lnet/selftest/console.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, 2014, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lnet/selftest/conctl.c
  32  *
  33  * Infrastructure of LST console
  34  *
  35  * Author: Liang Zhen <liangzhen@clusterfs.com>
  36  */
  37
  38 #include <libcfs/libcfs.h>
  39 #include <lnet/lib-lnet.h>
  40 #include "console.h"
  41 #include "conrpc.h"
  42
  43 #define LST_NODE_STATE_COUNTER(nd, p)                   \
  44 do {                                                    \
  45         if ((nd)->nd_state == LST_NODE_ACTIVE)          \
  46                 (p)->nle_nactive ++;                    \
  47         else if ((nd)->nd_state == LST_NODE_BUSY)       \
  48                 (p)->nle_nbusy ++;                      \
  49         else if ((nd)->nd_state == LST_NODE_DOWN)       \
  50                 (p)->nle_ndown ++;                      \
  51         else                                            \
  52                 (p)->nle_nunknown ++;                   \
  53         (p)->nle_nnode ++;                              \
  54 } while (0)
  55
  56 struct lstcon_session console_session;
  57
  58 static void
  59 lstcon_node_get(struct lstcon_node *nd)
  60 {
  61         LASSERT (nd->nd_ref >= 1);
  62
  63         nd->nd_ref++;
  64 }
  65
  66 static int
  67 lstcon_node_find(struct lnet_process_id id, struct lstcon_node **ndpp,
  68                  int create)
  69 {
  70         struct lstcon_ndlink *ndl;
  71         unsigned int idx = LNET_NIDADDR(id.nid) % LST_GLOBAL_HASHSIZE;
  72
  73         LASSERT(id.nid != LNET_NID_ANY);
  74
  75         list_for_each_entry(ndl, &console_session.ses_ndl_hash[idx],
  76                             ndl_hlink) {
  77                 if (ndl->ndl_node->nd_id.nid != id.nid ||
  78                     ndl->ndl_node->nd_id.pid != id.pid)
  79                         continue;
  80
  81                 lstcon_node_get(ndl->ndl_node);
  82                 *ndpp = ndl->ndl_node;
  83                 return 0;
  84         }
  85
  86         if (!create)
  87                 return -ENOENT;
  88
  89         LIBCFS_ALLOC(*ndpp, sizeof(**ndpp) + sizeof(*ndl));
  90         if (*ndpp == NULL)
  91                 return -ENOMEM;
  92
  93         ndl = (struct lstcon_ndlink *)(*ndpp + 1);
  94
  95         ndl->ndl_node = *ndpp;
  96
  97         ndl->ndl_node->nd_ref   = 1;
  98         ndl->ndl_node->nd_id    = id;
  99         ndl->ndl_node->nd_stamp = ktime_get();
 100         ndl->ndl_node->nd_state = LST_NODE_UNKNOWN;
 101         ndl->ndl_node->nd_timeout = 0;
 102         memset(&ndl->ndl_node->nd_ping, 0, sizeof(ndl->ndl_node->nd_ping));
 103
 104         /* queued in global hash & list, no refcount is taken by
 105          * global hash & list, if caller release his refcount,
 106          * node will be released */
 107         list_add_tail(&ndl->ndl_hlink, &console_session.ses_ndl_hash[idx]);
 108         list_add_tail(&ndl->ndl_link, &console_session.ses_ndl_list);
 109
 110         return 0;
 111 }
 112
 113 static void
 114 lstcon_node_put(struct lstcon_node *nd)
 115 {
 116         struct lstcon_ndlink *ndl;
 117
 118         LASSERT(nd->nd_ref > 0);
 119
 120         if (--nd->nd_ref > 0)
 121                 return;
 122
 123         ndl = (struct lstcon_ndlink *)(nd + 1);
 124
 125         LASSERT(!list_empty(&ndl->ndl_link));
 126         LASSERT(!list_empty(&ndl->ndl_hlink));
 127
 128         /* remove from session */
 129         list_del(&ndl->ndl_link);
 130         list_del(&ndl->ndl_hlink);
 131
 132         LIBCFS_FREE(nd, sizeof(*nd) + sizeof(*ndl));
 133 }
 134
 135 static int
 136 lstcon_ndlink_find(struct list_head *hash, struct lnet_process_id id,
 137                    struct lstcon_ndlink **ndlpp, int create)
 138 {
 139         unsigned int idx = LNET_NIDADDR(id.nid) % LST_NODE_HASHSIZE;
 140         struct lstcon_ndlink *ndl;
 141         struct lstcon_node *nd;
 142         int rc;
 143
 144         if (id.nid == LNET_NID_ANY)
 145                 return -EINVAL;
 146
 147         /* search in hash */
 148         list_for_each_entry(ndl, &hash[idx], ndl_hlink) {
 149                 if (ndl->ndl_node->nd_id.nid != id.nid ||
 150                     ndl->ndl_node->nd_id.pid != id.pid)
 151                         continue;
 152
 153                 *ndlpp = ndl;
 154                 return 0;
 155         }
 156
 157         if (create == 0)
 158                 return -ENOENT;
 159
 160         /* find or create in session hash */
 161         rc = lstcon_node_find(id, &nd, (create == 1) ? 1 : 0);
 162         if (rc != 0)
 163                 return rc;
 164
 165         LIBCFS_ALLOC(ndl, sizeof(*ndl));
 166         if (ndl == NULL) {
 167                 lstcon_node_put(nd);
 168                 return -ENOMEM;
 169         }
 170
 171         *ndlpp = ndl;
 172
 173         ndl->ndl_node = nd;
 174         INIT_LIST_HEAD(&ndl->ndl_link);
 175         list_add_tail(&ndl->ndl_hlink, &hash[idx]);
 176
 177         return 0;
 178 }
 179
 180 static void
 181 lstcon_ndlink_release(struct lstcon_ndlink *ndl)
 182 {
 183         LASSERT(list_empty(&ndl->ndl_link));
 184         LASSERT(!list_empty(&ndl->ndl_hlink));
 185
 186         list_del(&ndl->ndl_hlink); /* delete from hash */
 187         lstcon_node_put(ndl->ndl_node);
 188
 189         LIBCFS_FREE(ndl, sizeof(*ndl));
 190 }
 191
 192 static int
 193 lstcon_group_alloc(char *name, struct lstcon_group **grpp)
 194 {
 195         struct lstcon_group *grp;
 196         int i;
 197
 198         LIBCFS_ALLOC(grp, offsetof(struct lstcon_group,
 199                                    grp_ndl_hash[LST_NODE_HASHSIZE]));
 200         if (grp == NULL)
 201                 return -ENOMEM;
 202
 203         grp->grp_ref = 1;
 204         if (name != NULL) {
 205                 if (strlen(name) > sizeof(grp->grp_name)-1) {
 206                         LIBCFS_FREE(grp, offsetof(struct lstcon_group,
 207                                           grp_ndl_hash[LST_NODE_HASHSIZE]));
 208                         return -E2BIG;
 209                 }
 210                 strncpy(grp->grp_name, name, sizeof(grp->grp_name));
 211         }
 212
 213         INIT_LIST_HEAD(&grp->grp_link);
 214         INIT_LIST_HEAD(&grp->grp_ndl_list);
 215         INIT_LIST_HEAD(&grp->grp_trans_list);
 216
 217         for (i = 0; i < LST_NODE_HASHSIZE; i++)
 218                 INIT_LIST_HEAD(&grp->grp_ndl_hash[i]);
 219
 220         *grpp = grp;
 221
 222         return 0;
 223 }
 224
 225 void lstcon_group_addref(struct lstcon_group *grp)
 226 {
 227         grp->grp_ref++;
 228 }
 229
 230 static void lstcon_group_ndlink_release(struct lstcon_group *,
 231                                         struct lstcon_ndlink *);
 232
 233 static void
 234 lstcon_group_drain(struct lstcon_group *grp, int keep)
 235 {
 236         struct lstcon_ndlink *ndl;
 237         struct lstcon_ndlink *tmp;
 238
 239         list_for_each_entry_safe(ndl, tmp, &grp->grp_ndl_list, ndl_link) {
 240                 if ((ndl->ndl_node->nd_state & keep) == 0)
 241                         lstcon_group_ndlink_release(grp, ndl);
 242         }
 243 }
 244
 245 void lstcon_group_decref(struct lstcon_group *grp)
 246 {
 247         int i;
 248
 249         if (--grp->grp_ref > 0)
 250                 return;
 251
 252         if (!list_empty(&grp->grp_link))
 253                 list_del(&grp->grp_link);
 254
 255         lstcon_group_drain(grp, 0);
 256
 257         for (i = 0; i < LST_NODE_HASHSIZE; i++)
 258                 LASSERT(list_empty(&grp->grp_ndl_hash[i]));
 259
 260         LIBCFS_FREE(grp, offsetof(struct lstcon_group,
 261                                   grp_ndl_hash[LST_NODE_HASHSIZE]));
 262 }
 263
 264 int lstcon_group_find(const char *name, struct lstcon_group **grpp)
 265 {
 266         struct lstcon_group *grp;
 267
 268         list_for_each_entry(grp, &console_session.ses_grp_list, grp_link) {
 269                 if (strncmp(grp->grp_name, name, LST_NAME_SIZE) != 0)
 270                         continue;
 271
 272                 lstcon_group_addref(grp);  /* +1 ref for caller */
 273                 *grpp = grp;
 274                 return 0;
 275         }
 276
 277         return -ENOENT;
 278 }
 279
 280 static int
 281 lstcon_group_ndlink_find(struct lstcon_group *grp, struct lnet_process_id id,
 282                          struct lstcon_ndlink **ndlpp, int create)
 283 {
 284         int rc;
 285
 286         rc = lstcon_ndlink_find(&grp->grp_ndl_hash[0], id, ndlpp, create);
 287         if (rc != 0)
 288                 return rc;
 289
 290         if (!list_empty(&(*ndlpp)->ndl_link))
 291                 return 0;
 292
 293         list_add_tail(&(*ndlpp)->ndl_link, &grp->grp_ndl_list);
 294         grp->grp_nnode++;
 295
 296         return 0;
 297 }
 298
 299 static void
 300 lstcon_group_ndlink_release(struct lstcon_group *grp, struct lstcon_ndlink *ndl)
 301 {
 302         list_del_init(&ndl->ndl_link);
 303         lstcon_ndlink_release(ndl);
 304         grp->grp_nnode--;
 305 }
 306
 307 static void
 308 lstcon_group_ndlink_move(struct lstcon_group *old,
 309                          struct lstcon_group *new, struct lstcon_ndlink *ndl)
 310 {
 311         unsigned int idx = LNET_NIDADDR(ndl->ndl_node->nd_id.nid) %
 312                                         LST_NODE_HASHSIZE;
 313
 314         old->grp_nnode--;
 315
 316         list_move_tail(&ndl->ndl_hlink, &new->grp_ndl_hash[idx]);
 317         list_move_tail(&ndl->ndl_link, &new->grp_ndl_list);
 318         new->grp_nnode++;
 319 }
 320
 321 static void
 322 lstcon_group_move(struct lstcon_group *old, struct lstcon_group *new)
 323 {
 324         struct lstcon_ndlink *ndl;
 325
 326         while (!list_empty(&old->grp_ndl_list)) {
 327                 ndl = list_entry(old->grp_ndl_list.next,
 328                                  struct lstcon_ndlink, ndl_link);
 329                 lstcon_group_ndlink_move(old, new, ndl);
 330         }
 331 }
 332
 333 static int
 334 lstcon_sesrpc_condition(int transop, struct lstcon_node *nd, void *arg)
 335 {
 336         struct lstcon_group *grp = arg;
 337
 338         switch (transop) {
 339         case LST_TRANS_SESNEW:
 340                 if (nd->nd_state == LST_NODE_ACTIVE)
 341                         return 0;
 342                 break;
 343
 344         case LST_TRANS_SESEND:
 345                 if (nd->nd_state != LST_NODE_ACTIVE)
 346                         return 0;
 347
 348                 if (grp != NULL && nd->nd_ref > 1)
 349                         return 0;
 350                 break;
 351
 352         case LST_TRANS_SESQRY:
 353                 break;
 354
 355         default:
 356                 LBUG();
 357         }
 358
 359         return 1;
 360 }
 361
 362 static int
 363 lstcon_sesrpc_readent(int transop, struct srpc_msg *msg,
 364                       struct lstcon_rpc_ent __user *ent_up)
 365 {
 366         struct srpc_debug_reply *rep;
 367
 368         switch (transop) {
 369         case LST_TRANS_SESNEW:
 370         case LST_TRANS_SESEND:
 371                 return 0;
 372
 373         case LST_TRANS_SESQRY:
 374                 rep = &msg->msg_body.dbg_reply;
 375
 376                 if (copy_to_user(&ent_up->rpe_priv[0],
 377                                      &rep->dbg_timeout, sizeof(int)) ||
 378                     copy_to_user(&ent_up->rpe_payload[0],
 379                                      &rep->dbg_name, LST_NAME_SIZE))
 380                         return -EFAULT;
 381
 382                 return 0;
 383
 384         default:
 385                 LBUG();
 386         }
 387
 388         return 0;
 389 }
 390
 391 static int
 392 lstcon_group_nodes_add(struct lstcon_group *grp,
 393                        int count, struct lnet_process_id __user *ids_up,
 394                        unsigned int *featp,
 395                        struct list_head __user *result_up)
 396 {
 397         struct lstcon_rpc_trans *trans;
 398         struct lstcon_ndlink *ndl;
 399         struct lstcon_group *tmp;
 400         struct lnet_process_id id;
 401         int i;
 402         int rc;
 403
 404         rc = lstcon_group_alloc(NULL, &tmp);
 405         if (rc != 0) {
 406                 CERROR("Out of memory\n");
 407                 return -ENOMEM;
 408         }
 409
 410         for (i = 0 ; i < count; i++) {
 411                 if (copy_from_user(&id, &ids_up[i], sizeof(id))) {
 412                         rc = -EFAULT;
 413                         break;
 414                 }
 415
 416                 /* skip if it's in this group already */
 417                 rc = lstcon_group_ndlink_find(grp, id, &ndl, 0);
 418                 if (rc == 0)
 419                         continue;
 420
 421                 /* add to tmp group */
 422                 rc = lstcon_group_ndlink_find(tmp, id, &ndl, 1);
 423                 if (rc != 0) {
 424                         CERROR("Can't create ndlink, out of memory\n");
 425                         break;
 426                 }
 427         }
 428
 429         if (rc != 0) {
 430                 lstcon_group_decref(tmp);
 431                 return rc;
 432         }
 433
 434         rc = lstcon_rpc_trans_ndlist(&tmp->grp_ndl_list,
 435                                      &tmp->grp_trans_list, LST_TRANS_SESNEW,
 436                                      tmp, lstcon_sesrpc_condition, &trans);
 437         if (rc != 0) {
 438                 CERROR("Can't create transaction: %d\n", rc);
 439                 lstcon_group_decref(tmp);
 440                 return rc;
 441         }
 442
 443         /* post all RPCs */
 444         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
 445
 446         rc = lstcon_rpc_trans_interpreter(trans, result_up,
 447                                           lstcon_sesrpc_readent);
 448         *featp = trans->tas_features;
 449
 450         /* destroy all RPGs */
 451         lstcon_rpc_trans_destroy(trans);
 452
 453         lstcon_group_move(tmp, grp);
 454         lstcon_group_decref(tmp);
 455
 456         return rc;
 457 }
 458
 459 static int
 460 lstcon_group_nodes_remove(struct lstcon_group *grp,
 461                           int count, struct lnet_process_id __user *ids_up,
 462                           struct list_head __user *result_up)
 463 {
 464         struct lstcon_rpc_trans *trans;
 465         struct lstcon_ndlink *ndl;
 466         struct lstcon_group *tmp;
 467         struct lnet_process_id id;
 468         int rc;
 469         int i;
 470
 471         /* End session and remove node from the group */
 472
 473         rc = lstcon_group_alloc(NULL, &tmp);
 474         if (rc != 0) {
 475                 CERROR("Out of memory\n");
 476                 return -ENOMEM;
 477         }
 478
 479         for (i = 0; i < count; i++) {
 480                 if (copy_from_user(&id, &ids_up[i], sizeof(id))) {
 481                         rc = -EFAULT;
 482                         goto error;
 483                 }
 484
 485                 /* move node to tmp group */
 486                 if (lstcon_group_ndlink_find(grp, id, &ndl, 0) == 0)
 487                         lstcon_group_ndlink_move(grp, tmp, ndl);
 488         }
 489
 490         rc = lstcon_rpc_trans_ndlist(&tmp->grp_ndl_list,
 491                                      &tmp->grp_trans_list, LST_TRANS_SESEND,
 492                                      tmp, lstcon_sesrpc_condition, &trans);
 493         if (rc != 0) {
 494                 CERROR("Can't create transaction: %d\n", rc);
 495                 goto error;
 496         }
 497
 498         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
 499
 500         rc = lstcon_rpc_trans_interpreter(trans, result_up, NULL);
 501
 502         lstcon_rpc_trans_destroy(trans);
 503         /* release nodes anyway, because we can't rollback status */
 504         lstcon_group_decref(tmp);
 505
 506         return rc;
 507 error:
 508         lstcon_group_move(tmp, grp);
 509         lstcon_group_decref(tmp);
 510
 511         return rc;
 512 }
 513
 514 int
 515 lstcon_group_add(char *name)
 516 {
 517         struct lstcon_group *grp;
 518         int rc;
 519
 520         rc = (lstcon_group_find(name, &grp) == 0)? -EEXIST: 0;
 521         if (rc != 0) {
 522                 /* find a group with same name */
 523                 lstcon_group_decref(grp);
 524                 return rc;
 525         }
 526
 527         rc = lstcon_group_alloc(name, &grp);
 528         if (rc != 0) {
 529                 CERROR("Can't allocate descriptor for group %s\n", name);
 530                 return -ENOMEM;
 531         }
 532
 533         list_add_tail(&grp->grp_link, &console_session.ses_grp_list);
 534
 535         return rc;
 536 }
 537
 538 int
 539 lstcon_nodes_add(char *name, int count, struct lnet_process_id __user *ids_up,
 540                  unsigned *featp, struct list_head __user *result_up)
 541 {
 542         struct lstcon_group         *grp;
 543         int                     rc;
 544
 545         LASSERT (count > 0);
 546         LASSERT (ids_up != NULL);
 547
 548         rc = lstcon_group_find(name, &grp);
 549         if (rc != 0) {
 550                 CDEBUG(D_NET, "Can't find group %s\n", name);
 551                 return rc;
 552         }
 553
 554         if (grp->grp_ref > 2) {
 555                 /* referred by other threads or test */
 556                 CDEBUG(D_NET, "Group %s is busy\n", name);
 557                 lstcon_group_decref(grp);
 558
 559                 return -EBUSY;
 560         }
 561
 562         rc = lstcon_group_nodes_add(grp, count, ids_up, featp, result_up);
 563
 564         lstcon_group_decref(grp);
 565
 566         return rc;
 567 }
 568
 569 int
 570 lstcon_group_del(char *name)
 571 {
 572         struct lstcon_rpc_trans *trans;
 573         struct lstcon_group *grp;
 574         int rc;
 575
 576         rc = lstcon_group_find(name, &grp);
 577         if (rc != 0) {
 578                 CDEBUG(D_NET, "Can't find group: %s\n", name);
 579                 return rc;
 580         }
 581
 582         if (grp->grp_ref > 2) {
 583                 /* referred by others threads or test */
 584                 CDEBUG(D_NET, "Group %s is busy\n", name);
 585                 lstcon_group_decref(grp);
 586                 return -EBUSY;
 587         }
 588
 589         rc = lstcon_rpc_trans_ndlist(&grp->grp_ndl_list,
 590                                      &grp->grp_trans_list, LST_TRANS_SESEND,
 591                                      grp, lstcon_sesrpc_condition, &trans);
 592         if (rc != 0) {
 593                 CERROR("Can't create transaction: %d\n", rc);
 594                 lstcon_group_decref(grp);
 595                 return rc;
 596         }
 597
 598         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
 599
 600         lstcon_rpc_trans_destroy(trans);
 601
 602         lstcon_group_decref(grp);
 603         /* -ref for session, it's destroyed,
 604          * status can't be rolled back, destroy group anway */
 605         lstcon_group_decref(grp);
 606
 607         return rc;
 608 }
 609
 610 int
 611 lstcon_group_clean(char *name, int args)
 612 {
 613         struct lstcon_group *grp = NULL;
 614         int rc;
 615
 616         rc = lstcon_group_find(name, &grp);
 617         if (rc != 0) {
 618                 CDEBUG(D_NET, "Can't find group %s\n", name);
 619                 return rc;
 620         }
 621
 622         if (grp->grp_ref > 2) {
 623                 /* referred by test */
 624                 CDEBUG(D_NET, "Group %s is busy\n", name);
 625                 lstcon_group_decref(grp);
 626                 return -EBUSY;
 627         }
 628
 629         args = (LST_NODE_ACTIVE | LST_NODE_BUSY |
 630                 LST_NODE_DOWN | LST_NODE_UNKNOWN) & ~args;
 631
 632         lstcon_group_drain(grp, args);
 633
 634         lstcon_group_decref(grp);
 635         /* release empty group */
 636         if (list_empty(&grp->grp_ndl_list))
 637                 lstcon_group_decref(grp);
 638
 639         return 0;
 640 }
 641
 642 int
 643 lstcon_nodes_remove(char *name, int count,
 644                     struct lnet_process_id __user *ids_up,
 645                     struct list_head __user *result_up)
 646 {
 647         struct lstcon_group *grp = NULL;
 648         int rc;
 649
 650         rc = lstcon_group_find(name, &grp);
 651         if (rc != 0) {
 652                 CDEBUG(D_NET, "Can't find group: %s\n", name);
 653                 return rc;
 654         }
 655
 656         if (grp->grp_ref > 2) {
 657                 /* referred by test */
 658                 CDEBUG(D_NET, "Group %s is busy\n", name);
 659                 lstcon_group_decref(grp);
 660                 return -EBUSY;
 661         }
 662
 663         rc = lstcon_group_nodes_remove(grp, count, ids_up, result_up);
 664
 665         lstcon_group_decref(grp);
 666         /* release empty group */
 667         if (list_empty(&grp->grp_ndl_list))
 668                 lstcon_group_decref(grp);
 669
 670         return rc;
 671 }
 672
 673 int
 674 lstcon_group_refresh(char *name, struct list_head __user *result_up)
 675 {
 676         struct lstcon_rpc_trans *trans;
 677         struct lstcon_group *grp;
 678         int rc;
 679
 680         rc = lstcon_group_find(name, &grp);
 681         if (rc != 0) {
 682                 CDEBUG(D_NET, "Can't find group: %s\n", name);
 683                 return rc;
 684         }
 685
 686         if (grp->grp_ref > 2) {
 687                 /* referred by test */
 688                 CDEBUG(D_NET, "Group %s is busy\n", name);
 689                 lstcon_group_decref(grp);
 690                 return -EBUSY;
 691         }
 692
 693         /* re-invite all inactive nodes int the group */
 694         rc = lstcon_rpc_trans_ndlist(&grp->grp_ndl_list,
 695                                      &grp->grp_trans_list, LST_TRANS_SESNEW,
 696                                      grp, lstcon_sesrpc_condition, &trans);
 697         if (rc != 0) {
 698                 /* local error, return */
 699                 CDEBUG(D_NET, "Can't create transaction: %d\n", rc);
 700                 lstcon_group_decref(grp);
 701                 return rc;
 702         }
 703
 704         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
 705
 706         rc = lstcon_rpc_trans_interpreter(trans, result_up, NULL);
 707
 708         lstcon_rpc_trans_destroy(trans);
 709         /* -ref for me */
 710         lstcon_group_decref(grp);
 711
 712         return rc;
 713 }
 714
 715 static int
 716 lstcon_nodes_getent(struct list_head *head, int *index_p,
 717                     int *count_p, struct lstcon_node_ent __user *dents_up)
 718 {
 719         struct lstcon_ndlink *ndl;
 720         struct lstcon_node *nd;
 721         int count = 0;
 722         int index = 0;
 723
 724         LASSERT(index_p != NULL && count_p != NULL);
 725         LASSERT(dents_up != NULL);
 726         LASSERT(*index_p >= 0);
 727         LASSERT(*count_p > 0);
 728
 729         list_for_each_entry(ndl, head, ndl_link) {
 730                 if (index++ < *index_p)
 731                         continue;
 732
 733                 if (count >= *count_p)
 734                         break;
 735
 736                 nd = ndl->ndl_node;
 737                 if (copy_to_user(&dents_up[count].nde_id,
 738                                      &nd->nd_id, sizeof(nd->nd_id)) ||
 739                     copy_to_user(&dents_up[count].nde_state,
 740                                      &nd->nd_state, sizeof(nd->nd_state)))
 741                         return -EFAULT;
 742
 743                 count ++;
 744         }
 745
 746         if (index <= *index_p)
 747                 return -ENOENT;
 748
 749         *count_p = count;
 750         *index_p = index;
 751
 752         return 0;
 753 }
 754
 755 int
 756 lstcon_group_info(char *name, struct lstcon_ndlist_ent __user *gents_p,
 757                   int *index_p, int *count_p,
 758                   struct lstcon_node_ent __user *dents_up)
 759 {
 760         struct lstcon_ndlist_ent *gentp;
 761         struct lstcon_group *grp;
 762         struct lstcon_ndlink *ndl;
 763         int rc;
 764
 765         rc = lstcon_group_find(name, &grp);
 766         if (rc != 0) {
 767                 CDEBUG(D_NET, "Can't find group %s\n", name);
 768                 return rc;
 769         }
 770
 771         if (dents_up != NULL) {
 772                 /* verbose query */
 773                 rc = lstcon_nodes_getent(&grp->grp_ndl_list,
 774                                          index_p, count_p, dents_up);
 775                 lstcon_group_decref(grp);
 776
 777                 return rc;
 778         }
 779
 780         /* non-verbose query */
 781         CFS_ALLOC_PTR(gentp);
 782         if (gentp == NULL) {
 783                 CERROR("Can't allocate ndlist_ent\n");
 784                 lstcon_group_decref(grp);
 785
 786                 return -ENOMEM;
 787         }
 788
 789         list_for_each_entry(ndl, &grp->grp_ndl_list, ndl_link)
 790                 LST_NODE_STATE_COUNTER(ndl->ndl_node, gentp);
 791
 792         rc = copy_to_user(gents_p, gentp,
 793                           sizeof(struct lstcon_ndlist_ent)) ? -EFAULT : 0;
 794
 795         CFS_FREE_PTR(gentp);
 796
 797         lstcon_group_decref(grp);
 798
 799         return 0;
 800 }
 801
 802 static int
 803 lstcon_batch_find(const char *name, struct lstcon_batch **batpp)
 804 {
 805         struct lstcon_batch *bat;
 806
 807         list_for_each_entry(bat, &console_session.ses_bat_list, bat_link) {
 808                 if (strncmp(bat->bat_name, name, LST_NAME_SIZE) == 0) {
 809                         *batpp = bat;
 810                         return 0;
 811                 }
 812         }
 813
 814         return -ENOENT;
 815 }
 816
 817 int
 818 lstcon_batch_add(char *name)
 819 {
 820         struct lstcon_batch *bat;
 821         int i;
 822         int rc;
 823
 824         rc = (lstcon_batch_find(name, &bat) == 0)? -EEXIST: 0;
 825         if (rc != 0) {
 826                 CDEBUG(D_NET, "Batch %s already exists\n", name);
 827                 return rc;
 828         }
 829
 830         LIBCFS_ALLOC(bat, sizeof(*bat));
 831         if (bat == NULL) {
 832                 CERROR("Can't allocate descriptor for batch %s\n", name);
 833                 return -ENOMEM;
 834         }
 835
 836         CFS_ALLOC_PTR_ARRAY(bat->bat_cli_hash, LST_NODE_HASHSIZE);
 837         if (bat->bat_cli_hash == NULL) {
 838                 CERROR("Can't allocate hash for batch %s\n", name);
 839                 LIBCFS_FREE(bat, sizeof(*bat));
 840
 841                 return -ENOMEM;
 842         }
 843
 844         CFS_ALLOC_PTR_ARRAY(bat->bat_srv_hash, LST_NODE_HASHSIZE);
 845         if (bat->bat_srv_hash == NULL) {
 846                 CERROR("Can't allocate hash for batch %s\n", name);
 847                 LIBCFS_FREE(bat->bat_cli_hash, LST_NODE_HASHSIZE);
 848                 LIBCFS_FREE(bat, sizeof(*bat));
 849
 850                 return -ENOMEM;
 851         }
 852
 853         if (strlen(name) > sizeof(bat->bat_name)-1) {
 854                 LIBCFS_FREE(bat->bat_srv_hash, LST_NODE_HASHSIZE);
 855                 LIBCFS_FREE(bat->bat_cli_hash, LST_NODE_HASHSIZE);
 856                 LIBCFS_FREE(bat, sizeof(*bat));
 857                 return -E2BIG;
 858         }
 859         strncpy(bat->bat_name, name, sizeof(bat->bat_name));
 860         bat->bat_hdr.tsb_index = 0;
 861         bat->bat_hdr.tsb_id.bat_id = ++console_session.ses_id_cookie;
 862
 863         bat->bat_ntest = 0;
 864         bat->bat_state = LST_BATCH_IDLE;
 865
 866         INIT_LIST_HEAD(&bat->bat_cli_list);
 867         INIT_LIST_HEAD(&bat->bat_srv_list);
 868         INIT_LIST_HEAD(&bat->bat_test_list);
 869         INIT_LIST_HEAD(&bat->bat_trans_list);
 870
 871         for (i = 0; i < LST_NODE_HASHSIZE; i++) {
 872                 INIT_LIST_HEAD(&bat->bat_cli_hash[i]);
 873                 INIT_LIST_HEAD(&bat->bat_srv_hash[i]);
 874         }
 875
 876         list_add_tail(&bat->bat_link, &console_session.ses_bat_list);
 877
 878         return rc;
 879 }
 880
 881 int
 882 lstcon_batch_list(int index, int len, char __user *name_up)
 883 {
 884         struct lstcon_batch *bat;
 885
 886         LASSERT(name_up != NULL);
 887         LASSERT(index >= 0);
 888
 889         list_for_each_entry(bat, &console_session.ses_bat_list, bat_link) {
 890                 if (index-- == 0) {
 891                         return copy_to_user(name_up, bat->bat_name, len) ?
 892                                             -EFAULT : 0;
 893                 }
 894         }
 895
 896         return -ENOENT;
 897 }
 898
 899 int
 900 lstcon_batch_info(char *name, struct lstcon_test_batch_ent __user *ent_up,
 901                   int server, int testidx, int *index_p, int *ndent_p,
 902                   struct lstcon_node_ent __user *dents_up)
 903 {
 904         struct lstcon_test_batch_ent *entp;
 905         struct list_head *clilst;
 906         struct list_head *srvlst;
 907         struct lstcon_test *test = NULL;
 908         struct lstcon_batch *bat;
 909         struct lstcon_ndlink *ndl;
 910         int rc;
 911
 912         rc = lstcon_batch_find(name, &bat);
 913         if (rc != 0) {
 914                 CDEBUG(D_NET, "Can't find batch %s\n", name);
 915                 return -ENOENT;
 916         }
 917
 918         if (testidx > 0) {
 919                 /* query test, test index start from 1 */
 920                 list_for_each_entry(test, &bat->bat_test_list, tes_link) {
 921                         if (testidx-- == 1)
 922                                 break;
 923                 }
 924
 925                 if (testidx > 0) {
 926                         CDEBUG(D_NET, "Can't find specified test in batch\n");
 927                         return -ENOENT;
 928                 }
 929         }
 930
 931         clilst = (test == NULL) ? &bat->bat_cli_list :
 932                                   &test->tes_src_grp->grp_ndl_list;
 933         srvlst = (test == NULL) ? &bat->bat_srv_list :
 934                                   &test->tes_dst_grp->grp_ndl_list;
 935
 936         if (dents_up != NULL) {
 937                 rc = lstcon_nodes_getent((server ? srvlst: clilst),
 938                                          index_p, ndent_p, dents_up);
 939                 return rc;
 940         }
 941
 942         /* non-verbose query */
 943         CFS_ALLOC_PTR(entp);
 944         if (entp == NULL)
 945                 return -ENOMEM;
 946
 947         if (test == NULL) {
 948                 entp->u.tbe_batch.bae_ntest = bat->bat_ntest;
 949                 entp->u.tbe_batch.bae_state = bat->bat_state;
 950
 951         } else {
 952
 953                 entp->u.tbe_test.tse_type   = test->tes_type;
 954                 entp->u.tbe_test.tse_loop   = test->tes_loop;
 955                 entp->u.tbe_test.tse_concur = test->tes_concur;
 956         }
 957
 958         list_for_each_entry(ndl, clilst, ndl_link)
 959                 LST_NODE_STATE_COUNTER(ndl->ndl_node, &entp->tbe_cli_nle);
 960
 961         list_for_each_entry(ndl, srvlst, ndl_link)
 962                 LST_NODE_STATE_COUNTER(ndl->ndl_node, &entp->tbe_srv_nle);
 963
 964         rc = copy_to_user(ent_up, entp,
 965                           sizeof(struct lstcon_test_batch_ent)) ? -EFAULT : 0;
 966
 967         CFS_FREE_PTR(entp)
 968
 969         return rc;
 970 }
 971
 972 static int
 973 lstcon_batrpc_condition(int transop, struct lstcon_node *nd, void *arg)
 974 {
 975         switch (transop) {
 976         case LST_TRANS_TSBRUN:
 977                 if (nd->nd_state != LST_NODE_ACTIVE)
 978                         return -ENETDOWN;
 979                 break;
 980
 981         case LST_TRANS_TSBSTOP:
 982                 if (nd->nd_state != LST_NODE_ACTIVE)
 983                         return 0;
 984                 break;
 985
 986         case LST_TRANS_TSBCLIQRY:
 987         case LST_TRANS_TSBSRVQRY:
 988                 break;
 989         }
 990
 991         return 1;
 992 }
 993
 994 static int
 995 lstcon_batch_op(struct lstcon_batch *bat, int transop,
 996                 struct list_head __user *result_up)
 997 {
 998         struct lstcon_rpc_trans *trans;
 999         int                 rc;
1000
1001         rc = lstcon_rpc_trans_ndlist(&bat->bat_cli_list,
1002                                      &bat->bat_trans_list, transop,
1003                                      bat, lstcon_batrpc_condition, &trans);
1004         if (rc != 0) {
1005                 CERROR("Can't create transaction: %d\n", rc);
1006                 return rc;
1007         }
1008
1009         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
1010
1011         rc = lstcon_rpc_trans_interpreter(trans, result_up, NULL);
1012
1013         lstcon_rpc_trans_destroy(trans);
1014
1015         return rc;
1016 }
1017
1018 int
1019 lstcon_batch_run(char *name, int timeout, struct list_head __user *result_up)
1020 {
1021         struct lstcon_batch *bat;
1022         int rc;
1023
1024         if (lstcon_batch_find(name, &bat) != 0) {
1025                 CDEBUG(D_NET, "Can't find batch %s\n", name);
1026                 return -ENOENT;
1027         }
1028
1029         bat->bat_arg = timeout;
1030
1031         rc = lstcon_batch_op(bat, LST_TRANS_TSBRUN, result_up);
1032
1033         /* mark batch as running if it's started in any node */
1034         if (lstcon_tsbop_stat_success(lstcon_trans_stat(), 0) != 0)
1035                 bat->bat_state = LST_BATCH_RUNNING;
1036
1037         return rc;
1038 }
1039
1040 int
1041 lstcon_batch_stop(char *name, int force, struct list_head __user *result_up)
1042 {
1043         struct lstcon_batch *bat;
1044         int rc;
1045
1046         if (lstcon_batch_find(name, &bat) != 0) {
1047                 CDEBUG(D_NET, "Can't find batch %s\n", name);
1048                 return -ENOENT;
1049         }
1050
1051         bat->bat_arg = force;
1052
1053         rc = lstcon_batch_op(bat, LST_TRANS_TSBSTOP, result_up);
1054
1055         /* mark batch as stopped if all RPCs finished */
1056         if (lstcon_tsbop_stat_failure(lstcon_trans_stat(), 0) == 0)
1057                 bat->bat_state = LST_BATCH_IDLE;
1058
1059         return rc;
1060 }
1061
1062 static void
1063 lstcon_batch_destroy(struct lstcon_batch *bat)
1064 {
1065         struct lstcon_ndlink *ndl;
1066         struct lstcon_test *test;
1067         int i;
1068
1069         list_del(&bat->bat_link);
1070
1071         while (!list_empty(&bat->bat_test_list)) {
1072                 test = list_entry(bat->bat_test_list.next,
1073                                   struct lstcon_test, tes_link);
1074                 LASSERT(list_empty(&test->tes_trans_list));
1075
1076                 list_del(&test->tes_link);
1077
1078                 lstcon_group_decref(test->tes_src_grp);
1079                 lstcon_group_decref(test->tes_dst_grp);
1080
1081                 LIBCFS_FREE(test, offsetof(struct lstcon_test,
1082                                            tes_param[test->tes_paramlen]));
1083         }
1084
1085         LASSERT(list_empty(&bat->bat_trans_list));
1086
1087         while (!list_empty(&bat->bat_cli_list)) {
1088                 ndl = list_entry(bat->bat_cli_list.next,
1089                                  struct lstcon_ndlink, ndl_link);
1090                 list_del_init(&ndl->ndl_link);
1091
1092                 lstcon_ndlink_release(ndl);
1093         }
1094
1095         while (!list_empty(&bat->bat_srv_list)) {
1096                 ndl = list_entry(bat->bat_srv_list.next,
1097                                  struct lstcon_ndlink, ndl_link);
1098                 list_del_init(&ndl->ndl_link);
1099
1100                 lstcon_ndlink_release(ndl);
1101         }
1102
1103         for (i = 0; i < LST_NODE_HASHSIZE; i++) {
1104                 LASSERT(list_empty(&bat->bat_cli_hash[i]));
1105                 LASSERT(list_empty(&bat->bat_srv_hash[i]));
1106         }
1107
1108         LIBCFS_FREE(bat->bat_cli_hash,
1109                     sizeof(struct list_head) * LST_NODE_HASHSIZE);
1110         LIBCFS_FREE(bat->bat_srv_hash,
1111                     sizeof(struct list_head) * LST_NODE_HASHSIZE);
1112         LIBCFS_FREE(bat, sizeof(*bat));
1113 }
1114
1115 static int
1116 lstcon_testrpc_condition(int transop, struct lstcon_node *nd, void *arg)
1117 {
1118         struct lstcon_test *test = arg;
1119         struct lstcon_batch *batch;
1120         struct lstcon_ndlink *ndl;
1121         struct list_head *hash;
1122         struct list_head *head;
1123
1124         LASSERT(test != NULL);
1125
1126         batch = test->tes_batch;
1127         LASSERT(batch != NULL);
1128
1129         if (test->tes_oneside &&
1130             transop == LST_TRANS_TSBSRVADD)
1131                 return 0;
1132
1133         if (nd->nd_state != LST_NODE_ACTIVE)
1134                 return -ENETDOWN;
1135
1136         if (transop == LST_TRANS_TSBCLIADD) {
1137                 hash = batch->bat_cli_hash;
1138                 head = &batch->bat_cli_list;
1139
1140         } else {
1141                 LASSERT (transop == LST_TRANS_TSBSRVADD);
1142
1143                 hash = batch->bat_srv_hash;
1144                 head = &batch->bat_srv_list;
1145         }
1146
1147         LASSERT (nd->nd_id.nid != LNET_NID_ANY);
1148
1149         if (lstcon_ndlink_find(hash, nd->nd_id, &ndl, 1) != 0)
1150                 return -ENOMEM;
1151
1152         if (list_empty(&ndl->ndl_link))
1153                 list_add_tail(&ndl->ndl_link, head);
1154
1155         return 1;
1156 }
1157
1158 static int
1159 lstcon_test_nodes_add(struct lstcon_test *test,
1160                       struct list_head __user *result_up)
1161 {
1162         struct lstcon_rpc_trans *trans;
1163         struct lstcon_group *grp;
1164         int transop;
1165         int rc;
1166
1167         LASSERT (test->tes_src_grp != NULL);
1168         LASSERT (test->tes_dst_grp != NULL);
1169
1170         transop = LST_TRANS_TSBSRVADD;
1171         grp  = test->tes_dst_grp;
1172 again:
1173         rc = lstcon_rpc_trans_ndlist(&grp->grp_ndl_list,
1174                                      &test->tes_trans_list, transop,
1175                                      test, lstcon_testrpc_condition, &trans);
1176         if (rc != 0) {
1177                 CERROR("Can't create transaction: %d\n", rc);
1178                 return rc;
1179         }
1180
1181         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
1182
1183         if (lstcon_trans_stat()->trs_rpc_errno != 0 ||
1184             lstcon_trans_stat()->trs_fwk_errno != 0) {
1185                 lstcon_rpc_trans_interpreter(trans, result_up, NULL);
1186
1187                 lstcon_rpc_trans_destroy(trans);
1188                 /* return if any error */
1189                 CDEBUG(D_NET, "Failed to add test %s, "
1190                               "RPC error %d, framework error %d\n",
1191                        transop == LST_TRANS_TSBCLIADD ? "client" : "server",
1192                        lstcon_trans_stat()->trs_rpc_errno,
1193                        lstcon_trans_stat()->trs_fwk_errno);
1194
1195                 return rc;
1196         }
1197
1198         lstcon_rpc_trans_destroy(trans);
1199
1200         if (transop == LST_TRANS_TSBCLIADD)
1201                 return rc;
1202
1203         transop = LST_TRANS_TSBCLIADD;
1204         grp = test->tes_src_grp;
1205         test->tes_cliidx = 0;
1206
1207         /* requests to test clients */
1208         goto again;
1209 }
1210
1211 static int
1212 lstcon_verify_batch(const char *name, struct lstcon_batch **batch)
1213 {
1214         int rc;
1215
1216         rc = lstcon_batch_find(name, batch);
1217         if (rc != 0) {
1218                 CDEBUG(D_NET, "Can't find batch %s\n", name);
1219                 return rc;
1220         }
1221
1222         if ((*batch)->bat_state != LST_BATCH_IDLE) {
1223                 CDEBUG(D_NET, "Can't change running batch %s\n", name);
1224                 return -EINVAL;
1225         }
1226
1227         return 0;
1228 }
1229
1230 static int
1231 lstcon_verify_group(const char *name, struct lstcon_group **grp)
1232 {
1233         int rc;
1234         struct lstcon_ndlink *ndl;
1235
1236         rc = lstcon_group_find(name, grp);
1237         if (rc != 0) {
1238                 CDEBUG(D_NET, "can't find group %s\n", name);
1239                 return rc;
1240         }
1241
1242         list_for_each_entry(ndl, &(*grp)->grp_ndl_list, ndl_link) {
1243                 if (ndl->ndl_node->nd_state == LST_NODE_ACTIVE) {
1244                         return 0;
1245                 }
1246         }
1247
1248         CDEBUG(D_NET, "Group %s has no ACTIVE nodes\n", name);
1249
1250         return -EINVAL;
1251 }
1252
1253 int
1254 lstcon_test_add(char *batch_name, int type, int loop,
1255                 int concur, int dist, int span,
1256                 char *src_name, char *dst_name,
1257                 void *param, int paramlen, int *retp,
1258                 struct list_head __user *result_up)
1259 {
1260         struct lstcon_test *test = NULL;
1261         int rc;
1262         struct lstcon_group *src_grp = NULL;
1263         struct lstcon_group *dst_grp = NULL;
1264         struct lstcon_batch *batch = NULL;
1265
1266         /*
1267          * verify that a batch of the given name exists, and the groups
1268          * that will be part of the batch exist and have at least one
1269          * active node
1270          */
1271         rc = lstcon_verify_batch(batch_name, &batch);
1272         if (rc != 0)
1273                 goto out;
1274
1275         rc = lstcon_verify_group(src_name, &src_grp);
1276         if (rc != 0)
1277                 goto out;
1278
1279         rc = lstcon_verify_group(dst_name, &dst_grp);
1280         if (rc != 0)
1281                 goto out;
1282
1283         if (dst_grp->grp_userland)
1284                 *retp = 1;
1285
1286         LIBCFS_ALLOC(test, offsetof(struct lstcon_test, tes_param[paramlen]));
1287         if (!test) {
1288                 CERROR("Can't allocate test descriptor\n");
1289                 rc = -ENOMEM;
1290
1291                 goto out;
1292         }
1293
1294         test->tes_hdr.tsb_id    = batch->bat_hdr.tsb_id;
1295         test->tes_batch         = batch;
1296         test->tes_type          = type;
1297         test->tes_oneside       = 0; /* TODO */
1298         test->tes_loop          = loop;
1299         test->tes_concur        = concur;
1300         test->tes_stop_onerr    = 1; /* TODO */
1301         test->tes_span          = span;
1302         test->tes_dist          = dist;
1303         test->tes_cliidx        = 0; /* just used for creating RPC */
1304         test->tes_src_grp       = src_grp;
1305         test->tes_dst_grp       = dst_grp;
1306         INIT_LIST_HEAD(&test->tes_trans_list);
1307
1308         if (param != NULL) {
1309                 test->tes_paramlen = paramlen;
1310                 memcpy(&test->tes_param[0], param, paramlen);
1311         }
1312
1313         rc = lstcon_test_nodes_add(test, result_up);
1314
1315         if (rc != 0)
1316                 goto out;
1317
1318         if (lstcon_trans_stat()->trs_rpc_errno != 0 ||
1319             lstcon_trans_stat()->trs_fwk_errno != 0)
1320                 CDEBUG(D_NET, "Failed to add test %d to batch %s\n", type,
1321                        batch_name);
1322
1323         /* add to test list anyway, so user can check what's going on */
1324         list_add_tail(&test->tes_link, &batch->bat_test_list);
1325
1326         batch->bat_ntest++;
1327         test->tes_hdr.tsb_index = batch->bat_ntest;
1328
1329         /*  hold groups so nobody can change them */
1330         return rc;
1331 out:
1332         if (test != NULL)
1333                 LIBCFS_FREE(test, offsetof(struct lstcon_test,
1334                                            tes_param[paramlen]));
1335
1336         if (dst_grp != NULL)
1337                 lstcon_group_decref(dst_grp);
1338
1339         if (src_grp != NULL)
1340                 lstcon_group_decref(src_grp);
1341
1342         return rc;
1343 }
1344
1345 static int
1346 lstcon_test_find(struct lstcon_batch *batch, int idx,
1347                  struct lstcon_test **testpp)
1348 {
1349         struct lstcon_test *test;
1350
1351         list_for_each_entry(test, &batch->bat_test_list, tes_link) {
1352                 if (idx == test->tes_hdr.tsb_index) {
1353                         *testpp = test;
1354                         return 0;
1355                 }
1356         }
1357
1358         return -ENOENT;
1359 }
1360
1361 static int
1362 lstcon_tsbrpc_readent(int transop, struct srpc_msg *msg,
1363                       struct lstcon_rpc_ent __user *ent_up)
1364 {
1365         struct srpc_batch_reply *rep = &msg->msg_body.bat_reply;
1366
1367         LASSERT (transop == LST_TRANS_TSBCLIQRY ||
1368                  transop == LST_TRANS_TSBSRVQRY);
1369
1370         /* positive errno, framework error code */
1371         if (copy_to_user(&ent_up->rpe_priv[0],
1372                              &rep->bar_active, sizeof(rep->bar_active)))
1373                 return -EFAULT;
1374
1375         return 0;
1376 }
1377
1378 int
1379 lstcon_test_batch_query(char *name, int testidx, int client,
1380                         int timeout, struct list_head __user *result_up)
1381 {
1382         struct lstcon_rpc_trans *trans;
1383         struct list_head *translist;
1384         struct list_head *ndlist;
1385         struct lstcon_tsb_hdr *hdr;
1386         struct lstcon_batch *batch;
1387         struct lstcon_test *test = NULL;
1388         int transop;
1389         int rc;
1390
1391         rc = lstcon_batch_find(name, &batch);
1392         if (rc != 0) {
1393                 CDEBUG(D_NET, "Can't find batch: %s\n", name);
1394                 return rc;
1395         }
1396
1397         if (testidx == 0) {
1398                 translist = &batch->bat_trans_list;
1399                 ndlist    = &batch->bat_cli_list;
1400                 hdr       = &batch->bat_hdr;
1401
1402         } else {
1403                 /* query specified test only */
1404                 rc = lstcon_test_find(batch, testidx, &test);
1405                 if (rc != 0) {
1406                         CDEBUG(D_NET, "Can't find test: %d\n", testidx);
1407                         return rc;
1408                 }
1409
1410                 translist = &test->tes_trans_list;
1411                 ndlist    = &test->tes_src_grp->grp_ndl_list;
1412                 hdr       = &test->tes_hdr;
1413         }
1414
1415         transop = client ? LST_TRANS_TSBCLIQRY : LST_TRANS_TSBSRVQRY;
1416
1417         rc = lstcon_rpc_trans_ndlist(ndlist, translist, transop, hdr,
1418                                      lstcon_batrpc_condition, &trans);
1419         if (rc != 0) {
1420                 CERROR("Can't create transaction: %d\n", rc);
1421                 return rc;
1422         }
1423
1424         lstcon_rpc_trans_postwait(trans, timeout);
1425
1426         if (testidx == 0 && /* query a batch, not a test */
1427             lstcon_rpc_stat_failure(lstcon_trans_stat(), 0) == 0 &&
1428             lstcon_tsbqry_stat_run(lstcon_trans_stat(), 0) == 0) {
1429                 /* all RPCs finished, and no active test */
1430                 batch->bat_state = LST_BATCH_IDLE;
1431         }
1432
1433         rc = lstcon_rpc_trans_interpreter(trans, result_up,
1434                                           lstcon_tsbrpc_readent);
1435         lstcon_rpc_trans_destroy(trans);
1436
1437         return rc;
1438 }
1439
1440 static int
1441 lstcon_statrpc_readent(int transop, struct srpc_msg *msg,
1442                        struct lstcon_rpc_ent __user *ent_up)
1443 {
1444         struct srpc_stat_reply *rep = &msg->msg_body.stat_reply;
1445         struct sfw_counters __user *sfwk_stat;
1446         struct srpc_counters __user *srpc_stat;
1447         struct lnet_counters_common __user *lnet_stat;
1448
1449         if (rep->str_status != 0)
1450                 return 0;
1451
1452         sfwk_stat = (struct sfw_counters __user *)&ent_up->rpe_payload[0];
1453         srpc_stat = (struct srpc_counters __user *)
1454                 ((char __user *)sfwk_stat + sizeof(*sfwk_stat));
1455         lnet_stat = (struct lnet_counters_common __user *)
1456                 ((char __user *)srpc_stat + sizeof(*srpc_stat));
1457
1458         if (copy_to_user(sfwk_stat, &rep->str_fw, sizeof(*sfwk_stat)) ||
1459             copy_to_user(srpc_stat, &rep->str_rpc, sizeof(*srpc_stat)) ||
1460             copy_to_user(lnet_stat, &rep->str_lnet, sizeof(*lnet_stat)))
1461                 return -EFAULT;
1462
1463         return 0;
1464 }
1465
1466 static int
1467 lstcon_ndlist_stat(struct list_head *ndlist,
1468                    int timeout, struct list_head __user *result_up)
1469 {
1470         LIST_HEAD(head);
1471         struct lstcon_rpc_trans *trans;
1472         int rc;
1473
1474         rc = lstcon_rpc_trans_ndlist(ndlist, &head,
1475                                      LST_TRANS_STATQRY, NULL, NULL, &trans);
1476         if (rc != 0) {
1477                 CERROR("Can't create transaction: %d\n", rc);
1478                 return rc;
1479         }
1480
1481         lstcon_rpc_trans_postwait(trans, LST_VALIDATE_TIMEOUT(timeout));
1482
1483         rc = lstcon_rpc_trans_interpreter(trans, result_up,
1484                                           lstcon_statrpc_readent);
1485         lstcon_rpc_trans_destroy(trans);
1486
1487         return rc;
1488 }
1489
1490 int
1491 lstcon_group_stat(char *grp_name, int timeout,
1492                   struct list_head __user *result_up)
1493 {
1494         struct lstcon_group *grp;
1495         int rc;
1496
1497         rc = lstcon_group_find(grp_name, &grp);
1498         if (rc != 0) {
1499                 CDEBUG(D_NET, "Can't find group %s\n", grp_name);
1500                 return rc;
1501         }
1502
1503         rc = lstcon_ndlist_stat(&grp->grp_ndl_list, timeout, result_up);
1504
1505         lstcon_group_decref(grp);
1506
1507         return rc;
1508 }
1509
1510 int
1511 lstcon_nodes_stat(int count, struct lnet_process_id __user *ids_up,
1512                   int timeout, struct list_head __user *result_up)
1513 {
1514         struct lstcon_ndlink *ndl;
1515         struct lstcon_group *tmp;
1516         struct lnet_process_id id;
1517         int i;
1518         int rc;
1519
1520         rc = lstcon_group_alloc(NULL, &tmp);
1521         if (rc != 0) {
1522                 CERROR("Out of memory\n");
1523                 return -ENOMEM;
1524         }
1525
1526         for (i = 0 ; i < count; i++) {
1527                 if (copy_from_user(&id, &ids_up[i], sizeof(id))) {
1528                         rc = -EFAULT;
1529                         break;
1530                 }
1531
1532                 /* add to tmp group */
1533                 rc = lstcon_group_ndlink_find(tmp, id, &ndl, 2);
1534                 if (rc != 0) {
1535                         CDEBUG((rc == -ENOMEM) ? D_ERROR : D_NET,
1536                                "Failed to find or create %s: %d\n",
1537                                libcfs_id2str(id), rc);
1538                         break;
1539                 }
1540         }
1541
1542         if (rc != 0) {
1543                 lstcon_group_decref(tmp);
1544                 return rc;
1545         }
1546
1547         rc = lstcon_ndlist_stat(&tmp->grp_ndl_list, timeout, result_up);
1548
1549         lstcon_group_decref(tmp);
1550
1551         return rc;
1552 }
1553
1554 static int
1555 lstcon_debug_ndlist(struct list_head *ndlist,
1556                     struct list_head *translist,
1557                     int timeout, struct list_head __user *result_up)
1558 {
1559         struct lstcon_rpc_trans *trans;
1560         int rc;
1561
1562         rc = lstcon_rpc_trans_ndlist(ndlist, translist, LST_TRANS_SESQRY,
1563                                      NULL, lstcon_sesrpc_condition, &trans);
1564         if (rc != 0) {
1565                 CERROR("Can't create transaction: %d\n", rc);
1566                 return rc;
1567         }
1568
1569         lstcon_rpc_trans_postwait(trans, LST_VALIDATE_TIMEOUT(timeout));
1570
1571         rc = lstcon_rpc_trans_interpreter(trans, result_up,
1572                                           lstcon_sesrpc_readent);
1573         lstcon_rpc_trans_destroy(trans);
1574
1575         return rc;
1576 }
1577
1578 int
1579 lstcon_session_debug(int timeout, struct list_head __user *result_up)
1580 {
1581         return lstcon_debug_ndlist(&console_session.ses_ndl_list,
1582                                    NULL, timeout, result_up);
1583 }
1584
1585 int
1586 lstcon_batch_debug(int timeout, char *name,
1587                    int client, struct list_head __user *result_up)
1588 {
1589         struct lstcon_batch *bat;
1590         int rc;
1591
1592         rc = lstcon_batch_find(name, &bat);
1593         if (rc != 0)
1594                 return -ENOENT;
1595
1596         rc = lstcon_debug_ndlist(client ? &bat->bat_cli_list :
1597                                           &bat->bat_srv_list,
1598                                  NULL, timeout, result_up);
1599
1600         return rc;
1601 }
1602
1603 int
1604 lstcon_group_debug(int timeout, char *name,
1605                    struct list_head __user *result_up)
1606 {
1607         struct lstcon_group *grp;
1608         int rc;
1609
1610         rc = lstcon_group_find(name, &grp);
1611         if (rc != 0)
1612                 return -ENOENT;
1613
1614         rc = lstcon_debug_ndlist(&grp->grp_ndl_list, NULL,
1615                                  timeout, result_up);
1616         lstcon_group_decref(grp);
1617
1618         return rc;
1619 }
1620
1621 int
1622 lstcon_nodes_debug(int timeout, int count,
1623                    struct lnet_process_id __user *ids_up,
1624                    struct list_head __user *result_up)
1625 {
1626         struct lnet_process_id id;
1627         struct lstcon_ndlink *ndl;
1628         struct lstcon_group *grp;
1629         int i;
1630         int rc;
1631
1632         rc = lstcon_group_alloc(NULL, &grp);
1633         if (rc != 0) {
1634                 CDEBUG(D_NET, "Out of memory\n");
1635                 return rc;
1636         }
1637
1638         for (i = 0; i < count; i++) {
1639                 if (copy_from_user(&id, &ids_up[i], sizeof(id))) {
1640                         rc = -EFAULT;
1641                         break;
1642                 }
1643
1644                 /* node is added to tmp group */
1645                 rc = lstcon_group_ndlink_find(grp, id, &ndl, 1);
1646                 if (rc != 0) {
1647                         CERROR("Can't create node link\n");
1648                         break;
1649                 }
1650         }
1651
1652         if (rc != 0) {
1653                 lstcon_group_decref(grp);
1654                 return rc;
1655         }
1656
1657         rc = lstcon_debug_ndlist(&grp->grp_ndl_list, NULL,
1658                                  timeout, result_up);
1659
1660         lstcon_group_decref(grp);
1661
1662         return rc;
1663 }
1664
1665 int
1666 lstcon_session_match(struct lst_sid id)
1667 {
1668         struct lst_session_id sid;
1669
1670         sid.ses_stamp = id.ses_stamp;
1671         lnet_nid4_to_nid(id.ses_nid, &sid.ses_nid);
1672
1673         return (nid_same(&console_session.ses_id.ses_nid, &sid.ses_nid) &&
1674                 console_session.ses_id.ses_stamp == sid.ses_stamp) ?  1 : 0;
1675 }
1676
1677 static void
1678 lstcon_new_session_id(struct lst_session_id *sid)
1679 {
1680         struct lnet_processid id;
1681
1682         LASSERT(console_session.ses_state == LST_SESSION_NONE);
1683
1684         LNetGetId(1, &id);
1685         sid->ses_nid = id.nid;
1686         sid->ses_stamp = div_u64(ktime_get_ns(), NSEC_PER_MSEC);
1687 }
1688
1689 int
1690 lstcon_session_new(char *name, int key, unsigned feats,
1691                    int timeout, int force)
1692 {
1693         int     rc = 0;
1694         int     i;
1695
1696         if (console_session.ses_state != LST_SESSION_NONE) {
1697                 /* session exists */
1698                 if (!force) {
1699                         CNETERR("Session %s already exists\n",
1700                                 console_session.ses_name);
1701                         return -EEXIST;
1702                 }
1703
1704                 rc = lstcon_session_end();
1705
1706                 /* lstcon_session_end() only return local error */
1707                 if  (rc != 0)
1708                         return rc;
1709         }
1710
1711         if ((feats & ~LST_FEATS_MASK) != 0) {
1712                 CNETERR("Unknown session features %x\n",
1713                         (feats & ~LST_FEATS_MASK));
1714                 return -EINVAL;
1715         }
1716
1717         for (i = 0; i < LST_GLOBAL_HASHSIZE; i++)
1718                 LASSERT(list_empty(&console_session.ses_ndl_hash[i]));
1719
1720         lstcon_new_session_id(&console_session.ses_id);
1721
1722         console_session.ses_key     = key;
1723         console_session.ses_force   = !!force;
1724         console_session.ses_features = feats;
1725         console_session.ses_feats_updated = 0;
1726         console_session.ses_timeout = (timeout <= 0) ?
1727                                       LST_CONSOLE_TIMEOUT : timeout;
1728
1729         if (strlen(name) > sizeof(console_session.ses_name)-1)
1730                 return -E2BIG;
1731         strlcpy(console_session.ses_name, name,
1732                 sizeof(console_session.ses_name));
1733
1734         rc = lstcon_batch_add(LST_DEFAULT_BATCH);
1735         if (rc != 0)
1736                 return rc;
1737
1738         rc = lstcon_rpc_pinger_start();
1739         if (rc != 0) {
1740                 struct lstcon_batch *bat = NULL;
1741
1742                 lstcon_batch_find(LST_DEFAULT_BATCH, &bat);
1743                 lstcon_batch_destroy(bat);
1744
1745                 return rc;
1746         }
1747
1748         console_session.ses_state = LST_SESSION_ACTIVE;
1749
1750         return rc;
1751 }
1752
1753 int lstcon_session_end(void)
1754 {
1755         struct lstcon_rpc_trans *trans;
1756         struct lstcon_group *grp;
1757         struct lstcon_batch *bat;
1758         int rc = 0;
1759
1760         LASSERT (console_session.ses_state == LST_SESSION_ACTIVE);
1761
1762         rc = lstcon_rpc_trans_ndlist(&console_session.ses_ndl_list,
1763                                      NULL, LST_TRANS_SESEND, NULL,
1764                                      lstcon_sesrpc_condition, &trans);
1765         if (rc != 0) {
1766                 CERROR("Can't create transaction: %d\n", rc);
1767                 return rc;
1768         }
1769
1770         console_session.ses_shutdown = 1;
1771
1772         lstcon_rpc_pinger_stop();
1773
1774         lstcon_rpc_trans_postwait(trans, LST_TRANS_TIMEOUT);
1775
1776         lstcon_rpc_trans_destroy(trans);
1777         /* User can do nothing even rpc failed, so go on */
1778
1779         /* waiting for orphan rpcs to die */
1780         lstcon_rpc_cleanup_wait();
1781
1782         console_session.ses_id    = LST_INVALID_SID;
1783         console_session.ses_state = LST_SESSION_NONE;
1784         console_session.ses_key   = 0;
1785         console_session.ses_force = 0;
1786         console_session.ses_feats_updated = 0;
1787
1788         /* destroy all batches */
1789         while (!list_empty(&console_session.ses_bat_list)) {
1790                 bat = list_entry(console_session.ses_bat_list.next,
1791                                  struct lstcon_batch, bat_link);
1792
1793                 lstcon_batch_destroy(bat);
1794         }
1795
1796         /* destroy all groups */
1797         while (!list_empty(&console_session.ses_grp_list)) {
1798                 grp = list_entry(console_session.ses_grp_list.next,
1799                                  struct lstcon_group, grp_link);
1800                 LASSERT(grp->grp_ref == 1);
1801
1802                 lstcon_group_decref(grp);
1803         }
1804
1805         /* all nodes should be released */
1806         LASSERT(list_empty(&console_session.ses_ndl_list));
1807
1808         console_session.ses_shutdown = 0;
1809         console_session.ses_expired  = 0;
1810
1811         return rc;
1812 }
1813
1814 int
1815 lstcon_session_feats_check(unsigned feats)
1816 {
1817         int rc = 0;
1818
1819         if ((feats & ~LST_FEATS_MASK) != 0) {
1820                 CERROR("Can't support these features: %x\n",
1821                        (feats & ~LST_FEATS_MASK));
1822                 return -EPROTO;
1823         }
1824
1825         spin_lock(&console_session.ses_rpc_lock);
1826
1827         if (!console_session.ses_feats_updated) {
1828                 console_session.ses_feats_updated = 1;
1829                 console_session.ses_features = feats;
1830         }
1831
1832         if (console_session.ses_features != feats)
1833                 rc = -EPROTO;
1834
1835         spin_unlock(&console_session.ses_rpc_lock);
1836
1837         if (rc != 0) {
1838                 CERROR("remote features %x do not match with "
1839                        "session features %x of console\n",
1840                        feats, console_session.ses_features);
1841         }
1842
1843         return rc;
1844 }
1845
1846 static int
1847 lstcon_acceptor_handle(struct srpc_server_rpc *rpc)
1848 {
1849         struct srpc_msg *rep = &rpc->srpc_replymsg;
1850         struct srpc_msg *req = &rpc->srpc_reqstbuf->buf_msg;
1851         struct srpc_join_reqst *jreq = &req->msg_body.join_reqst;
1852         struct srpc_join_reply *jrep = &rep->msg_body.join_reply;
1853         struct lstcon_group *grp = NULL;
1854         struct lstcon_ndlink *ndl;
1855         int rc = 0;
1856
1857         sfw_unpack_message(req);
1858
1859         mutex_lock(&console_session.ses_mutex);
1860
1861         jrep->join_sid.ses_stamp = console_session.ses_id.ses_stamp;
1862         jrep->join_sid.ses_nid = lnet_nid_to_nid4(&console_session.ses_id.ses_nid);
1863
1864         if (LNET_NID_IS_ANY(&console_session.ses_id.ses_nid)) {
1865                 jrep->join_status = ESRCH;
1866                 goto out;
1867         }
1868
1869         if (lstcon_session_feats_check(req->msg_ses_feats) != 0) {
1870                 jrep->join_status = EPROTO;
1871                 goto out;
1872         }
1873
1874         if (jreq->join_sid.ses_nid != LNET_NID_ANY &&
1875              !lstcon_session_match(jreq->join_sid)) {
1876                 jrep->join_status = EBUSY;
1877                 goto out;
1878         }
1879
1880         if (lstcon_group_find(jreq->join_group, &grp) != 0) {
1881                 rc = lstcon_group_alloc(jreq->join_group, &grp);
1882                 if (rc != 0) {
1883                         CERROR("Out of memory\n");
1884                         goto out;
1885                 }
1886
1887                 list_add_tail(&grp->grp_link,
1888                               &console_session.ses_grp_list);
1889                 lstcon_group_addref(grp);
1890         }
1891
1892         if (grp->grp_ref > 2) {
1893                 /* Group in using */
1894                 jrep->join_status = EBUSY;
1895                 goto out;
1896         }
1897
1898         rc = lstcon_group_ndlink_find(grp, rpc->srpc_peer, &ndl, 0);
1899         if (rc == 0) {
1900                 jrep->join_status = EEXIST;
1901                 goto out;
1902         }
1903
1904         rc = lstcon_group_ndlink_find(grp, rpc->srpc_peer, &ndl, 1);
1905         if (rc != 0) {
1906                 CERROR("Out of memory\n");
1907                 goto out;
1908         }
1909
1910         ndl->ndl_node->nd_state   = LST_NODE_ACTIVE;
1911         ndl->ndl_node->nd_timeout = console_session.ses_timeout;
1912
1913         if (grp->grp_userland == 0)
1914                 grp->grp_userland = 1;
1915
1916         strlcpy(jrep->join_session, console_session.ses_name,
1917                 sizeof(jrep->join_session));
1918         jrep->join_timeout = console_session.ses_timeout;
1919         jrep->join_status  = 0;
1920
1921 out:
1922         rep->msg_ses_feats = console_session.ses_features;
1923         if (grp != NULL)
1924                 lstcon_group_decref(grp);
1925
1926         mutex_unlock(&console_session.ses_mutex);
1927
1928         return rc;
1929 }
1930
1931 static struct srpc_service lstcon_acceptor_service;
1932
1933 static void lstcon_init_acceptor_service(void)
1934 {
1935         /* initialize selftest console acceptor service table */
1936         lstcon_acceptor_service.sv_name    = "join session";
1937         lstcon_acceptor_service.sv_handler = lstcon_acceptor_handle;
1938         lstcon_acceptor_service.sv_id      = SRPC_SERVICE_JOIN;
1939         lstcon_acceptor_service.sv_wi_total = SFW_FRWK_WI_MAX;
1940 }
1941
1942 static struct notifier_block lstcon_ioctl_handler = {
1943         .notifier_call = lstcon_ioctl_entry,
1944 };
1945
1946 /* initialize console */
1947 int
1948 lstcon_console_init(void)
1949 {
1950         int     i;
1951         int     rc;
1952
1953         console_session.ses_id              = LST_INVALID_SID;
1954         console_session.ses_state           = LST_SESSION_NONE;
1955         console_session.ses_timeout         = 0;
1956         console_session.ses_force           = 0;
1957         console_session.ses_expired         = 0;
1958         console_session.ses_feats_updated   = 0;
1959         console_session.ses_features        = LST_FEATS_MASK;
1960         console_session.ses_laststamp = ktime_get_real_seconds();
1961
1962         mutex_init(&console_session.ses_mutex);
1963
1964         INIT_LIST_HEAD(&console_session.ses_ndl_list);
1965         INIT_LIST_HEAD(&console_session.ses_grp_list);
1966         INIT_LIST_HEAD(&console_session.ses_bat_list);
1967         INIT_LIST_HEAD(&console_session.ses_trans_list);
1968
1969         CFS_ALLOC_PTR_ARRAY(console_session.ses_ndl_hash,
1970                                LST_GLOBAL_HASHSIZE);
1971         if (console_session.ses_ndl_hash == NULL)
1972                 return -ENOMEM;
1973
1974         for (i = 0; i < LST_GLOBAL_HASHSIZE; i++)
1975                 INIT_LIST_HEAD(&console_session.ses_ndl_hash[i]);
1976
1977
1978         /* initialize acceptor service table */
1979         lstcon_init_acceptor_service();
1980
1981         rc = srpc_add_service(&lstcon_acceptor_service);
1982         LASSERT(rc != -EBUSY);
1983         if (rc != 0) {
1984                 CFS_FREE_PTR_ARRAY(console_session.ses_ndl_hash,
1985                                    LST_GLOBAL_HASHSIZE);
1986                 return rc;
1987         }
1988
1989         rc = srpc_service_add_buffers(&lstcon_acceptor_service,
1990                                       lstcon_acceptor_service.sv_wi_total);
1991         if (rc != 0) {
1992                 rc = -ENOMEM;
1993                 goto out;
1994         }
1995
1996         rc = lstcon_init_netlink();
1997         if (rc < 0)
1998                 goto out;
1999
2000         rc = blocking_notifier_chain_register(&libcfs_ioctl_list,
2001                                               &lstcon_ioctl_handler);
2002         if (rc < 0) {
2003                 lstcon_fini_netlink();
2004                 goto out;
2005         }
2006
2007         lstcon_rpc_module_init();
2008         return 0;
2009
2010 out:
2011         srpc_shutdown_service(&lstcon_acceptor_service);
2012         srpc_remove_service(&lstcon_acceptor_service);
2013
2014         CFS_FREE_PTR_ARRAY(console_session.ses_ndl_hash, LST_GLOBAL_HASHSIZE);
2015
2016         srpc_wait_service_shutdown(&lstcon_acceptor_service);
2017
2018         return rc;
2019 }
2020
2021 int
2022 lstcon_console_fini(void)
2023 {
2024         int i;
2025
2026         blocking_notifier_chain_unregister(&libcfs_ioctl_list,
2027                                            &lstcon_ioctl_handler);
2028         lstcon_fini_netlink();
2029
2030         mutex_lock(&console_session.ses_mutex);
2031
2032         srpc_shutdown_service(&lstcon_acceptor_service);
2033         srpc_remove_service(&lstcon_acceptor_service);
2034
2035         if (console_session.ses_state != LST_SESSION_NONE)
2036                 lstcon_session_end();
2037
2038         lstcon_rpc_module_fini();
2039
2040         mutex_unlock(&console_session.ses_mutex);
2041
2042         LASSERT(list_empty(&console_session.ses_ndl_list));
2043         LASSERT(list_empty(&console_session.ses_grp_list));
2044         LASSERT(list_empty(&console_session.ses_bat_list));
2045         LASSERT(list_empty(&console_session.ses_trans_list));
2046
2047         for (i = 0; i < LST_NODE_HASHSIZE; i++)
2048                 LASSERT(list_empty(&console_session.ses_ndl_hash[i]));
2049
2050         CFS_FREE_PTR_ARRAY(console_session.ses_ndl_hash,
2051                            LST_GLOBAL_HASHSIZE);
2052
2053         srpc_wait_service_shutdown(&lstcon_acceptor_service);
2054
2055         return 0;
2056 }
2057