)
1658                 CWARN("Zero rpc_timeout specified - test RPC never expire.\n");
1659
1660         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1661
1662         sfw_data.fw_session     = NULL;
1663         sfw_data.fw_active_srpc = NULL;
1664         spin_lock_init(&sfw_data.fw_lock);
1665         atomic_set(&sfw_data.fw_nzombies, 0);
1666         INIT_LIST_HEAD(&sfw_data.fw_tests);
1667         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1668         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1669
1670         brw_init_test_client();
1671         brw_init_test_service();
1672         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1673         LASSERT(rc == 0);
1674
1675         ping_init_test_client();
1676         ping_init_test_service();
1677         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1678         LASSERT(rc == 0);
1679
1680         error = 0;
1681         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1682                 sv = tsc->tsc_srv_service;
1683
1684                 rc = srpc_add_service(sv);
1685                 LASSERT(rc != -EBUSY);
1686                 if (rc != 0) {
1687                         CWARN("Failed to add %s service: %d\n",
1688                               sv->sv_name, rc);
1689                         error = rc;
1690                 }
1691         }
1692
1693         for (i = 0; ; i++) {
1694                 sv = &sfw_services[i];
1695                 if (sv->sv_name == NULL)
1696                         break;
1697
1698                 sv->sv_bulk_ready = NULL;
1699                 sv->sv_handler    = sfw_handle_server_rpc;
1700                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1701                 if (sv->sv_id == SRPC_SERVICE_TEST)
1702                         sv->sv_bulk_ready = sfw_bulk_ready;
1703
1704                 rc = srpc_add_service(sv);
1705                 LASSERT(rc != -EBUSY);
1706                 if (rc != 0) {
1707                         CWARN("Failed to add %s service: %d\n", sv->sv_name,
1708                               rc);
1709                         error = rc;
1710                 }
1711
1712                 /* about to sfw_shutdown, no need to add buffer */
1713                 if (error)
1714                         continue;
1715
1716                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1717                 if (rc != 0) {
1718                         CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
1719                               sv->sv_name, sv->sv_wi_total, rc);
1720                         error = -ENOMEM;
1721                 }
1722         }
1723
1724         if (error != 0)
1725                 sfw_shutdown();
1726         return error;
1727 }
1728
1729 void
1730 sfw_shutdown(void)
1731 {
1732         struct srpc_service *sv;
1733         struct sfw_test_case *tsc;
1734         int i;
1735
1736         spin_lock(&sfw_data.fw_lock);
1737
1738         sfw_data.fw_shuttingdown = 1;
1739         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1740                        "waiting for active RPC to finish.\n");
1741
1742         if (sfw_del_session_timer() != 0)
1743                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1744                                "waiting for session timer to explode.\n");
1745
1746         sfw_deactivate_session();
1747         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1748                        sfw_data.fw_lock,
1749                        "waiting for %d zombie sessions to die.\n",
1750                        atomic_read(&sfw_data.fw_nzombies));
1751
1752         spin_unlock(&sfw_data.fw_lock);
1753
1754         for (i = 0; ; i++) {
1755                 sv = &sfw_services[i];
1756                 if (sv->sv_name == NULL)
1757                         break;
1758
1759                 srpc_shutdown_service(sv);
1760                 srpc_remove_service(sv);
1761         }
1762
1763         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1764                 sv = tsc->tsc_srv_service;
1765                 srpc_shutdown_service(sv);
1766                 srpc_remove_service(sv);
1767         }
1768
1769         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1770                 struct srpc_client_rpc *rpc;
1771
1772                 rpc = list_first_entry(&sfw_data.fw_zombie_rpcs,
1773                                        struct srpc_client_rpc, crpc_list);
1774                 list_del(&rpc->crpc_list);
1775
1776                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1777         }
1778
1779         for (i = 0; ; i++) {
1780                 sv = &sfw_services[i];
1781                 if (sv->sv_name == NULL)
1782                         break;
1783
1784                 srpc_wait_service_shutdown(sv);
1785         }
1786
1787         while (!list_empty(&sfw_data.fw_tests)) {
1788                 tsc = list_first_entry(&sfw_data.fw_tests,
1789                                        struct sfw_test_case, tsc_list);
1790
1791                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1792
1793                 list_del(&tsc->tsc_list);
1794                 LIBCFS_FREE(tsc, sizeof(*tsc));
1795         }
1796 }