lnet/selftest/conrpc.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lnet/selftest/conctl.c
  37  *
  38  * Console framework rpcs
  39  *
  40  * Author: Liang Zhen <liangzhen@clusterfs.com>
  41  */
  42
  43 #ifdef __KERNEL__
  44
  45 #include <libcfs/libcfs.h>
  46 #include <lnet/lib-lnet.h>
  47 #include "timer.h"
  48 #include "conrpc.h"
  49 #include "console.h"
  50
  51 void lstcon_rpc_stat_reply(int, srpc_msg_t *,
  52                            lstcon_node_t *, lstcon_trans_stat_t *);
  53
  54 static void
  55 lstcon_rpc_done(srpc_client_rpc_t *rpc)
  56 {
  57         lstcon_rpc_t *crpc = (lstcon_rpc_t *)rpc->crpc_priv;
  58
  59         LASSERT (crpc != NULL && rpc == crpc->crp_rpc);
  60         LASSERT (crpc->crp_posted && !crpc->crp_finished);
  61
  62         spin_lock(&rpc->crpc_lock);
  63
  64         if (crpc->crp_trans == NULL) {
  65                 /* Orphan RPC is not in any transaction,
  66                  * I'm just a poor body and nobody loves me */
  67                 spin_unlock(&rpc->crpc_lock);
  68
  69                 /* release it */
  70                 lstcon_rpc_put(crpc);
  71                 return;
  72         }
  73
  74         /* not an orphan RPC */
  75         crpc->crp_finished = 1;
  76
  77         if (crpc->crp_stamp == 0) {
  78                 /* not aborted */
  79                 LASSERT (crpc->crp_status == 0);
  80
  81                 crpc->crp_stamp  = cfs_time_current();
  82                 crpc->crp_status = rpc->crpc_status;
  83         }
  84
  85         /* wakeup (transaction)thread if I'm the last RPC in the transaction */
  86         if (atomic_dec_and_test(&crpc->crp_trans->tas_remaining))
  87                 cfs_waitq_signal(&crpc->crp_trans->tas_waitq);
  88
  89         spin_unlock(&rpc->crpc_lock);
  90 }
  91
  92 int
  93 lstcon_rpc_init(lstcon_node_t *nd, int service,
  94                 int npg, int cached, lstcon_rpc_t *crpc)
  95 {
  96
  97         crpc->crp_rpc = sfw_create_rpc(nd->nd_id, service,
  98                                        npg, npg * CFS_PAGE_SIZE,
  99                                        lstcon_rpc_done, (void *)crpc);
 100         if (crpc->crp_rpc == NULL)
 101                 return -ENOMEM;
 102
 103         crpc->crp_trans    = NULL;
 104         crpc->crp_node     = nd;
 105         crpc->crp_posted   = 0;
 106         crpc->crp_finished = 0;
 107         crpc->crp_unpacked = 0;
 108         crpc->crp_status   = 0;
 109         crpc->crp_stamp    = 0;
 110         crpc->crp_static   = !cached;
 111         CFS_INIT_LIST_HEAD(&crpc->crp_link);
 112
 113         atomic_inc(&console_session.ses_rpc_counter);
 114
 115         return 0;
 116 }
 117
 118 int
 119 lstcon_rpc_prep(lstcon_node_t *nd, int service,
 120                 int npg, lstcon_rpc_t **crpcpp)
 121 {
 122         lstcon_rpc_t  *crpc = NULL;
 123         int            rc;
 124
 125         spin_lock(&console_session.ses_rpc_lock);
 126
 127         if (!list_empty(&console_session.ses_rpc_freelist)) {
 128                 crpc = list_entry(console_session.ses_rpc_freelist.next,
 129                                   lstcon_rpc_t, crp_link);
 130                 list_del_init(&crpc->crp_link);
 131         }
 132
 133         spin_unlock(&console_session.ses_rpc_lock);
 134
 135         if (crpc == NULL) {
 136                 LIBCFS_ALLOC(crpc, sizeof(*crpc));
 137                 if (crpc == NULL)
 138                         return -ENOMEM;
 139         }
 140
 141         rc = lstcon_rpc_init(nd, service, npg, 1, crpc);
 142         if (rc == 0) {
 143                 *crpcpp = crpc;
 144                 return 0;
 145         }
 146
 147         LIBCFS_FREE(crpc, sizeof(*crpc));
 148
 149         return rc;
 150 }
 151
 152 void
 153 lstcon_rpc_put(lstcon_rpc_t *crpc)
 154 {
 155         srpc_bulk_t *bulk = &crpc->crp_rpc->crpc_bulk;
 156         int          i;
 157
 158         LASSERT (list_empty(&crpc->crp_link));
 159
 160         for (i = 0; i < bulk->bk_niov; i++) {
 161                 if (bulk->bk_iovs[i].kiov_page == NULL)
 162                         continue;
 163
 164                 cfs_free_page(bulk->bk_iovs[i].kiov_page);
 165         }
 166
 167         srpc_client_rpc_decref(crpc->crp_rpc);
 168
 169         if (crpc->crp_static) {
 170                 /* Static RPC, not allocated */
 171                 memset(crpc, 0, sizeof(*crpc));
 172                 crpc->crp_static = 1;
 173
 174         } else {
 175                 spin_lock(&console_session.ses_rpc_lock);
 176
 177                 list_add(&crpc->crp_link, &console_session.ses_rpc_freelist);
 178
 179                 spin_unlock(&console_session.ses_rpc_lock);
 180         }
 181
 182         /* RPC is not alive now */
 183         atomic_dec(&console_session.ses_rpc_counter);
 184 }
 185
 186 void
 187 lstcon_rpc_post(lstcon_rpc_t *crpc)
 188 {
 189         lstcon_rpc_trans_t *trans = crpc->crp_trans;
 190
 191         LASSERT (trans != NULL);
 192
 193         atomic_inc(&trans->tas_remaining);
 194         crpc->crp_posted = 1;
 195
 196         sfw_post_rpc(crpc->crp_rpc);
 197 }
 198
 199 static char *
 200 lstcon_rpc_trans_name(int transop)
 201 {
 202         if (transop == LST_TRANS_SESNEW)
 203                 return "SESNEW";
 204
 205         if (transop == LST_TRANS_SESEND)
 206                 return "SESEND";
 207
 208         if (transop == LST_TRANS_SESQRY)
 209                 return "SESQRY";
 210
 211         if (transop == LST_TRANS_SESPING)
 212                 return "SESPING";
 213
 214         if (transop == LST_TRANS_TSBCLIADD)
 215                 return "TSBCLIADD";
 216
 217         if (transop == LST_TRANS_TSBSRVADD)
 218                 return "TSBSRVADD";
 219
 220         if (transop == LST_TRANS_TSBRUN)
 221                 return "TSBRUN";
 222
 223         if (transop == LST_TRANS_TSBSTOP)
 224                 return "TSBSTOP";
 225
 226         if (transop == LST_TRANS_TSBCLIQRY)
 227                 return "TSBCLIQRY";
 228
 229         if (transop == LST_TRANS_TSBSRVQRY)
 230                 return "TSBSRVQRY";
 231
 232         if (transop == LST_TRANS_STATQRY)
 233                 return "STATQRY";
 234
 235         return "Unknown";
 236 }
 237
 238 int
 239 lstcon_rpc_trans_prep(struct list_head *translist,
 240                       int transop, lstcon_rpc_trans_t **transpp)
 241 {
 242         lstcon_rpc_trans_t *trans;
 243
 244         if (translist != NULL) {
 245                 list_for_each_entry(trans, translist, tas_link) {
 246                         /* Can't enqueue two private transaction on
 247                          * the same object */
 248                         if ((trans->tas_opc & transop) == LST_TRANS_PRIVATE)
 249                                 return -EPERM;
 250                 }
 251         }
 252
 253         /* create a trans group */
 254         LIBCFS_ALLOC(trans, sizeof(*trans));
 255         if (trans == NULL)
 256                 return -ENOMEM;
 257
 258         trans->tas_opc = transop;
 259
 260         if (translist == NULL)
 261                 CFS_INIT_LIST_HEAD(&trans->tas_olink);
 262         else
 263                 list_add_tail(&trans->tas_olink, translist);
 264
 265         list_add_tail(&trans->tas_link, &console_session.ses_trans_list);
 266
 267         CFS_INIT_LIST_HEAD(&trans->tas_rpcs_list);
 268         atomic_set(&trans->tas_remaining, 0);
 269         cfs_waitq_init(&trans->tas_waitq);
 270
 271         *transpp = trans;
 272
 273         return 0;
 274 }
 275
 276 void
 277 lstcon_rpc_trans_addreq(lstcon_rpc_trans_t *trans, lstcon_rpc_t *crpc)
 278 {
 279         list_add_tail(&crpc->crp_link, &trans->tas_rpcs_list);
 280         crpc->crp_trans = trans;
 281 }
 282
 283 void
 284 lstcon_rpc_trans_abort(lstcon_rpc_trans_t *trans, int error)
 285 {
 286         srpc_client_rpc_t *rpc;
 287         lstcon_rpc_t      *crpc;
 288         lstcon_node_t     *nd;
 289
 290         list_for_each_entry (crpc, &trans->tas_rpcs_list, crp_link) {
 291                 rpc = crpc->crp_rpc;
 292
 293                 spin_lock(&rpc->crpc_lock);
 294
 295                 if (!crpc->crp_posted || crpc->crp_stamp != 0) {
 296                         /* rpc done or aborted already */
 297                         spin_unlock(&rpc->crpc_lock);
 298                         continue;
 299                 }
 300
 301                 crpc->crp_stamp  = cfs_time_current();
 302                 crpc->crp_status = error;
 303
 304                 spin_unlock(&rpc->crpc_lock);
 305
 306                 sfw_abort_rpc(rpc);
 307
 308                 if  (error != ETIMEDOUT)
 309                         continue;
 310
 311                 nd = crpc->crp_node;
 312                 if (cfs_time_after(nd->nd_stamp, crpc->crp_stamp))
 313                         continue;
 314
 315                 nd->nd_stamp = crpc->crp_stamp;
 316                 nd->nd_state = LST_NODE_DOWN;
 317         }
 318 }
 319
 320 static int
 321 lstcon_rpc_trans_check(lstcon_rpc_trans_t *trans)
 322 {
 323         if (console_session.ses_shutdown &&
 324             !list_empty(&trans->tas_olink)) /* It's not an end session RPC */
 325                 return 1;
 326
 327         return (atomic_read(&trans->tas_remaining) == 0) ? 1: 0;
 328 }
 329
 330 int
 331 lstcon_rpc_trans_postwait(lstcon_rpc_trans_t *trans, int timeout)
 332 {
 333         lstcon_rpc_t  *crpc;
 334         int            rc;
 335
 336         if (list_empty(&trans->tas_rpcs_list))
 337                 return 0;
 338
 339         if (timeout < LST_TRANS_MIN_TIMEOUT)
 340                 timeout = LST_TRANS_MIN_TIMEOUT;
 341
 342         CDEBUG(D_NET, "Transaction %s started\n",
 343                lstcon_rpc_trans_name(trans->tas_opc));
 344
 345         /* post all requests */
 346         list_for_each_entry (crpc, &trans->tas_rpcs_list, crp_link) {
 347                 LASSERT (!crpc->crp_posted);
 348
 349                 lstcon_rpc_post(crpc);
 350         }
 351
 352         mutex_up(&console_session.ses_mutex);
 353
 354         rc = cfs_waitq_wait_event_interruptible_timeout(trans->tas_waitq,
 355                                               lstcon_rpc_trans_check(trans),
 356                                               timeout * HZ);
 357
 358         rc = (rc > 0)? 0: ((rc < 0)? -EINTR: -ETIMEDOUT);
 359
 360         mutex_down(&console_session.ses_mutex);
 361
 362         if (console_session.ses_shutdown)
 363                 rc = -ESHUTDOWN;
 364
 365         if (rc != 0) {
 366                 /* treat short timeout as canceled */
 367                 if (rc == -ETIMEDOUT && timeout < LST_TRANS_MIN_TIMEOUT * 2)
 368                         rc = -EINTR;
 369
 370                 lstcon_rpc_trans_abort(trans, rc);
 371         }
 372
 373         CDEBUG(D_NET, "Transaction %s stopped: %d\n",
 374                lstcon_rpc_trans_name(trans->tas_opc), rc);
 375
 376         lstcon_rpc_trans_stat(trans, lstcon_trans_stat());
 377
 378         return rc;
 379 }
 380
 381 int
 382 lstcon_rpc_get_reply(lstcon_rpc_t *crpc, srpc_msg_t **msgpp)
 383 {
 384         lstcon_node_t        *nd  = crpc->crp_node;
 385         srpc_client_rpc_t    *rpc = crpc->crp_rpc;
 386         srpc_generic_reply_t *rep;
 387
 388         LASSERT (nd != NULL && rpc != NULL);
 389         LASSERT (crpc->crp_stamp != 0);
 390
 391         if (crpc->crp_status != 0) {
 392                 *msgpp = NULL;
 393                 return crpc->crp_status;
 394         }
 395
 396         *msgpp = &rpc->crpc_replymsg;
 397         if (!crpc->crp_unpacked) {
 398                 sfw_unpack_message(*msgpp);
 399                 crpc->crp_unpacked = 1;
 400         }
 401
 402         if (cfs_time_after(nd->nd_stamp, crpc->crp_stamp))
 403                 return 0;
 404
 405         nd->nd_stamp = crpc->crp_stamp;
 406         rep = &(*msgpp)->msg_body.reply;
 407
 408         if (rep->sid.ses_nid == LNET_NID_ANY)
 409                 nd->nd_state = LST_NODE_UNKNOWN;
 410         else if (lstcon_session_match(rep->sid))
 411                 nd->nd_state = LST_NODE_ACTIVE;
 412         else
 413                 nd->nd_state = LST_NODE_BUSY;
 414
 415         return 0;
 416 }
 417
 418 void
 419 lstcon_rpc_trans_stat(lstcon_rpc_trans_t *trans, lstcon_trans_stat_t *stat)
 420 {
 421         lstcon_rpc_t      *crpc;
 422         srpc_client_rpc_t *rpc;
 423         srpc_msg_t        *rep;
 424         int                error;
 425
 426         LASSERT (stat != NULL);
 427
 428         memset(stat, 0, sizeof(*stat));
 429
 430         list_for_each_entry(crpc, &trans->tas_rpcs_list, crp_link) {
 431                 lstcon_rpc_stat_total(stat, 1);
 432
 433                 rpc = crpc->crp_rpc;
 434
 435                 LASSERT (crpc->crp_stamp != 0);
 436
 437                 error = lstcon_rpc_get_reply(crpc, &rep);
 438                 if (error != 0) {
 439                         lstcon_rpc_stat_failure(stat, 1);
 440                         if (stat->trs_rpc_errno == 0)
 441                                 stat->trs_rpc_errno = -error;
 442
 443                         continue;
 444                 }
 445
 446                 lstcon_rpc_stat_success(stat, 1);
 447
 448                 lstcon_rpc_stat_reply(trans->tas_opc, rep,
 449                                       crpc->crp_node, stat);
 450         }
 451
 452         CDEBUG(D_NET, "transaction %s : success %d, failure %d, total %d, "
 453                       "RPC error(%d), Framework error(%d)\n",
 454                lstcon_rpc_trans_name(trans->tas_opc),
 455                lstcon_rpc_stat_success(stat, 0),
 456                lstcon_rpc_stat_failure(stat, 0),
 457                lstcon_rpc_stat_total(stat, 0),
 458                stat->trs_rpc_errno, stat->trs_fwk_errno);
 459
 460         return;
 461 }
 462
 463 int
 464 lstcon_rpc_trans_interpreter(lstcon_rpc_trans_t *trans,
 465                              struct list_head *head_up,
 466                              lstcon_rpc_readent_func_t readent)
 467 {
 468         struct list_head      tmp;
 469         struct list_head     *next;
 470         lstcon_rpc_ent_t     *ent;
 471         srpc_generic_reply_t *rep;
 472         srpc_client_rpc_t    *rpc;
 473         lstcon_rpc_t         *crpc;
 474         srpc_msg_t           *msg;
 475         lstcon_node_t        *nd;
 476         cfs_duration_t        dur;
 477         struct timeval        tv;
 478         int                   error;
 479
 480         LASSERT (head_up != NULL);
 481
 482         next = head_up;
 483
 484         list_for_each_entry(crpc, &trans->tas_rpcs_list, crp_link) {
 485                 if (copy_from_user(&tmp, next, sizeof(struct list_head)))
 486                         return -EFAULT;
 487
 488                 if (tmp.next == head_up)
 489                         return 0;
 490
 491                 next = tmp.next;
 492
 493                 ent = list_entry(next, lstcon_rpc_ent_t, rpe_link);
 494
 495                 rpc = crpc->crp_rpc;
 496
 497                 LASSERT (crpc->crp_stamp != 0);
 498
 499                 error = lstcon_rpc_get_reply(crpc, &msg);
 500
 501                 nd = crpc->crp_node;
 502
 503                 dur = cfs_time_sub(crpc->crp_stamp,
 504                                    console_session.ses_id.ses_stamp);
 505                 cfs_duration_usec(dur, &tv);
 506
 507                 if (copy_to_user(&ent->rpe_peer,
 508                                  &nd->nd_id, sizeof(lnet_process_id_t)) ||
 509                     copy_to_user(&ent->rpe_stamp, &tv, sizeof(tv)) ||
 510                     copy_to_user(&ent->rpe_state,
 511                                  &nd->nd_state, sizeof(nd->nd_state)) ||
 512                     copy_to_user(&ent->rpe_rpc_errno, &error, sizeof(error)))
 513                         return -EFAULT;
 514
 515                 if (error != 0)
 516                         continue;
 517
 518                 /* RPC is done */
 519                 rep = (srpc_generic_reply_t *)&msg->msg_body.reply;
 520
 521                 if (copy_to_user(&ent->rpe_sid,
 522                                  &rep->sid, sizeof(lst_sid_t)) ||
 523                     copy_to_user(&ent->rpe_fwk_errno,
 524                                  &rep->status, sizeof(rep->status)))
 525                         return -EFAULT;
 526
 527                 if (readent == NULL)
 528                         continue;
 529
 530                 if ((error = readent(trans->tas_opc, msg, ent)) != 0)
 531                         return error;
 532         }
 533
 534         return 0;
 535 }
 536
 537 void
 538 lstcon_rpc_trans_destroy(lstcon_rpc_trans_t *trans)
 539 {
 540         srpc_client_rpc_t *rpc;
 541         lstcon_rpc_t      *crpc;
 542         lstcon_rpc_t      *tmp;
 543         int                count = 0;
 544
 545         list_for_each_entry_safe(crpc, tmp,
 546                                  &trans->tas_rpcs_list, crp_link) {
 547                 rpc = crpc->crp_rpc;
 548
 549                 spin_lock(&rpc->crpc_lock);
 550
 551                 /* free it if not posted or finished already */
 552                 if (!crpc->crp_posted || crpc->crp_finished) {
 553                         spin_unlock(&rpc->crpc_lock);
 554
 555                         list_del_init(&crpc->crp_link);
 556                         lstcon_rpc_put(crpc);
 557
 558                         continue;
 559                 }
 560
 561                 /* rpcs can be still not callbacked (even LNetMDUnlink is called)
 562                  * because huge timeout for inaccessible network, don't make
 563                  * user wait for them, just abandon them, they will be recycled
 564                  * in callback */
 565
 566                 LASSERT (crpc->crp_status != 0);
 567
 568                 crpc->crp_node  = NULL;
 569                 crpc->crp_trans = NULL;
 570                 list_del_init(&crpc->crp_link);
 571                 count ++;
 572
 573                 spin_unlock(&rpc->crpc_lock);
 574
 575                 atomic_dec(&trans->tas_remaining);
 576         }
 577
 578         LASSERT (atomic_read(&trans->tas_remaining) == 0);
 579
 580         list_del(&trans->tas_link);
 581         if (!list_empty(&trans->tas_olink))
 582                 list_del(&trans->tas_olink);
 583
 584         CDEBUG(D_NET, "Transaction %s destroyed with %d pending RPCs\n",
 585                lstcon_rpc_trans_name(trans->tas_opc), count);
 586
 587         LIBCFS_FREE(trans, sizeof(*trans));
 588
 589         return;
 590 }
 591
 592 int
 593 lstcon_sesrpc_prep(lstcon_node_t *nd, int transop, lstcon_rpc_t **crpc)
 594 {
 595         srpc_mksn_reqst_t *msrq;
 596         srpc_rmsn_reqst_t *rsrq;
 597         int                rc;
 598
 599         switch (transop) {
 600         case LST_TRANS_SESNEW:
 601                 rc = lstcon_rpc_prep(nd, SRPC_SERVICE_MAKE_SESSION, 0, crpc);
 602                 if (rc != 0)
 603                         return rc;
 604
 605                 msrq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.mksn_reqst;
 606                 msrq->mksn_sid     = console_session.ses_id;
 607                 msrq->mksn_force   = console_session.ses_force;
 608                 strncpy(msrq->mksn_name, console_session.ses_name,
 609                         strlen(console_session.ses_name));
 610                 break;
 611
 612         case LST_TRANS_SESEND:
 613                 rc = lstcon_rpc_prep(nd, SRPC_SERVICE_REMOVE_SESSION, 0, crpc);
 614                 if (rc != 0)
 615                         return rc;
 616
 617                 rsrq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.rmsn_reqst;
 618                 rsrq->rmsn_sid = console_session.ses_id;
 619                 break;
 620
 621         default:
 622                 LBUG();
 623         }
 624
 625         return 0;
 626 }
 627
 628 int
 629 lstcon_dbgrpc_prep(lstcon_node_t *nd, lstcon_rpc_t **crpc)
 630 {
 631         srpc_debug_reqst_t *drq;
 632         int                 rc;
 633
 634         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_DEBUG, 0, crpc);
 635         if (rc != 0)
 636                 return rc;
 637
 638         drq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.dbg_reqst;
 639
 640         drq->dbg_sid   = console_session.ses_id;
 641         drq->dbg_flags = 0;
 642
 643         return rc;
 644 }
 645
 646 int
 647 lstcon_batrpc_prep(lstcon_node_t *nd, int transop,
 648                    lstcon_tsb_hdr_t *tsb, lstcon_rpc_t **crpc)
 649 {
 650         lstcon_batch_t     *batch;
 651         srpc_batch_reqst_t *brq;
 652         int                 rc;
 653
 654         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_BATCH, 0, crpc);
 655         if (rc != 0)
 656                 return rc;
 657
 658         brq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.bat_reqst;
 659
 660         brq->bar_sid     = console_session.ses_id;
 661         brq->bar_bid     = tsb->tsb_id;
 662         brq->bar_testidx = tsb->tsb_index;
 663         brq->bar_opc     = transop == LST_TRANS_TSBRUN ? SRPC_BATCH_OPC_RUN :
 664                            (transop == LST_TRANS_TSBSTOP ? SRPC_BATCH_OPC_STOP:
 665                             SRPC_BATCH_OPC_QUERY);
 666
 667         if (transop != LST_TRANS_TSBRUN &&
 668             transop != LST_TRANS_TSBSTOP)
 669                 return 0;
 670
 671         LASSERT (tsb->tsb_index == 0);
 672
 673         batch = (lstcon_batch_t *)tsb;
 674         brq->bar_arg = batch->bat_arg;
 675
 676         return 0;
 677 }
 678
 679 int
 680 lstcon_statrpc_prep(lstcon_node_t *nd, lstcon_rpc_t **crpc)
 681 {
 682         srpc_stat_reqst_t *srq;
 683         int                rc;
 684
 685         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_QUERY_STAT, 0, crpc);
 686         if (rc != 0)
 687                 return rc;
 688
 689         srq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.stat_reqst;
 690
 691         srq->str_sid  = console_session.ses_id;
 692         srq->str_type = 0; /* XXX remove it */
 693
 694         return 0;
 695 }
 696
 697 lnet_process_id_t *
 698 lstcon_next_id(int idx, int nkiov, lnet_kiov_t *kiov)
 699 {
 700         lnet_process_id_t *pid;
 701         int                i;
 702
 703         i = idx / (CFS_PAGE_SIZE / sizeof(lnet_process_id_t));
 704
 705         LASSERT (i < nkiov);
 706
 707         pid = (lnet_process_id_t *)cfs_page_address(kiov[i].kiov_page);
 708
 709         return &pid[idx % (CFS_PAGE_SIZE / sizeof(lnet_process_id_t))];
 710 }
 711
 712 int
 713 lstcon_dstnodes_prep(lstcon_group_t *grp, int idx,
 714                      int dist, int span, int nkiov, lnet_kiov_t *kiov)
 715 {
 716         lnet_process_id_t *pid;
 717         lstcon_ndlink_t   *ndl;
 718         lstcon_node_t     *nd;
 719         int                start;
 720         int                end;
 721         int                i = 0;
 722
 723         LASSERT (dist >= 1);
 724         LASSERT (span >= 1);
 725         LASSERT (grp->grp_nnode >= 1);
 726
 727         if (span > grp->grp_nnode)
 728                 return -EINVAL;
 729
 730         start = ((idx / dist) * span) % grp->grp_nnode;
 731         end   = ((idx / dist) * span + span - 1) % grp->grp_nnode;
 732
 733         list_for_each_entry(ndl, &grp->grp_ndl_list, ndl_link) {
 734                 nd = ndl->ndl_node;
 735                 if (i < start) {
 736                         i ++;
 737                         continue;
 738                 }
 739
 740                 if (i > (end >= start ? end: grp->grp_nnode))
 741                         break;
 742
 743                 pid = lstcon_next_id((i - start), nkiov, kiov);
 744                 *pid = nd->nd_id;
 745                 i++;
 746         }
 747
 748         if (start <= end) /* done */
 749                 return 0;
 750
 751         list_for_each_entry(ndl, &grp->grp_ndl_list, ndl_link) {
 752                 if (i > grp->grp_nnode + end)
 753                         break;
 754
 755                 nd = ndl->ndl_node;
 756                 pid = lstcon_next_id((i - start), nkiov, kiov);
 757                 *pid = nd->nd_id;
 758                 i++;
 759         }
 760
 761         return 0;
 762 }
 763
 764 int
 765 lstcon_pingrpc_prep(lst_test_ping_param_t *param, srpc_test_reqst_t *req)
 766 {
 767         test_ping_req_t *prq = &req->tsr_u.ping;
 768
 769         prq->png_size   = param->png_size;
 770         prq->png_flags  = param->png_flags;
 771         /* TODO dest */
 772         return 0;
 773 }
 774
 775 int
 776 lstcon_bulkrpc_prep(lst_test_bulk_param_t *param, srpc_test_reqst_t *req)
 777 {
 778         test_bulk_req_t *brq = &req->tsr_u.bulk;
 779
 780         brq->blk_opc    = param->blk_opc;
 781         brq->blk_npg    = (param->blk_size + CFS_PAGE_SIZE - 1) / CFS_PAGE_SIZE;
 782         brq->blk_flags  = param->blk_flags;
 783
 784         return 0;
 785 }
 786
 787 int
 788 lstcon_testrpc_prep(lstcon_node_t *nd, int transop,
 789                     lstcon_test_t *test, lstcon_rpc_t **crpc)
 790 {
 791         lstcon_group_t    *sgrp = test->tes_src_grp;
 792         lstcon_group_t    *dgrp = test->tes_dst_grp;
 793         srpc_test_reqst_t *trq;
 794         srpc_bulk_t       *bulk;
 795         int                i;
 796         int                n  = 0;
 797         int                rc = 0;
 798
 799         if (transop == LST_TRANS_TSBCLIADD)
 800                 n = sfw_id_pages(test->tes_span);
 801
 802         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_TEST, n, crpc);
 803         if (rc != 0)
 804                 return rc;
 805
 806         trq  = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.tes_reqst;
 807
 808         if (transop == LST_TRANS_TSBSRVADD) {
 809                 int ndist = (sgrp->grp_nnode + test->tes_dist - 1) / test->tes_dist;
 810                 int nspan = (dgrp->grp_nnode + test->tes_span - 1) / test->tes_span;
 811                 int nmax = (ndist + nspan - 1) / nspan;
 812
 813                 trq->tsr_ndest = 0;
 814                 trq->tsr_loop  = nmax * test->tes_dist * test->tes_concur;
 815
 816         } else {
 817                 bulk = &(*crpc)->crp_rpc->crpc_bulk;
 818
 819                 for (i = 0; i < n; i++) {
 820                         bulk->bk_iovs[i].kiov_offset = 0;
 821                         bulk->bk_iovs[i].kiov_len    = CFS_PAGE_SIZE;
 822                         bulk->bk_iovs[i].kiov_page   = cfs_alloc_page(CFS_ALLOC_STD);
 823
 824                         if (bulk->bk_iovs[i].kiov_page != NULL)
 825                                 continue;
 826
 827                         lstcon_rpc_put(*crpc);
 828                         return -ENOMEM;
 829                 }
 830
 831                 bulk->bk_sink = 0;
 832
 833                 LASSERT (transop == LST_TRANS_TSBCLIADD);
 834
 835                 rc = lstcon_dstnodes_prep(test->tes_dst_grp,
 836                                           test->tes_cliidx++, test->tes_dist,
 837                                           test->tes_span, n, &bulk->bk_iovs[0]);
 838                 if (rc != 0) {
 839                         lstcon_rpc_put(*crpc);
 840                         return rc;
 841                 }
 842
 843                 trq->tsr_ndest = test->tes_span;
 844                 trq->tsr_loop  = test->tes_loop;
 845         }
 846
 847         trq->tsr_sid        = console_session.ses_id;
 848         trq->tsr_bid        = test->tes_hdr.tsb_id;
 849         trq->tsr_concur     = test->tes_concur;
 850         trq->tsr_is_client  = (transop == LST_TRANS_TSBCLIADD) ? 1 : 0;
 851         trq->tsr_stop_onerr = test->tes_stop_onerr;
 852
 853         switch (test->tes_type) {
 854         case LST_TEST_PING:
 855                 trq->tsr_service = SRPC_SERVICE_PING;
 856                 rc = lstcon_pingrpc_prep((lst_test_ping_param_t *)&test->tes_param[0], trq);
 857                 break;
 858         case LST_TEST_BULK:
 859                 trq->tsr_service = SRPC_SERVICE_BRW;
 860                 rc = lstcon_bulkrpc_prep((lst_test_bulk_param_t *)&test->tes_param[0], trq);
 861                 break;
 862         default:
 863                 LBUG();
 864                 break;
 865         }
 866
 867         return rc;
 868 }
 869
 870 void
 871 lstcon_rpc_stat_reply(int transop, srpc_msg_t *msg,
 872                       lstcon_node_t *nd, lstcon_trans_stat_t *stat)
 873 {
 874         srpc_mksn_reply_t  *mksn_rep;
 875         srpc_rmsn_reply_t  *rmsn_rep;
 876         srpc_debug_reply_t *dbg_rep;
 877         srpc_batch_reply_t *bat_rep;
 878         srpc_test_reply_t  *test_rep;
 879         srpc_stat_reply_t  *stat_rep;
 880         int                 errno = 0;
 881
 882         switch (transop) {
 883         case LST_TRANS_SESNEW:
 884                 mksn_rep = &msg->msg_body.mksn_reply;
 885
 886                 if (mksn_rep->mksn_status == 0) {
 887                         lstcon_sesop_stat_success(stat, 1);
 888                         /* session timeout on remote node */
 889                         nd->nd_timeout = mksn_rep->mksn_timeout;
 890                         return;
 891                 }
 892
 893                 LASSERT (mksn_rep->mksn_status == EBUSY ||
 894                          mksn_rep->mksn_status == EINVAL);
 895
 896                 lstcon_sesop_stat_failure(stat, 1);
 897                 errno = mksn_rep->mksn_status;
 898                 break;
 899
 900         case LST_TRANS_SESEND:
 901                 rmsn_rep = &msg->msg_body.rmsn_reply;
 902                 /* ESRCH is not an error for end session */
 903                 if (rmsn_rep->rmsn_status == 0 ||
 904                     rmsn_rep->rmsn_status == ESRCH) {
 905                         lstcon_sesop_stat_success(stat, 1);
 906                         return;
 907                 }
 908
 909                 LASSERT (rmsn_rep->rmsn_status == EBUSY ||
 910                          rmsn_rep->rmsn_status == EINVAL);
 911
 912                 lstcon_sesop_stat_failure(stat, 1);
 913                 errno = rmsn_rep->rmsn_status;
 914                 break;
 915
 916         case LST_TRANS_SESQRY:
 917         case LST_TRANS_SESPING:
 918                 dbg_rep = &msg->msg_body.dbg_reply;
 919
 920                 if (dbg_rep->dbg_status == ESRCH) {
 921                         lstcon_sesqry_stat_unknown(stat, 1);
 922                         return;
 923                 }
 924
 925                 LASSERT (dbg_rep->dbg_status == 0);
 926
 927                 if (lstcon_session_match(dbg_rep->dbg_sid))
 928                         lstcon_sesqry_stat_active(stat, 1);
 929                 else
 930                         lstcon_sesqry_stat_busy(stat, 1);
 931                 return;
 932
 933         case LST_TRANS_TSBRUN:
 934         case LST_TRANS_TSBSTOP:
 935                 bat_rep = &msg->msg_body.bat_reply;
 936
 937                 if (bat_rep->bar_status == 0) {
 938                         lstcon_tsbop_stat_success(stat, 1);
 939                         return;
 940                 }
 941
 942                 if (bat_rep->bar_status == EPERM &&
 943                     transop == LST_TRANS_TSBSTOP) {
 944                         lstcon_tsbop_stat_success(stat, 1);
 945                         return;
 946                 }
 947
 948                 lstcon_tsbop_stat_failure(stat, 1);
 949                 errno = bat_rep->bar_status;
 950                 break;
 951
 952         case LST_TRANS_TSBCLIQRY:
 953         case LST_TRANS_TSBSRVQRY:
 954                 bat_rep = &msg->msg_body.bat_reply;
 955
 956                 if (bat_rep->bar_active != 0)
 957                         lstcon_tsbqry_stat_run(stat, 1);
 958                 else
 959                         lstcon_tsbqry_stat_idle(stat, 1);
 960
 961                 if (bat_rep->bar_status == 0)
 962                         return;
 963
 964                 lstcon_tsbqry_stat_failure(stat, 1);
 965                 errno = bat_rep->bar_status;
 966                 break;
 967
 968         case LST_TRANS_TSBCLIADD:
 969         case LST_TRANS_TSBSRVADD:
 970                 test_rep = &msg->msg_body.tes_reply;
 971
 972                 if (test_rep->tsr_status == 0) {
 973                         lstcon_tsbop_stat_success(stat, 1);
 974                         return;
 975                 }
 976
 977                 lstcon_tsbop_stat_failure(stat, 1);
 978                 errno = test_rep->tsr_status;
 979                 break;
 980
 981         case LST_TRANS_STATQRY:
 982                 stat_rep = &msg->msg_body.stat_reply;
 983
 984                 if (stat_rep->str_status == 0) {
 985                         lstcon_statqry_stat_success(stat, 1);
 986                         return;
 987                 }
 988
 989                 lstcon_statqry_stat_failure(stat, 1);
 990                 errno = stat_rep->str_status;
 991                 break;
 992
 993         default:
 994                 LBUG();
 995         }
 996
 997         if (stat->trs_fwk_errno == 0)
 998                 stat->trs_fwk_errno = errno;
 999
1000         return;
1001 }
1002
1003 int
1004 lstcon_rpc_trans_ndlist(struct list_head *ndlist,
1005                         struct list_head *translist, int transop,
1006                         void *arg, lstcon_rpc_cond_func_t condition,
1007                         lstcon_rpc_trans_t **transpp)
1008 {
1009         lstcon_rpc_trans_t *trans;
1010         lstcon_ndlink_t    *ndl;
1011         lstcon_node_t      *nd;
1012         lstcon_rpc_t       *rpc;
1013         int                 rc;
1014
1015         /* Creating session RPG for list of nodes */
1016
1017         rc = lstcon_rpc_trans_prep(translist, transop, &trans);
1018         if (rc != 0) {
1019                 CERROR("Can't create transaction %d: %d\n", transop, rc);
1020                 return rc;
1021         }
1022
1023         list_for_each_entry(ndl, ndlist, ndl_link) {
1024                 rc = condition == NULL ? 1 :
1025                      condition(transop, ndl->ndl_node, arg);
1026
1027                 if (rc == 0)
1028                         continue;
1029
1030                 if (rc < 0) {
1031                         CDEBUG(D_NET, "Condition error while creating RPC "
1032                                       " for transaction %d: %d\n", transop, rc);
1033                         break;
1034                 }
1035
1036                 nd = ndl->ndl_node;
1037
1038                 switch (transop) {
1039                 case LST_TRANS_SESNEW:
1040                 case LST_TRANS_SESEND:
1041                         rc = lstcon_sesrpc_prep(nd, transop, &rpc);
1042                         break;
1043                 case LST_TRANS_SESQRY:
1044                 case LST_TRANS_SESPING:
1045                         rc = lstcon_dbgrpc_prep(nd, &rpc);
1046                         break;
1047                 case LST_TRANS_TSBCLIADD:
1048                 case LST_TRANS_TSBSRVADD:
1049                         rc = lstcon_testrpc_prep(nd, transop,
1050                                                  (lstcon_test_t *)arg, &rpc);
1051                         break;
1052                 case LST_TRANS_TSBRUN:
1053                 case LST_TRANS_TSBSTOP:
1054                 case LST_TRANS_TSBCLIQRY:
1055                 case LST_TRANS_TSBSRVQRY:
1056                         rc = lstcon_batrpc_prep(nd, transop,
1057                                                 (lstcon_tsb_hdr_t *)arg, &rpc);
1058                         break;
1059                 case LST_TRANS_STATQRY:
1060                         rc = lstcon_statrpc_prep(nd, &rpc);
1061                         break;
1062                 default:
1063                         rc = -EINVAL;
1064                         break;
1065                 }
1066
1067                 if (rc != 0) {
1068                         CERROR("Failed to create RPC for transaction %s: %d\n",
1069                                lstcon_rpc_trans_name(transop), rc);
1070                         break;
1071                 }
1072
1073                 lstcon_rpc_trans_addreq(trans, rpc);
1074         }
1075
1076         if (rc == 0) {
1077                 *transpp = trans;
1078                 return 0;
1079         }
1080
1081         lstcon_rpc_trans_destroy(trans);
1082
1083         return rc;
1084 }
1085
1086 void
1087 lstcon_rpc_pinger(void *arg)
1088 {
1089         stt_timer_t        *ptimer = (stt_timer_t *)arg;
1090         lstcon_rpc_trans_t *trans;
1091         lstcon_rpc_t       *crpc;
1092         srpc_msg_t         *rep;
1093         srpc_debug_reqst_t *drq;
1094         lstcon_ndlink_t    *ndl;
1095         lstcon_node_t      *nd;
1096         time_t              intv;
1097         int                 count = 0;
1098         int                 rc;
1099
1100         /* RPC pinger is a special case of transaction,
1101          * it's called by timer at 8 seconds interval.
1102          */
1103         mutex_down(&console_session.ses_mutex);
1104
1105         if (console_session.ses_shutdown || console_session.ses_expired) {
1106                 mutex_up(&console_session.ses_mutex);
1107                 return;
1108         }
1109
1110         if (!console_session.ses_expired &&
1111             cfs_time_current_sec() - console_session.ses_laststamp >
1112             console_session.ses_timeout)
1113                 console_session.ses_expired = 1;
1114
1115         trans = console_session.ses_ping;
1116
1117         LASSERT (trans != NULL);
1118
1119         list_for_each_entry(ndl, &console_session.ses_ndl_list, ndl_link) {
1120                 nd = ndl->ndl_node;
1121
1122                 if (console_session.ses_expired) {
1123                         /* idle console, end session on all nodes */
1124                         if (nd->nd_state != LST_NODE_ACTIVE)
1125                                 continue;
1126
1127                         rc = lstcon_sesrpc_prep(nd, LST_TRANS_SESEND, &crpc);
1128                         if (rc != 0) {
1129                                 CERROR("Out of memory\n");
1130                                 break;
1131                         }
1132
1133                         lstcon_rpc_trans_addreq(trans, crpc);
1134                         lstcon_rpc_post(crpc);
1135
1136                         continue;
1137                 }
1138
1139                 crpc = &nd->nd_ping;
1140
1141                 if (crpc->crp_rpc != NULL) {
1142                         LASSERT (crpc->crp_trans == trans);
1143                         LASSERT (!list_empty(&crpc->crp_link));
1144
1145                         spin_lock(&crpc->crp_rpc->crpc_lock);
1146
1147                         LASSERT (crpc->crp_posted);
1148
1149                         if (!crpc->crp_finished) {
1150                                 /* in flight */
1151                                 spin_unlock(&crpc->crp_rpc->crpc_lock);
1152                                 continue;
1153                         }
1154
1155                         spin_unlock(&crpc->crp_rpc->crpc_lock);
1156
1157                         lstcon_rpc_get_reply(crpc, &rep);
1158
1159                         list_del_init(&crpc->crp_link);
1160
1161                         lstcon_rpc_put(crpc);
1162                 }
1163
1164                 if (nd->nd_state != LST_NODE_ACTIVE)
1165                         continue;
1166
1167                 intv = cfs_duration_sec(cfs_time_sub(cfs_time_current(),
1168                                                      nd->nd_stamp));
1169                 if (intv < nd->nd_timeout / 2)
1170                         continue;
1171
1172                 rc = lstcon_rpc_init(nd, SRPC_SERVICE_DEBUG, 0, 0, crpc);
1173                 if (rc != 0) {
1174                         CERROR("Out of memory\n");
1175                         break;
1176                 }
1177
1178                 drq = &crpc->crp_rpc->crpc_reqstmsg.msg_body.dbg_reqst;
1179
1180                 drq->dbg_sid   = console_session.ses_id;
1181                 drq->dbg_flags = 0;
1182
1183                 lstcon_rpc_trans_addreq(trans, crpc);
1184                 lstcon_rpc_post(crpc);
1185
1186                 count ++;
1187         }
1188
1189         if (console_session.ses_expired) {
1190                 mutex_up(&console_session.ses_mutex);
1191                 return;
1192         }
1193
1194         CDEBUG(D_NET, "Ping %d nodes in session\n", count);
1195
1196         ptimer->stt_expires = cfs_time_current_sec() + LST_PING_INTERVAL;
1197         stt_add_timer(ptimer);
1198
1199         mutex_up(&console_session.ses_mutex);
1200 }
1201
1202 int
1203 lstcon_rpc_pinger_start(void)
1204 {
1205         stt_timer_t    *ptimer;
1206         int             rc;
1207
1208         LASSERT (list_empty(&console_session.ses_rpc_freelist));
1209         LASSERT (atomic_read(&console_session.ses_rpc_counter) == 0);
1210
1211         rc = lstcon_rpc_trans_prep(NULL, LST_TRANS_SESPING,
1212                                    &console_session.ses_ping);
1213         if (rc != 0) {
1214                 CERROR("Failed to create console pinger\n");
1215                 return rc;
1216         }
1217
1218         ptimer = &console_session.ses_ping_timer;
1219         ptimer->stt_expires = cfs_time_current_sec() + LST_PING_INTERVAL;
1220
1221         stt_add_timer(ptimer);
1222
1223         return 0;
1224 }
1225
1226 void
1227 lstcon_rpc_pinger_stop(void)
1228 {
1229         LASSERT (console_session.ses_shutdown);
1230
1231         stt_del_timer(&console_session.ses_ping_timer);
1232
1233         lstcon_rpc_trans_abort(console_session.ses_ping, -ESHUTDOWN);
1234         lstcon_rpc_trans_stat(console_session.ses_ping, lstcon_trans_stat());
1235         lstcon_rpc_trans_destroy(console_session.ses_ping);
1236
1237         memset(lstcon_trans_stat(), 0, sizeof(lstcon_trans_stat_t));
1238
1239         console_session.ses_ping = NULL;
1240 }
1241
1242 void
1243 lstcon_rpc_cleanup_wait(void)
1244 {
1245         lstcon_rpc_trans_t *trans;
1246         lstcon_rpc_t       *crpc;
1247         struct list_head   *pacer;
1248         struct list_head    zlist;
1249
1250         /* Called with hold of global mutex */
1251
1252         LASSERT (console_session.ses_shutdown);
1253
1254         while (!list_empty(&console_session.ses_trans_list)) {
1255                 list_for_each(pacer, &console_session.ses_trans_list) {
1256                         trans = list_entry(pacer, lstcon_rpc_trans_t, tas_link);
1257
1258                         CDEBUG(D_NET, "Session closed, wakeup transaction %s\n",
1259                                lstcon_rpc_trans_name(trans->tas_opc));
1260
1261                         cfs_waitq_signal(&trans->tas_waitq);
1262                 }
1263
1264                 mutex_up(&console_session.ses_mutex);
1265
1266                 CWARN("Session is shutting down, "
1267                       "waiting for termination of transactions\n");
1268                 cfs_pause(cfs_time_seconds(1));
1269
1270                 mutex_down(&console_session.ses_mutex);
1271         }
1272
1273         spin_lock(&console_session.ses_rpc_lock);
1274
1275         lst_wait_until((atomic_read(&console_session.ses_rpc_counter) == 0),
1276                        console_session.ses_rpc_lock,
1277                        "Network is not accessable or target is down, "
1278                        "waiting for %d console RPCs to being recycled\n",
1279                        atomic_read(&console_session.ses_rpc_counter));
1280
1281         list_add(&zlist, &console_session.ses_rpc_freelist);
1282         list_del_init(&console_session.ses_rpc_freelist);
1283
1284         spin_unlock(&console_session.ses_rpc_lock);
1285
1286         while (!list_empty(&zlist)) {
1287                 crpc = list_entry(zlist.next, lstcon_rpc_t, crp_link);
1288
1289                 list_del(&crpc->crp_link);
1290                 LIBCFS_FREE(crpc, sizeof(lstcon_rpc_t));
1291         }
1292 }
1293
1294 int
1295 lstcon_rpc_module_init(void)
1296 {
1297         CFS_INIT_LIST_HEAD(&console_session.ses_ping_timer.stt_list);
1298         console_session.ses_ping_timer.stt_func = lstcon_rpc_pinger;
1299         console_session.ses_ping_timer.stt_data = &console_session.ses_ping_timer;
1300
1301         console_session.ses_ping = NULL;
1302
1303         spin_lock_init(&console_session.ses_rpc_lock);
1304         atomic_set(&console_session.ses_rpc_counter, 0);
1305         CFS_INIT_LIST_HEAD(&console_session.ses_rpc_freelist);
1306
1307         return 0;
1308 }
1309
1310 void
1311 lstcon_rpc_module_fini(void)
1312 {
1313         LASSERT (list_empty(&console_session.ses_rpc_freelist));
1314         LASSERT (atomic_read(&console_session.ses_rpc_counter) == 0);
1315 }
1316
1317 #endif