drivers/staging/lustre/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, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lnet/selftest/framework.c
  33  *
  34  * Author: Isaac Huang <isaac@clusterfs.com>
  35  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LNET
  39
  40 #include "selftest.h"
  41
  42 struct lst_sid LST_INVALID_SID = {LNET_NID_ANY, -1};
  43
  44 static int session_timeout = 100;
  45 module_param(session_timeout, int, 0444);
  46 MODULE_PARM_DESC(session_timeout, "test session timeout in seconds (100 by default, 0 == never)");
  47
  48 static int rpc_timeout = 64;
  49 module_param(rpc_timeout, int, 0644);
  50 MODULE_PARM_DESC(rpc_timeout, "rpc timeout in seconds (64 by default, 0 == never)");
  51
  52 #define sfw_unpack_id(id)               \
  53 do {                                    \
  54         __swab64s(&(id).nid);           \
  55         __swab32s(&(id).pid);           \
  56 } while (0)
  57
  58 #define sfw_unpack_sid(sid)             \
  59 do {                                    \
  60         __swab64s(&(sid).ses_nid);      \
  61         __swab64s(&(sid).ses_stamp);    \
  62 } while (0)
  63
  64 #define sfw_unpack_fw_counters(fc)        \
  65 do {                                      \
  66         __swab32s(&(fc).running_ms);      \
  67         __swab32s(&(fc).active_batches);  \
  68         __swab32s(&(fc).zombie_sessions); \
  69         __swab32s(&(fc).brw_errors);      \
  70         __swab32s(&(fc).ping_errors);     \
  71 } while (0)
  72
  73 #define sfw_unpack_rpc_counters(rc)     \
  74 do {                                    \
  75         __swab32s(&(rc).errors);        \
  76         __swab32s(&(rc).rpcs_sent);     \
  77         __swab32s(&(rc).rpcs_rcvd);     \
  78         __swab32s(&(rc).rpcs_dropped);  \
  79         __swab32s(&(rc).rpcs_expired);  \
  80         __swab64s(&(rc).bulk_get);      \
  81         __swab64s(&(rc).bulk_put);      \
  82 } while (0)
  83
  84 #define sfw_unpack_lnet_counters(lc)    \
  85 do {                                    \
  86         __swab32s(&(lc).errors);        \
  87         __swab32s(&(lc).msgs_max);      \
  88         __swab32s(&(lc).msgs_alloc);    \
  89         __swab32s(&(lc).send_count);    \
  90         __swab32s(&(lc).recv_count);    \
  91         __swab32s(&(lc).drop_count);    \
  92         __swab32s(&(lc).route_count);   \
  93         __swab64s(&(lc).send_length);   \
  94         __swab64s(&(lc).recv_length);   \
  95         __swab64s(&(lc).drop_length);   \
  96         __swab64s(&(lc).route_length);  \
  97 } while (0)
  98
  99 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive))
 100 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive))
 101
 102 static struct smoketest_framework {
 103         struct list_head  fw_zombie_rpcs;     /* RPCs to be recycled */
 104         struct list_head  fw_zombie_sessions; /* stopping sessions */
 105         struct list_head  fw_tests;           /* registered test cases */
 106         atomic_t          fw_nzombies;        /* # zombie sessions */
 107         spinlock_t        fw_lock;            /* serialise */
 108         struct sfw_session        *fw_session;        /* _the_ session */
 109         int               fw_shuttingdown;    /* shutdown in progress */
 110         struct srpc_server_rpc *fw_active_srpc;/* running RPC */
 111 } sfw_data;
 112
 113 /* forward ref's */
 114 int sfw_stop_batch(struct sfw_batch *tsb, int force);
 115 void sfw_destroy_session(struct sfw_session *sn);
 116
 117 static inline struct sfw_test_case *
 118 sfw_find_test_case(int id)
 119 {
 120         struct sfw_test_case *tsc;
 121
 122         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 123         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 124
 125         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 126                 if (tsc->tsc_srv_service->sv_id == id)
 127                         return tsc;
 128         }
 129
 130         return NULL;
 131 }
 132
 133 static int
 134 sfw_register_test(struct srpc_service *service,
 135                   struct sfw_test_client_ops *cliops)
 136 {
 137         struct sfw_test_case *tsc;
 138
 139         if (sfw_find_test_case(service->sv_id)) {
 140                 CERROR("Failed to register test %s (%d)\n",
 141                        service->sv_name, service->sv_id);
 142                 return -EEXIST;
 143         }
 144
 145         LIBCFS_ALLOC(tsc, sizeof(struct sfw_test_case));
 146         if (!tsc)
 147                 return -ENOMEM;
 148
 149         tsc->tsc_cli_ops = cliops;
 150         tsc->tsc_srv_service = service;
 151
 152         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 153         return 0;
 154 }
 155
 156 static void
 157 sfw_add_session_timer(void)
 158 {
 159         struct sfw_session *sn = sfw_data.fw_session;
 160         struct stt_timer *timer = &sn->sn_timer;
 161
 162         LASSERT(!sfw_data.fw_shuttingdown);
 163
 164         if (!sn || !sn->sn_timeout)
 165                 return;
 166
 167         LASSERT(!sn->sn_timer_active);
 168
 169         sn->sn_timer_active = 1;
 170         timer->stt_expires = ktime_get_real_seconds() + sn->sn_timeout;
 171         stt_add_timer(timer);
 172 }
 173
 174 static int
 175 sfw_del_session_timer(void)
 176 {
 177         struct sfw_session *sn = sfw_data.fw_session;
 178
 179         if (!sn || !sn->sn_timer_active)
 180                 return 0;
 181
 182         LASSERT(sn->sn_timeout);
 183
 184         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 185                 sn->sn_timer_active = 0;
 186                 return 0;
 187         }
 188
 189         return EBUSY; /* racing with sfw_session_expired() */
 190 }
 191
 192 static void
 193 sfw_deactivate_session(void)
 194 __must_hold(&sfw_data.fw_lock)
 195 {
 196         struct sfw_session *sn = sfw_data.fw_session;
 197         int nactive = 0;
 198         struct sfw_batch *tsb;
 199         struct sfw_test_case *tsc;
 200
 201         if (!sn)
 202                 return;
 203
 204         LASSERT(!sn->sn_timer_active);
 205
 206         sfw_data.fw_session = NULL;
 207         atomic_inc(&sfw_data.fw_nzombies);
 208         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 209
 210         spin_unlock(&sfw_data.fw_lock);
 211
 212         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 213                 srpc_abort_service(tsc->tsc_srv_service);
 214         }
 215
 216         spin_lock(&sfw_data.fw_lock);
 217
 218         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 219                 if (sfw_batch_active(tsb)) {
 220                         nactive++;
 221                         sfw_stop_batch(tsb, 1);
 222                 }
 223         }
 224
 225         if (nactive)
 226                 return; /* wait for active batches to stop */
 227
 228         list_del_init(&sn->sn_list);
 229         spin_unlock(&sfw_data.fw_lock);
 230
 231         sfw_destroy_session(sn);
 232
 233         spin_lock(&sfw_data.fw_lock);
 234 }
 235
 236 static void
 237 sfw_session_expired(void *data)
 238 {
 239         struct sfw_session *sn = data;
 240
 241         spin_lock(&sfw_data.fw_lock);
 242
 243         LASSERT(sn->sn_timer_active);
 244         LASSERT(sn == sfw_data.fw_session);
 245
 246         CWARN("Session expired! sid: %s-%llu, name: %s\n",
 247               libcfs_nid2str(sn->sn_id.ses_nid),
 248               sn->sn_id.ses_stamp, &sn->sn_name[0]);
 249
 250         sn->sn_timer_active = 0;
 251         sfw_deactivate_session();
 252
 253         spin_unlock(&sfw_data.fw_lock);
 254 }
 255
 256 static inline void
 257 sfw_init_session(struct sfw_session *sn, struct lst_sid sid,
 258                  unsigned int features, const char *name)
 259 {
 260         struct stt_timer *timer = &sn->sn_timer;
 261
 262         memset(sn, 0, sizeof(struct sfw_session));
 263         INIT_LIST_HEAD(&sn->sn_list);
 264         INIT_LIST_HEAD(&sn->sn_batches);
 265         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 266         atomic_set(&sn->sn_brw_errors, 0);
 267         atomic_set(&sn->sn_ping_errors, 0);
 268         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 269
 270         sn->sn_timer_active = 0;
 271         sn->sn_id = sid;
 272         sn->sn_features = features;
 273         sn->sn_timeout = session_timeout;
 274         sn->sn_started = cfs_time_current();
 275
 276         timer->stt_data = sn;
 277         timer->stt_func = sfw_session_expired;
 278         INIT_LIST_HEAD(&timer->stt_list);
 279 }
 280
 281 /* completion handler for incoming framework RPCs */
 282 static void
 283 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 284 {
 285         struct srpc_service *sv = rpc->srpc_scd->scd_svc;
 286         int status = rpc->srpc_status;
 287
 288         CDEBUG(D_NET, "Incoming framework RPC done: service %s, peer %s, status %s:%d\n",
 289                sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 290                swi_state2str(rpc->srpc_wi.swi_state),
 291                status);
 292
 293         if (rpc->srpc_bulk)
 294                 sfw_free_pages(rpc);
 295 }
 296
 297 static void
 298 sfw_client_rpc_fini(struct srpc_client_rpc *rpc)
 299 {
 300         LASSERT(!rpc->crpc_bulk.bk_niov);
 301         LASSERT(list_empty(&rpc->crpc_list));
 302         LASSERT(!atomic_read(&rpc->crpc_refcount));
 303
 304         CDEBUG(D_NET, "Outgoing framework RPC done: service %d, peer %s, status %s:%d:%d\n",
 305                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 306                swi_state2str(rpc->crpc_wi.swi_state),
 307                rpc->crpc_aborted, rpc->crpc_status);
 308
 309         spin_lock(&sfw_data.fw_lock);
 310
 311         /* my callers must finish all RPCs before shutting me down */
 312         LASSERT(!sfw_data.fw_shuttingdown);
 313         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 314
 315         spin_unlock(&sfw_data.fw_lock);
 316 }
 317
 318 static struct sfw_batch *
 319 sfw_find_batch(struct lst_bid bid)
 320 {
 321         struct sfw_session *sn = sfw_data.fw_session;
 322         struct sfw_batch *bat;
 323
 324         LASSERT(sn);
 325
 326         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 327                 if (bat->bat_id.bat_id == bid.bat_id)
 328                         return bat;
 329         }
 330
 331         return NULL;
 332 }
 333
 334 static struct sfw_batch *
 335 sfw_bid2batch(struct lst_bid bid)
 336 {
 337         struct sfw_session *sn = sfw_data.fw_session;
 338         struct sfw_batch *bat;
 339
 340         LASSERT(sn);
 341
 342         bat = sfw_find_batch(bid);
 343         if (bat)
 344                 return bat;
 345
 346         LIBCFS_ALLOC(bat, sizeof(struct sfw_batch));
 347         if (!bat)
 348                 return NULL;
 349
 350         bat->bat_error = 0;
 351         bat->bat_session = sn;
 352         bat->bat_id = bid;
 353         atomic_set(&bat->bat_nactive, 0);
 354         INIT_LIST_HEAD(&bat->bat_tests);
 355
 356         list_add_tail(&bat->bat_list, &sn->sn_batches);
 357         return bat;
 358 }
 359
 360 static int
 361 sfw_get_stats(struct srpc_stat_reqst *request, struct srpc_stat_reply *reply)
 362 {
 363         struct sfw_session *sn = sfw_data.fw_session;
 364         struct sfw_counters *cnt = &reply->str_fw;
 365         struct sfw_batch *bat;
 366
 367         reply->str_sid = !sn ? LST_INVALID_SID : sn->sn_id;
 368
 369         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 370                 reply->str_status = EINVAL;
 371                 return 0;
 372         }
 373
 374         if (!sn || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 375                 reply->str_status = ESRCH;
 376                 return 0;
 377         }
 378
 379         lnet_counters_get(&reply->str_lnet);
 380         srpc_get_counters(&reply->str_rpc);
 381
 382         /*
 383          * send over the msecs since the session was started
 384          * with 32 bits to send, this is ~49 days
 385          */
 386         cnt->running_ms = jiffies_to_msecs(jiffies - sn->sn_started);
 387         cnt->brw_errors = atomic_read(&sn->sn_brw_errors);
 388         cnt->ping_errors = atomic_read(&sn->sn_ping_errors);
 389         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 390
 391         cnt->active_batches = 0;
 392         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 393                 if (atomic_read(&bat->bat_nactive) > 0)
 394                         cnt->active_batches++;
 395         }
 396
 397         reply->str_status = 0;
 398         return 0;
 399 }
 400
 401 int
 402 sfw_make_session(struct srpc_mksn_reqst *request, struct srpc_mksn_reply *reply)
 403 {
 404         struct sfw_session *sn = sfw_data.fw_session;
 405         struct srpc_msg *msg = container_of(request, struct srpc_msg,
 406                                        msg_body.mksn_reqst);
 407         int cplen = 0;
 408
 409         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 410                 reply->mksn_sid = !sn ? LST_INVALID_SID : sn->sn_id;
 411                 reply->mksn_status = EINVAL;
 412                 return 0;
 413         }
 414
 415         if (sn) {
 416                 reply->mksn_status = 0;
 417                 reply->mksn_sid = sn->sn_id;
 418                 reply->mksn_timeout = sn->sn_timeout;
 419
 420                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 421                         atomic_inc(&sn->sn_refcount);
 422                         return 0;
 423                 }
 424
 425                 if (!request->mksn_force) {
 426                         reply->mksn_status = EBUSY;
 427                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 428                                         sizeof(reply->mksn_name));
 429                         if (cplen >= sizeof(reply->mksn_name))
 430                                 return -E2BIG;
 431                         return 0;
 432                 }
 433         }
 434
 435         /*
 436          * reject the request if it requires unknown features
 437          * NB: old version will always accept all features because it's not
 438          * aware of srpc_msg::msg_ses_feats, it's a defect but it's also
 439          * harmless because it will return zero feature to console, and it's
 440          * console's responsibility to make sure all nodes in a session have
 441          * same feature mask.
 442          */
 443         if (msg->msg_ses_feats & ~LST_FEATS_MASK) {
 444                 reply->mksn_status = EPROTO;
 445                 return 0;
 446         }
 447
 448         /* brand new or create by force */
 449         LIBCFS_ALLOC(sn, sizeof(struct sfw_session));
 450         if (!sn) {
 451                 CERROR("dropping RPC mksn under memory pressure\n");
 452                 return -ENOMEM;
 453         }
 454
 455         sfw_init_session(sn, request->mksn_sid,
 456                          msg->msg_ses_feats, &request->mksn_name[0]);
 457
 458         spin_lock(&sfw_data.fw_lock);
 459
 460         sfw_deactivate_session();
 461         LASSERT(!sfw_data.fw_session);
 462         sfw_data.fw_session = sn;
 463
 464         spin_unlock(&sfw_data.fw_lock);
 465
 466         reply->mksn_status = 0;
 467         reply->mksn_sid = sn->sn_id;
 468         reply->mksn_timeout = sn->sn_timeout;
 469         return 0;
 470 }
 471
 472 static int
 473 sfw_remove_session(struct srpc_rmsn_reqst *request,
 474                    struct srpc_rmsn_reply *reply)
 475 {
 476         struct sfw_session *sn = sfw_data.fw_session;
 477
 478         reply->rmsn_sid = !sn ? LST_INVALID_SID : sn->sn_id;
 479
 480         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 481                 reply->rmsn_status = EINVAL;
 482                 return 0;
 483         }
 484
 485         if (!sn || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 486                 reply->rmsn_status = !sn ? ESRCH : EBUSY;
 487                 return 0;
 488         }
 489
 490         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 491                 reply->rmsn_status = 0;
 492                 return 0;
 493         }
 494
 495         spin_lock(&sfw_data.fw_lock);
 496         sfw_deactivate_session();
 497         spin_unlock(&sfw_data.fw_lock);
 498
 499         reply->rmsn_status = 0;
 500         reply->rmsn_sid = LST_INVALID_SID;
 501         LASSERT(!sfw_data.fw_session);
 502         return 0;
 503 }
 504
 505 static int
 506 sfw_debug_session(struct srpc_debug_reqst *request,
 507                   struct srpc_debug_reply *reply)
 508 {
 509         struct sfw_session *sn = sfw_data.fw_session;
 510
 511         if (!sn) {
 512                 reply->dbg_status = ESRCH;
 513                 reply->dbg_sid = LST_INVALID_SID;
 514                 return 0;
 515         }
 516
 517         reply->dbg_status = 0;
 518         reply->dbg_sid = sn->sn_id;
 519         reply->dbg_timeout = sn->sn_timeout;
 520         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 521             >= sizeof(reply->dbg_name))
 522                 return -E2BIG;
 523
 524         return 0;
 525 }
 526
 527 static void
 528 sfw_test_rpc_fini(struct srpc_client_rpc *rpc)
 529 {
 530         struct sfw_test_unit *tsu = rpc->crpc_priv;
 531         struct sfw_test_instance *tsi = tsu->tsu_instance;
 532
 533         /* Called with hold of tsi->tsi_lock */
 534         LASSERT(list_empty(&rpc->crpc_list));
 535         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 536 }
 537
 538 static inline int
 539 sfw_test_buffers(struct sfw_test_instance *tsi)
 540 {
 541         struct sfw_test_case *tsc;
 542         struct srpc_service *svc;
 543         int nbuf;
 544
 545         LASSERT(tsi);
 546         tsc = sfw_find_test_case(tsi->tsi_service);
 547         LASSERT(tsc);
 548         svc = tsc->tsc_srv_service;
 549         LASSERT(svc);
 550
 551         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 552         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 553 }
 554
 555 static int
 556 sfw_load_test(struct sfw_test_instance *tsi)
 557 {
 558         struct sfw_test_case *tsc;
 559         struct srpc_service *svc;
 560         int nbuf;
 561         int rc;
 562
 563         LASSERT(tsi);
 564         tsc = sfw_find_test_case(tsi->tsi_service);
 565         nbuf = sfw_test_buffers(tsi);
 566         LASSERT(tsc);
 567         svc = tsc->tsc_srv_service;
 568
 569         if (tsi->tsi_is_client) {
 570                 tsi->tsi_ops = tsc->tsc_cli_ops;
 571                 return 0;
 572         }
 573
 574         rc = srpc_service_add_buffers(svc, nbuf);
 575         if (rc) {
 576                 CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
 577                       svc->sv_name, nbuf, rc);
 578                 /*
 579                  * NB: this error handler is not strictly correct, because
 580                  * it may release more buffers than already allocated,
 581                  * but it doesn't matter because request portal should
 582                  * be lazy portal and will grow buffers if necessary.
 583                  */
 584                 srpc_service_remove_buffers(svc, nbuf);
 585                 return -ENOMEM;
 586         }
 587
 588         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 589                nbuf * (srpc_serv_is_framework(svc) ?
 590                        2 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 591         return 0;
 592 }
 593
 594 static void
 595 sfw_unload_test(struct sfw_test_instance *tsi)
 596 {
 597         struct sfw_test_case *tsc;
 598
 599         LASSERT(tsi);
 600         tsc = sfw_find_test_case(tsi->tsi_service);
 601         LASSERT(tsc);
 602
 603         if (tsi->tsi_is_client)
 604                 return;
 605
 606         /*
 607          * shrink buffers, because request portal is lazy portal
 608          * which can grow buffers at runtime so we may leave
 609          * some buffers behind, but never mind...
 610          */
 611         srpc_service_remove_buffers(tsc->tsc_srv_service,
 612                                     sfw_test_buffers(tsi));
 613 }
 614
 615 static void
 616 sfw_destroy_test_instance(struct sfw_test_instance *tsi)
 617 {
 618         struct srpc_client_rpc *rpc;
 619         struct sfw_test_unit *tsu;
 620
 621         if (!tsi->tsi_is_client)
 622                 goto clean;
 623
 624         tsi->tsi_ops->tso_fini(tsi);
 625
 626         LASSERT(!tsi->tsi_stopping);
 627         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 628         LASSERT(!sfw_test_active(tsi));
 629
 630         while (!list_empty(&tsi->tsi_units)) {
 631                 tsu = list_entry(tsi->tsi_units.next,
 632                                  struct sfw_test_unit, tsu_list);
 633                 list_del(&tsu->tsu_list);
 634                 LIBCFS_FREE(tsu, sizeof(*tsu));
 635         }
 636
 637         while (!list_empty(&tsi->tsi_free_rpcs)) {
 638                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 639                                  struct srpc_client_rpc, crpc_list);
 640                 list_del(&rpc->crpc_list);
 641                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 642         }
 643
 644 clean:
 645         sfw_unload_test(tsi);
 646         LIBCFS_FREE(tsi, sizeof(*tsi));
 647 }
 648
 649 static void
 650 sfw_destroy_batch(struct sfw_batch *tsb)
 651 {
 652         struct sfw_test_instance *tsi;
 653
 654         LASSERT(!sfw_batch_active(tsb));
 655         LASSERT(list_empty(&tsb->bat_list));
 656
 657         while (!list_empty(&tsb->bat_tests)) {
 658                 tsi = list_entry(tsb->bat_tests.next,
 659                                  struct sfw_test_instance, tsi_list);
 660                 list_del_init(&tsi->tsi_list);
 661                 sfw_destroy_test_instance(tsi);
 662         }
 663
 664         LIBCFS_FREE(tsb, sizeof(struct sfw_batch));
 665 }
 666
 667 void
 668 sfw_destroy_session(struct sfw_session *sn)
 669 {
 670         struct sfw_batch *batch;
 671
 672         LASSERT(list_empty(&sn->sn_list));
 673         LASSERT(sn != sfw_data.fw_session);
 674
 675         while (!list_empty(&sn->sn_batches)) {
 676                 batch = list_entry(sn->sn_batches.next,
 677                                    struct sfw_batch, bat_list);
 678                 list_del_init(&batch->bat_list);
 679                 sfw_destroy_batch(batch);
 680         }
 681
 682         LIBCFS_FREE(sn, sizeof(*sn));
 683         atomic_dec(&sfw_data.fw_nzombies);
 684 }
 685
 686 static void
 687 sfw_unpack_addtest_req(struct srpc_msg *msg)
 688 {
 689         struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
 690
 691         LASSERT(msg->msg_type == SRPC_MSG_TEST_REQST);
 692         LASSERT(req->tsr_is_client);
 693
 694         if (msg->msg_magic == SRPC_MSG_MAGIC)
 695                 return; /* no flipping needed */
 696
 697         LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 698
 699         if (req->tsr_service == SRPC_SERVICE_BRW) {
 700                 if (!(msg->msg_ses_feats & LST_FEAT_BULK_LEN)) {
 701                         struct test_bulk_req *bulk = &req->tsr_u.bulk_v0;
 702
 703                         __swab32s(&bulk->blk_opc);
 704                         __swab32s(&bulk->blk_npg);
 705                         __swab32s(&bulk->blk_flags);
 706
 707                 } else {
 708                         struct test_bulk_req_v1 *bulk = &req->tsr_u.bulk_v1;
 709
 710                         __swab16s(&bulk->blk_opc);
 711                         __swab16s(&bulk->blk_flags);
 712                         __swab32s(&bulk->blk_offset);
 713                         __swab32s(&bulk->blk_len);
 714                 }
 715
 716                 return;
 717         }
 718
 719         if (req->tsr_service == SRPC_SERVICE_PING) {
 720                 struct test_ping_req *ping = &req->tsr_u.ping;
 721
 722                 __swab32s(&ping->png_size);
 723                 __swab32s(&ping->png_flags);
 724                 return;
 725         }
 726
 727         LBUG();
 728 }
 729
 730 static int
 731 sfw_add_test_instance(struct sfw_batch *tsb, struct srpc_server_rpc *rpc)
 732 {
 733         struct srpc_msg *msg = &rpc->srpc_reqstbuf->buf_msg;
 734         struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
 735         struct srpc_bulk *bk = rpc->srpc_bulk;
 736         int ndest = req->tsr_ndest;
 737         struct sfw_test_unit *tsu;
 738         struct sfw_test_instance *tsi;
 739         int i;
 740         int rc;
 741
 742         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 743         if (!tsi) {
 744                 CERROR("Can't allocate test instance for batch: %llu\n",
 745                        tsb->bat_id.bat_id);
 746                 return -ENOMEM;
 747         }
 748
 749         spin_lock_init(&tsi->tsi_lock);
 750         atomic_set(&tsi->tsi_nactive, 0);
 751         INIT_LIST_HEAD(&tsi->tsi_units);
 752         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 753         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 754
 755         tsi->tsi_stopping = 0;
 756         tsi->tsi_batch = tsb;
 757         tsi->tsi_loop = req->tsr_loop;
 758         tsi->tsi_concur = req->tsr_concur;
 759         tsi->tsi_service = req->tsr_service;
 760         tsi->tsi_is_client = !!(req->tsr_is_client);
 761         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 762
 763         rc = sfw_load_test(tsi);
 764         if (rc) {
 765                 LIBCFS_FREE(tsi, sizeof(*tsi));
 766                 return rc;
 767         }
 768
 769         LASSERT(!sfw_batch_active(tsb));
 770
 771         if (!tsi->tsi_is_client) {
 772                 /* it's test server, just add it to tsb */
 773                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 774                 return 0;
 775         }
 776
 777         LASSERT(bk);
 778         LASSERT(bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 779         LASSERT((unsigned int)bk->bk_len >=
 780                 sizeof(struct lnet_process_id_packed) * ndest);
 781
 782         sfw_unpack_addtest_req(msg);
 783         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 784
 785         for (i = 0; i < ndest; i++) {
 786                 struct lnet_process_id_packed *dests;
 787                 struct lnet_process_id_packed id;
 788                 int j;
 789
 790                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].bv_page);
 791                 LASSERT(dests);  /* my pages are within KVM always */
 792                 id = dests[i % SFW_ID_PER_PAGE];
 793                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 794                         sfw_unpack_id(id);
 795
 796                 for (j = 0; j < tsi->tsi_concur; j++) {
 797                         LIBCFS_ALLOC(tsu, sizeof(struct sfw_test_unit));
 798                         if (!tsu) {
 799                                 rc = -ENOMEM;
 800                                 CERROR("Can't allocate tsu for %d\n",
 801                                        tsi->tsi_service);
 802                                 goto error;
 803                         }
 804
 805                         tsu->tsu_dest.nid = id.nid;
 806                         tsu->tsu_dest.pid = id.pid;
 807                         tsu->tsu_instance = tsi;
 808                         tsu->tsu_private = NULL;
 809                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 810                 }
 811         }
 812
 813         rc = tsi->tsi_ops->tso_init(tsi);
 814         if (!rc) {
 815                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 816                 return 0;
 817         }
 818
 819 error:
 820         LASSERT(rc);
 821         sfw_destroy_test_instance(tsi);
 822         return rc;
 823 }
 824
 825 static void
 826 sfw_test_unit_done(struct sfw_test_unit *tsu)
 827 {
 828         struct sfw_test_instance *tsi = tsu->tsu_instance;
 829         struct sfw_batch *tsb = tsi->tsi_batch;
 830         struct sfw_session *sn = tsb->bat_session;
 831
 832         LASSERT(sfw_test_active(tsi));
 833
 834         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 835                 return;
 836
 837         /* the test instance is done */
 838         spin_lock(&tsi->tsi_lock);
 839
 840         tsi->tsi_stopping = 0;
 841
 842         spin_unlock(&tsi->tsi_lock);
 843
 844         spin_lock(&sfw_data.fw_lock);
 845
 846         if (!atomic_dec_and_test(&tsb->bat_nactive) ||  /* tsb still active */
 847             sn == sfw_data.fw_session) {                /* sn also active */
 848                 spin_unlock(&sfw_data.fw_lock);
 849                 return;
 850         }
 851
 852         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 853
 854         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 855                 if (sfw_batch_active(tsb)) {
 856                         spin_unlock(&sfw_data.fw_lock);
 857                         return;
 858                 }
 859         }
 860
 861         list_del_init(&sn->sn_list);
 862         spin_unlock(&sfw_data.fw_lock);
 863
 864         sfw_destroy_session(sn);
 865 }
 866
 867 static void
 868 sfw_test_rpc_done(struct srpc_client_rpc *rpc)
 869 {
 870         struct sfw_test_unit *tsu = rpc->crpc_priv;
 871         struct sfw_test_instance *tsi = tsu->tsu_instance;
 872         int done = 0;
 873
 874         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 875
 876         spin_lock(&tsi->tsi_lock);
 877
 878         LASSERT(sfw_test_active(tsi));
 879         LASSERT(!list_empty(&rpc->crpc_list));
 880
 881         list_del_init(&rpc->crpc_list);
 882
 883         /* batch is stopping or loop is done or get error */
 884         if (tsi->tsi_stopping || !tsu->tsu_loop ||
 885             (rpc->crpc_status && tsi->tsi_stoptsu_onerr))
 886                 done = 1;
 887
 888         /* dec ref for poster */
 889         srpc_client_rpc_decref(rpc);
 890
 891         spin_unlock(&tsi->tsi_lock);
 892
 893         if (!done) {
 894                 swi_schedule_workitem(&tsu->tsu_worker);
 895                 return;
 896         }
 897
 898         sfw_test_unit_done(tsu);
 899 }
 900
 901 int
 902 sfw_create_test_rpc(struct sfw_test_unit *tsu, struct lnet_process_id peer,
 903                     unsigned int features, int nblk, int blklen,
 904                     struct srpc_client_rpc **rpcpp)
 905 {
 906         struct srpc_client_rpc *rpc = NULL;
 907         struct sfw_test_instance *tsi = tsu->tsu_instance;
 908
 909         spin_lock(&tsi->tsi_lock);
 910
 911         LASSERT(sfw_test_active(tsi));
 912                 /* pick request from buffer */
 913         rpc = list_first_entry_or_null(&tsi->tsi_free_rpcs,
 914                                        struct srpc_client_rpc, crpc_list);
 915         if (rpc) {
 916                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 917                 list_del_init(&rpc->crpc_list);
 918         }
 919
 920         spin_unlock(&tsi->tsi_lock);
 921
 922         if (!rpc) {
 923                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 924                                              blklen, sfw_test_rpc_done,
 925                                              sfw_test_rpc_fini, tsu);
 926         } else {
 927                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 928                                      blklen, sfw_test_rpc_done,
 929                                      sfw_test_rpc_fini, tsu);
 930         }
 931
 932         if (!rpc) {
 933                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 934                 return -ENOMEM;
 935         }
 936
 937         rpc->crpc_reqstmsg.msg_ses_feats = features;
 938         *rpcpp = rpc;
 939
 940         return 0;
 941 }
 942
 943 static int
 944 sfw_run_test(struct swi_workitem *wi)
 945 {
 946         struct sfw_test_unit *tsu = wi->swi_workitem.wi_data;
 947         struct sfw_test_instance *tsi = tsu->tsu_instance;
 948         struct srpc_client_rpc *rpc = NULL;
 949
 950         LASSERT(wi == &tsu->tsu_worker);
 951
 952         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc)) {
 953                 LASSERT(!rpc);
 954                 goto test_done;
 955         }
 956
 957         LASSERT(rpc);
 958
 959         spin_lock(&tsi->tsi_lock);
 960
 961         if (tsi->tsi_stopping) {
 962                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 963                 spin_unlock(&tsi->tsi_lock);
 964                 goto test_done;
 965         }
 966
 967         if (tsu->tsu_loop > 0)
 968                 tsu->tsu_loop--;
 969
 970         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
 971         spin_unlock(&tsi->tsi_lock);
 972
 973         spin_lock(&rpc->crpc_lock);
 974         rpc->crpc_timeout = rpc_timeout;
 975         srpc_post_rpc(rpc);
 976         spin_unlock(&rpc->crpc_lock);
 977         return 0;
 978
 979 test_done:
 980         /*
 981          * No one can schedule me now since:
 982          * - previous RPC, if any, has done and
 983          * - no new RPC is initiated.
 984          * - my batch is still active; no one can run it again now.
 985          * Cancel pending schedules and prevent future schedule attempts:
 986          */
 987         swi_exit_workitem(wi);
 988         sfw_test_unit_done(tsu);
 989         return 1;
 990 }
 991
 992 static int
 993 sfw_run_batch(struct sfw_batch *tsb)
 994 {
 995         struct swi_workitem *wi;
 996         struct sfw_test_unit *tsu;
 997         struct sfw_test_instance *tsi;
 998
 999         if (sfw_batch_active(tsb)) {
1000                 CDEBUG(D_NET, "Batch already active: %llu (%d)\n",
1001                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1002                 return 0;
1003         }
1004
1005         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1006                 if (!tsi->tsi_is_client) /* skip server instances */
1007                         continue;
1008
1009                 LASSERT(!tsi->tsi_stopping);
1010                 LASSERT(!sfw_test_active(tsi));
1011
1012                 atomic_inc(&tsb->bat_nactive);
1013
1014                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1015                         atomic_inc(&tsi->tsi_nactive);
1016                         tsu->tsu_loop = tsi->tsi_loop;
1017                         wi = &tsu->tsu_worker;
1018                         swi_init_workitem(wi, tsu, sfw_run_test,
1019                                           lst_sched_test[lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1020                         swi_schedule_workitem(wi);
1021                 }
1022         }
1023
1024         return 0;
1025 }
1026
1027 int
1028 sfw_stop_batch(struct sfw_batch *tsb, int force)
1029 {
1030         struct sfw_test_instance *tsi;
1031         struct srpc_client_rpc *rpc;
1032
1033         if (!sfw_batch_active(tsb)) {
1034                 CDEBUG(D_NET, "Batch %llu inactive\n", tsb->bat_id.bat_id);
1035                 return 0;
1036         }
1037
1038         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1039                 spin_lock(&tsi->tsi_lock);
1040
1041                 if (!tsi->tsi_is_client ||
1042                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1043                         spin_unlock(&tsi->tsi_lock);
1044                         continue;
1045                 }
1046
1047                 tsi->tsi_stopping = 1;
1048
1049                 if (!force) {
1050                         spin_unlock(&tsi->tsi_lock);
1051                         continue;
1052                 }
1053
1054                 /* abort launched rpcs in the test */
1055                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1056                         spin_lock(&rpc->crpc_lock);
1057
1058                         srpc_abort_rpc(rpc, -EINTR);
1059
1060                         spin_unlock(&rpc->crpc_lock);
1061                 }
1062
1063                 spin_unlock(&tsi->tsi_lock);
1064         }
1065
1066         return 0;
1067 }
1068
1069 static int
1070 sfw_query_batch(struct sfw_batch *tsb, int testidx,
1071                 struct srpc_batch_reply *reply)
1072 {
1073         struct sfw_test_instance *tsi;
1074
1075         if (testidx < 0)
1076                 return -EINVAL;
1077
1078         if (!testidx) {
1079                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1080                 return 0;
1081         }
1082
1083         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1084                 if (testidx-- > 1)
1085                         continue;
1086
1087                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1088                 return 0;
1089         }
1090
1091         return -ENOENT;
1092 }
1093
1094 void
1095 sfw_free_pages(struct srpc_server_rpc *rpc)
1096 {
1097         srpc_free_bulk(rpc->srpc_bulk);
1098         rpc->srpc_bulk = NULL;
1099 }
1100
1101 int
1102 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1103                 int sink)
1104 {
1105         LASSERT(!rpc->srpc_bulk);
1106         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1107
1108         rpc->srpc_bulk = srpc_alloc_bulk(cpt, 0, npages, len, sink);
1109         if (!rpc->srpc_bulk)
1110                 return -ENOMEM;
1111
1112         return 0;
1113 }
1114
1115 static int
1116 sfw_add_test(struct srpc_server_rpc *rpc)
1117 {
1118         struct sfw_session *sn = sfw_data.fw_session;
1119         struct srpc_test_reply *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1120         struct srpc_test_reqst *request;
1121         int rc;
1122         struct sfw_batch *bat;
1123
1124         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1125         reply->tsr_sid = !sn ? LST_INVALID_SID : sn->sn_id;
1126
1127         if (!request->tsr_loop ||
1128             !request->tsr_concur ||
1129             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1130             request->tsr_ndest > SFW_MAX_NDESTS ||
1131             (request->tsr_is_client && !request->tsr_ndest) ||
1132             request->tsr_concur > SFW_MAX_CONCUR ||
1133             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1134             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1135                 reply->tsr_status = EINVAL;
1136                 return 0;
1137         }
1138
1139         if (!sn || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1140             !sfw_find_test_case(request->tsr_service)) {
1141                 reply->tsr_status = ENOENT;
1142                 return 0;
1143         }
1144
1145         bat = sfw_bid2batch(request->tsr_bid);
1146         if (!bat) {
1147                 CERROR("dropping RPC %s from %s under memory pressure\n",
1148                        rpc->srpc_scd->scd_svc->sv_name,
1149                        libcfs_id2str(rpc->srpc_peer));
1150                 return -ENOMEM;
1151         }
1152
1153         if (sfw_batch_active(bat)) {
1154                 reply->tsr_status = EBUSY;
1155                 return 0;
1156         }
1157
1158         if (request->tsr_is_client && !rpc->srpc_bulk) {
1159                 /* rpc will be resumed later in sfw_bulk_ready */
1160                 int npg = sfw_id_pages(request->tsr_ndest);
1161                 int len;
1162
1163                 if (!(sn->sn_features & LST_FEAT_BULK_LEN)) {
1164                         len = npg * PAGE_SIZE;
1165
1166                 } else {
1167                         len = sizeof(struct lnet_process_id_packed) *
1168                               request->tsr_ndest;
1169                 }
1170
1171                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1172         }
1173
1174         rc = sfw_add_test_instance(bat, rpc);
1175         CDEBUG(!rc ? D_NET : D_WARNING,
1176                "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1177                !rc ? "Added" : "Failed to add", request->tsr_service,
1178                request->tsr_is_client ? "client" : "server",
1179                request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1180
1181         reply->tsr_status = (rc < 0) ? -rc : rc;
1182         return 0;
1183 }
1184
1185 static int
1186 sfw_control_batch(struct srpc_batch_reqst *request,
1187                   struct srpc_batch_reply *reply)
1188 {
1189         struct sfw_session *sn = sfw_data.fw_session;
1190         int rc = 0;
1191         struct sfw_batch *bat;
1192
1193         reply->bar_sid = !sn ? LST_INVALID_SID : sn->sn_id;
1194
1195         if (!sn || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1196                 reply->bar_status = ESRCH;
1197                 return 0;
1198         }
1199
1200         bat = sfw_find_batch(request->bar_bid);
1201         if (!bat) {
1202                 reply->bar_status = ENOENT;
1203                 return 0;
1204         }
1205
1206         switch (request->bar_opc) {
1207         case SRPC_BATCH_OPC_RUN:
1208                 rc = sfw_run_batch(bat);
1209                 break;
1210
1211         case SRPC_BATCH_OPC_STOP:
1212                 rc = sfw_stop_batch(bat, request->bar_arg);
1213                 break;
1214
1215         case SRPC_BATCH_OPC_QUERY:
1216                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1217                 break;
1218
1219         default:
1220                 return -EINVAL; /* drop it */
1221         }
1222
1223         reply->bar_status = (rc < 0) ? -rc : rc;
1224         return 0;
1225 }
1226
1227 static int
1228 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1229 {
1230         struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1231         struct srpc_msg *reply = &rpc->srpc_replymsg;
1232         struct srpc_msg *request = &rpc->srpc_reqstbuf->buf_msg;
1233         unsigned int features = LST_FEATS_MASK;
1234         int rc = 0;
1235
1236         LASSERT(!sfw_data.fw_active_srpc);
1237         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1238
1239         spin_lock(&sfw_data.fw_lock);
1240
1241         if (sfw_data.fw_shuttingdown) {
1242                 spin_unlock(&sfw_data.fw_lock);
1243                 return -ESHUTDOWN;
1244         }
1245
1246         /* Remove timer to avoid racing with it or expiring active session */
1247         if (sfw_del_session_timer()) {
1248                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1249                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1250                 spin_unlock(&sfw_data.fw_lock);
1251                 return -EAGAIN;
1252         }
1253
1254         sfw_data.fw_active_srpc = rpc;
1255         spin_unlock(&sfw_data.fw_lock);
1256
1257         sfw_unpack_message(request);
1258         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1259
1260         /* rpc module should have checked this */
1261         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1262
1263         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1264             sv->sv_id != SRPC_SERVICE_DEBUG) {
1265                 struct sfw_session *sn = sfw_data.fw_session;
1266
1267                 if (sn &&
1268                     sn->sn_features != request->msg_ses_feats) {
1269                         CNETERR("Features of framework RPC don't match features of current session: %x/%x\n",
1270                                 request->msg_ses_feats, sn->sn_features);
1271                         reply->msg_body.reply.status = EPROTO;
1272                         reply->msg_body.reply.sid = sn->sn_id;
1273                         goto out;
1274                 }
1275
1276         } else if (request->msg_ses_feats & ~LST_FEATS_MASK) {
1277                 /*
1278                  * NB: at this point, old version will ignore features and
1279                  * create new session anyway, so console should be able
1280                  * to handle this
1281                  */
1282                 reply->msg_body.reply.status = EPROTO;
1283                 goto out;
1284         }
1285
1286         switch (sv->sv_id) {
1287         default:
1288                 LBUG();
1289         case SRPC_SERVICE_TEST:
1290                 rc = sfw_add_test(rpc);
1291                 break;
1292
1293         case SRPC_SERVICE_BATCH:
1294                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1295                                        &reply->msg_body.bat_reply);
1296                 break;
1297
1298         case SRPC_SERVICE_QUERY_STAT:
1299                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1300                                    &reply->msg_body.stat_reply);
1301                 break;
1302
1303         case SRPC_SERVICE_DEBUG:
1304                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1305                                        &reply->msg_body.dbg_reply);
1306                 break;
1307
1308         case SRPC_SERVICE_MAKE_SESSION:
1309                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1310                                       &reply->msg_body.mksn_reply);
1311                 break;
1312
1313         case SRPC_SERVICE_REMOVE_SESSION:
1314                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1315                                         &reply->msg_body.rmsn_reply);
1316                 break;
1317         }
1318
1319         if (sfw_data.fw_session)
1320                 features = sfw_data.fw_session->sn_features;
1321  out:
1322         reply->msg_ses_feats = features;
1323         rpc->srpc_done = sfw_server_rpc_done;
1324         spin_lock(&sfw_data.fw_lock);
1325
1326         if (!sfw_data.fw_shuttingdown)
1327                 sfw_add_session_timer();
1328
1329         sfw_data.fw_active_srpc = NULL;
1330         spin_unlock(&sfw_data.fw_lock);
1331         return rc;
1332 }
1333
1334 static int
1335 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1336 {
1337         struct srpc_service *sv = rpc->srpc_scd->scd_svc;
1338         int rc;
1339
1340         LASSERT(rpc->srpc_bulk);
1341         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1342         LASSERT(!sfw_data.fw_active_srpc);
1343         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1344
1345         spin_lock(&sfw_data.fw_lock);
1346
1347         if (status) {
1348                 CERROR("Bulk transfer failed for RPC: service %s, peer %s, status %d\n",
1349                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1350                 spin_unlock(&sfw_data.fw_lock);
1351                 return -EIO;
1352         }
1353
1354         if (sfw_data.fw_shuttingdown) {
1355                 spin_unlock(&sfw_data.fw_lock);
1356                 return -ESHUTDOWN;
1357         }
1358
1359         if (sfw_del_session_timer()) {
1360                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1361                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1362                 spin_unlock(&sfw_data.fw_lock);
1363                 return -EAGAIN;
1364         }
1365
1366         sfw_data.fw_active_srpc = rpc;
1367         spin_unlock(&sfw_data.fw_lock);
1368
1369         rc = sfw_add_test(rpc);
1370
1371         spin_lock(&sfw_data.fw_lock);
1372
1373         if (!sfw_data.fw_shuttingdown)
1374                 sfw_add_session_timer();
1375
1376         sfw_data.fw_active_srpc = NULL;
1377         spin_unlock(&sfw_data.fw_lock);
1378         return rc;
1379 }
1380
1381 struct srpc_client_rpc *
1382 sfw_create_rpc(struct lnet_process_id peer, int service,
1383                unsigned int features, int nbulkiov, int bulklen,
1384                void (*done)(struct srpc_client_rpc *), void *priv)
1385 {
1386         struct srpc_client_rpc *rpc = NULL;
1387
1388         spin_lock(&sfw_data.fw_lock);
1389
1390         LASSERT(!sfw_data.fw_shuttingdown);
1391         LASSERT(service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1392
1393         if (!nbulkiov && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1394                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1395                                  struct srpc_client_rpc, crpc_list);
1396                 list_del(&rpc->crpc_list);
1397
1398                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1399                                      done, sfw_client_rpc_fini, priv);
1400         }
1401
1402         spin_unlock(&sfw_data.fw_lock);
1403
1404         if (!rpc) {
1405                 rpc = srpc_create_client_rpc(peer, service,
1406                                              nbulkiov, bulklen, done,
1407                                              nbulkiov ?  NULL :
1408                                              sfw_client_rpc_fini,
1409                                              priv);
1410         }
1411
1412         if (rpc) /* "session" is concept in framework */
1413                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1414
1415         return rpc;
1416 }
1417
1418 void
1419 sfw_unpack_message(struct srpc_msg *msg)
1420 {
1421         if (msg->msg_magic == SRPC_MSG_MAGIC)
1422                 return; /* no flipping needed */
1423
1424         /* srpc module should guarantee I wouldn't get crap */
1425         LASSERT(msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1426
1427         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1428                 struct srpc_stat_reqst *req = &msg->msg_body.stat_reqst;
1429
1430                 __swab32s(&req->str_type);
1431                 __swab64s(&req->str_rpyid);
1432                 sfw_unpack_sid(req->str_sid);
1433                 return;
1434         }
1435
1436         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1437                 struct srpc_stat_reply *rep = &msg->msg_body.stat_reply;
1438
1439                 __swab32s(&rep->str_status);
1440                 sfw_unpack_sid(rep->str_sid);
1441                 sfw_unpack_fw_counters(rep->str_fw);
1442                 sfw_unpack_rpc_counters(rep->str_rpc);
1443                 sfw_unpack_lnet_counters(rep->str_lnet);
1444                 return;
1445         }
1446
1447         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1448                 struct srpc_mksn_reqst *req = &msg->msg_body.mksn_reqst;
1449
1450                 __swab64s(&req->mksn_rpyid);
1451                 __swab32s(&req->mksn_force);
1452                 sfw_unpack_sid(req->mksn_sid);
1453                 return;
1454         }
1455
1456         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1457                 struct srpc_mksn_reply *rep = &msg->msg_body.mksn_reply;
1458
1459                 __swab32s(&rep->mksn_status);
1460                 __swab32s(&rep->mksn_timeout);
1461                 sfw_unpack_sid(rep->mksn_sid);
1462                 return;
1463         }
1464
1465         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1466                 struct srpc_rmsn_reqst *req = &msg->msg_body.rmsn_reqst;
1467
1468                 __swab64s(&req->rmsn_rpyid);
1469                 sfw_unpack_sid(req->rmsn_sid);
1470                 return;
1471         }
1472
1473         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1474                 struct srpc_rmsn_reply *rep = &msg->msg_body.rmsn_reply;
1475
1476                 __swab32s(&rep->rmsn_status);
1477                 sfw_unpack_sid(rep->rmsn_sid);
1478                 return;
1479         }
1480
1481         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1482                 struct srpc_debug_reqst *req = &msg->msg_body.dbg_reqst;
1483
1484                 __swab64s(&req->dbg_rpyid);
1485                 __swab32s(&req->dbg_flags);
1486                 sfw_unpack_sid(req->dbg_sid);
1487                 return;
1488         }
1489
1490         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1491                 struct srpc_debug_reply *rep = &msg->msg_body.dbg_reply;
1492
1493                 __swab32s(&rep->dbg_nbatch);
1494                 __swab32s(&rep->dbg_timeout);
1495                 sfw_unpack_sid(rep->dbg_sid);
1496                 return;
1497         }
1498
1499         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1500                 struct srpc_batch_reqst *req = &msg->msg_body.bat_reqst;
1501
1502                 __swab32s(&req->bar_opc);
1503                 __swab64s(&req->bar_rpyid);
1504                 __swab32s(&req->bar_testidx);
1505                 __swab32s(&req->bar_arg);
1506                 sfw_unpack_sid(req->bar_sid);
1507                 __swab64s(&req->bar_bid.bat_id);
1508                 return;
1509         }
1510
1511         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1512                 struct srpc_batch_reply *rep = &msg->msg_body.bat_reply;
1513
1514                 __swab32s(&rep->bar_status);
1515                 sfw_unpack_sid(rep->bar_sid);
1516                 return;
1517         }
1518
1519         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1520                 struct srpc_test_reqst *req = &msg->msg_body.tes_reqst;
1521
1522                 __swab64s(&req->tsr_rpyid);
1523                 __swab64s(&req->tsr_bulkid);
1524                 __swab32s(&req->tsr_loop);
1525                 __swab32s(&req->tsr_ndest);
1526                 __swab32s(&req->tsr_concur);
1527                 __swab32s(&req->tsr_service);
1528                 sfw_unpack_sid(req->tsr_sid);
1529                 __swab64s(&req->tsr_bid.bat_id);
1530                 return;
1531         }
1532
1533         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1534                 struct srpc_test_reply *rep = &msg->msg_body.tes_reply;
1535
1536                 __swab32s(&rep->tsr_status);
1537                 sfw_unpack_sid(rep->tsr_sid);
1538                 return;
1539         }
1540
1541         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1542                 struct srpc_join_reqst *req = &msg->msg_body.join_reqst;
1543
1544                 __swab64s(&req->join_rpyid);
1545                 sfw_unpack_sid(req->join_sid);
1546                 return;
1547         }
1548
1549         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1550                 struct srpc_join_reply *rep = &msg->msg_body.join_reply;
1551
1552                 __swab32s(&rep->join_status);
1553                 __swab32s(&rep->join_timeout);
1554                 sfw_unpack_sid(rep->join_sid);
1555                 return;
1556         }
1557
1558         LBUG();
1559 }
1560
1561 void
1562 sfw_abort_rpc(struct srpc_client_rpc *rpc)
1563 {
1564         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1565         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1566
1567         spin_lock(&rpc->crpc_lock);
1568         srpc_abort_rpc(rpc, -EINTR);
1569         spin_unlock(&rpc->crpc_lock);
1570 }
1571
1572 void
1573 sfw_post_rpc(struct srpc_client_rpc *rpc)
1574 {
1575         spin_lock(&rpc->crpc_lock);
1576
1577         LASSERT(!rpc->crpc_closed);
1578         LASSERT(!rpc->crpc_aborted);
1579         LASSERT(list_empty(&rpc->crpc_list));
1580         LASSERT(!sfw_data.fw_shuttingdown);
1581
1582         rpc->crpc_timeout = rpc_timeout;
1583         srpc_post_rpc(rpc);
1584
1585         spin_unlock(&rpc->crpc_lock);
1586 }
1587
1588 static struct srpc_service sfw_services[] = {
1589         {
1590                 /* sv_id */    SRPC_SERVICE_DEBUG,
1591                 /* sv_name */  "debug",
1592                 0
1593         },
1594         {
1595                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1596                 /* sv_name */  "query stats",
1597                 0
1598         },
1599         {
1600                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1601                 /* sv_name */  "make session",
1602                 0
1603         },
1604         {
1605                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1606                 /* sv_name */  "remove session",
1607                 0
1608         },
1609         {
1610                 /* sv_id */    SRPC_SERVICE_BATCH,
1611                 /* sv_name */  "batch service",
1612                 0
1613         },
1614         {
1615                 /* sv_id */    SRPC_SERVICE_TEST,
1616                 /* sv_name */  "test service",
1617                 0
1618         },
1619         {
1620                 /* sv_id */    0,
1621                 /* sv_name */  NULL,
1622                 0
1623         }
1624 };
1625
1626 int
1627 sfw_startup(void)
1628 {
1629         int i;
1630         int rc;
1631         int error;
1632         struct srpc_service *sv;
1633         struct sfw_test_case *tsc;
1634
1635         if (session_timeout < 0) {
1636                 CERROR("Session timeout must be non-negative: %d\n",
1637                        session_timeout);
1638                 return -EINVAL;
1639         }
1640
1641         if (rpc_timeout < 0) {
1642                 CERROR("RPC timeout must be non-negative: %d\n",
1643                        rpc_timeout);
1644                 return -EINVAL;
1645         }
1646
1647         if (!session_timeout)
1648                 CWARN("Zero session_timeout specified - test sessions never expire.\n");
1649
1650         if (!rpc_timeout)
1651                 CWARN("Zero rpc_timeout specified - test RPC never expire.\n");
1652
1653         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1654
1655         sfw_data.fw_session = NULL;
1656         sfw_data.fw_active_srpc = NULL;
1657         spin_lock_init(&sfw_data.fw_lock);
1658         atomic_set(&sfw_data.fw_nzombies, 0);
1659         INIT_LIST_HEAD(&sfw_data.fw_tests);
1660         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1661         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1662
1663         brw_init_test_client();
1664         brw_init_test_service();
1665         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1666         LASSERT(!rc);
1667
1668         ping_init_test_client();
1669         ping_init_test_service();
1670         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1671         LASSERT(!rc);
1672
1673         error = 0;
1674         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1675                 sv = tsc->tsc_srv_service;
1676
1677                 rc = srpc_add_service(sv);
1678                 LASSERT(rc != -EBUSY);
1679                 if (rc) {
1680                         CWARN("Failed to add %s service: %d\n",
1681                               sv->sv_name, rc);
1682                         error = rc;
1683                 }
1684         }
1685
1686         for (i = 0; ; i++) {
1687                 sv = &sfw_services[i];
1688                 if (!sv->sv_name)
1689                         break;
1690
1691                 sv->sv_bulk_ready = NULL;
1692                 sv->sv_handler = sfw_handle_server_rpc;
1693                 sv->sv_wi_total = SFW_FRWK_WI_MAX;
1694                 if (sv->sv_id == SRPC_SERVICE_TEST)
1695                         sv->sv_bulk_ready = sfw_bulk_ready;
1696
1697                 rc = srpc_add_service(sv);
1698                 LASSERT(rc != -EBUSY);
1699                 if (rc) {
1700                         CWARN("Failed to add %s service: %d\n",
1701                               sv->sv_name, rc);
1702                         error = rc;
1703                 }
1704
1705                 /* about to sfw_shutdown, no need to add buffer */
1706                 if (error)
1707                         continue;
1708
1709                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1710                 if (rc) {
1711                         CWARN("Failed to reserve enough buffers: service %s, %d needed: %d\n",
1712                               sv->sv_name, sv->sv_wi_total, rc);
1713                         error = -ENOMEM;
1714                 }
1715         }
1716
1717         if (error)
1718                 sfw_shutdown();
1719         return error;
1720 }
1721
1722 void
1723 sfw_shutdown(void)
1724 {
1725         struct srpc_service *sv;
1726         struct sfw_test_case    *tsc;
1727         int i;
1728
1729         spin_lock(&sfw_data.fw_lock);
1730
1731         sfw_data.fw_shuttingdown = 1;
1732         lst_wait_until(!sfw_data.fw_active_srpc, sfw_data.fw_lock,
1733                        "waiting for active RPC to finish.\n");
1734
1735         if (sfw_del_session_timer())
1736                 lst_wait_until(!sfw_data.fw_session, sfw_data.fw_lock,
1737                                "waiting for session timer to explode.\n");
1738
1739         sfw_deactivate_session();
1740         lst_wait_until(!atomic_read(&sfw_data.fw_nzombies),
1741                        sfw_data.fw_lock,
1742                        "waiting for %d zombie sessions to die.\n",
1743                        atomic_read(&sfw_data.fw_nzombies));
1744
1745         spin_unlock(&sfw_data.fw_lock);
1746
1747         for (i = 0; ; i++) {
1748                 sv = &sfw_services[i];
1749                 if (!sv->sv_name)
1750                         break;
1751
1752                 srpc_shutdown_service(sv);
1753                 srpc_remove_service(sv);
1754         }
1755
1756         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1757                 sv = tsc->tsc_srv_service;
1758                 srpc_shutdown_service(sv);
1759                 srpc_remove_service(sv);
1760         }
1761
1762         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1763                 struct srpc_client_rpc *rpc;
1764
1765                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1766                                  struct srpc_client_rpc, crpc_list);
1767                 list_del(&rpc->crpc_list);
1768
1769                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1770         }
1771
1772         for (i = 0; ; i++) {
1773                 sv = &sfw_services[i];
1774                 if (!sv->sv_name)
1775                         break;
1776
1777                 srpc_wait_service_shutdown(sv);
1778         }
1779
1780         while (!list_empty(&sfw_data.fw_tests)) {
1781                 tsc = list_entry(sfw_data.fw_tests.next,
1782                                  struct sfw_test_case, tsc_list);
1783
1784                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1785
1786                 list_del(&tsc->tsc_list);
1787                 LIBCFS_FREE(tsc, sizeof(*tsc));
1788         }
1789 }