Whamcloud - gitweb
7c5eb0e14d7a5cb719e3aa9a692585969724cb60
[fs/lustre-release.git] / lnet / selftest / conrpc.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  * 
4  * Author: Liang Zhen <liangzhen@clusterfs.com>
5  *
6  * This file is part of Lustre, http://www.lustre.org
7  *
8  * Console framework rpcs
9  */
10 #ifdef __KERNEL__
11
12 #include <libcfs/libcfs.h>
13 #include <lnet/lib-lnet.h>
14 #include "timer.h"
15 #include "conrpc.h"
16 #include "console.h"
17
18 void lstcon_rpc_stat_reply(int, srpc_msg_t *,
19                            lstcon_node_t *, lstcon_trans_stat_t *);
20
21 static void
22 lstcon_rpc_done(srpc_client_rpc_t *rpc)
23 {
24         lstcon_rpc_t *crpc = (lstcon_rpc_t *)rpc->crpc_priv;
25
26         LASSERT (!list_empty(&crpc->crp_link));
27         LASSERT (crpc != NULL && rpc == crpc->crp_rpc);
28         LASSERT (crpc->crp_posted && !crpc->crp_finished);
29
30         spin_lock(&rpc->crpc_lock);
31
32         if (crpc->crp_trans == NULL) {
33                 /* orphan RPC */
34                 spin_lock(&console_session.ses_rpc_lock);
35
36                 /* delete from orphan rpcs list */
37                 console_session.ses_rpc_pending --;
38                 list_del_init(&crpc->crp_link);
39
40                 spin_unlock(&console_session.ses_rpc_lock);
41
42                 spin_unlock(&rpc->crpc_lock);
43
44                 /* release it */
45                 lstcon_rpc_put(crpc);
46                 return;
47         }
48
49         /* not an orphan RPC */
50         crpc->crp_finished = 1;
51
52         if (crpc->crp_stamp == 0) {
53                 /* not aborted */
54                 LASSERT (crpc->crp_status == 0);
55
56                 crpc->crp_stamp  = cfs_time_current();
57                 crpc->crp_status = rpc->crpc_status;
58         }
59
60         /* wakeup thread waiting on the group if 
61          * it's the last rpc in the group */
62         if (atomic_dec_and_test(&crpc->crp_trans->tas_remaining))
63                 cfs_waitq_signal(&crpc->crp_trans->tas_waitq);
64
65         spin_unlock(&rpc->crpc_lock);
66 }
67
68 int
69 lstcon_rpc_init(lstcon_node_t *nd, int service,
70                 int npg, int cached, lstcon_rpc_t *crpc)
71 {
72
73         crpc->crp_rpc = sfw_create_rpc(nd->nd_id, service, 
74                                        npg, npg * CFS_PAGE_SIZE,
75                                        lstcon_rpc_done, (void *)crpc);
76         if (crpc->crp_rpc == NULL)
77                 return -ENOMEM;
78
79         crpc->crp_trans    = NULL;
80         crpc->crp_node     = nd;
81         crpc->crp_posted   = 0;
82         crpc->crp_finished = 0;
83         crpc->crp_unpacked = 0;
84         crpc->crp_status   = 0;
85         crpc->crp_stamp    = 0;
86         crpc->crp_static   = !cached;
87         CFS_INIT_LIST_HEAD(&crpc->crp_link);
88
89         return 0;
90 }
91
92 int
93 lstcon_rpc_prep(lstcon_node_t *nd, int service,
94                 int npg, lstcon_rpc_t **crpcpp)
95 {
96         lstcon_rpc_t  *crpc = NULL;
97         int            rc;
98
99         spin_lock(&console_session.ses_rpc_lock);
100
101         if (!list_empty(&console_session.ses_rpc_freelist)) {
102                 crpc = list_entry(console_session.ses_rpc_freelist.next,
103                                   lstcon_rpc_t, crp_link);
104                 list_del(&crpc->crp_link);
105         }
106
107         spin_unlock(&console_session.ses_rpc_lock);
108
109         if (crpc == NULL) {
110                 LIBCFS_ALLOC(crpc, sizeof(*crpc));
111                 if (crpc == NULL)
112                         return -ENOMEM;
113         }
114
115         rc = lstcon_rpc_init(nd, service, npg, 1, crpc);
116         if (rc == 0) {
117                 *crpcpp = crpc;
118                 return 0;
119         }
120
121         LIBCFS_FREE(crpc, sizeof(*crpc));
122
123         return rc;
124 }
125
126 void
127 lstcon_rpc_put(lstcon_rpc_t *crpc)
128 {
129         srpc_bulk_t *bulk = &crpc->crp_rpc->crpc_bulk;
130         int          i;
131
132         LASSERT (list_empty(&crpc->crp_link));
133
134         for (i = 0; i < bulk->bk_niov; i++) {
135                 if (bulk->bk_iovs[i].kiov_page == NULL)
136                         continue;
137
138                 cfs_free_page(bulk->bk_iovs[i].kiov_page);
139         }
140
141         srpc_client_rpc_decref(crpc->crp_rpc);
142
143         if (crpc->crp_static) {
144                 memset(crpc, 0, sizeof(*crpc));
145                 crpc->crp_static = 1;
146                 return;
147         }
148
149         spin_lock(&console_session.ses_rpc_lock);
150
151         list_add(&crpc->crp_link, &console_session.ses_rpc_freelist);
152
153         spin_unlock(&console_session.ses_rpc_lock);
154 }
155
156 void
157 lstcon_rpc_post(lstcon_rpc_t *crpc)
158 {
159         lstcon_rpc_trans_t *trans = crpc->crp_trans;
160
161         LASSERT (trans != NULL);
162
163         atomic_inc(&trans->tas_remaining);
164         crpc->crp_posted = 1;
165
166         sfw_post_rpc(crpc->crp_rpc);
167 }
168
169 static char *
170 lstcon_rpc_trans_name(int transop)
171 {
172         if (transop == LST_TRANS_SESNEW)
173                 return "SESNEW";
174
175         if (transop == LST_TRANS_SESEND)
176                 return "SESEND";
177
178         if (transop == LST_TRANS_SESQRY)
179                 return "SESQRY";
180
181         if (transop == LST_TRANS_SESPING)
182                 return "SESPING";
183
184         if (transop == LST_TRANS_TSBCLIADD)
185                 return "TSBCLIADD";
186
187         if (transop == LST_TRANS_TSBSRVADD)
188                 return "TSBSRVADD";
189
190         if (transop == LST_TRANS_TSBRUN)
191                 return "TSBRUN";
192
193         if (transop == LST_TRANS_TSBSTOP)
194                 return "TSBSTOP";
195
196         if (transop == LST_TRANS_TSBCLIQRY)
197                 return "TSBCLIQRY";
198
199         if (transop == LST_TRANS_TSBSRVQRY)
200                 return "TSBSRVQRY";
201
202         if (transop == LST_TRANS_STATQRY)
203                 return "STATQRY";
204
205         return "Unknown";
206 }
207
208 int
209 lstcon_rpc_trans_prep(struct list_head *translist,
210                       int transop, lstcon_rpc_trans_t **transpp)
211 {
212         lstcon_rpc_trans_t *trans;
213
214         if (translist != NULL) {
215                 list_for_each_entry(trans, translist, tas_link) {
216                         /* Can't enqueue two private transaction on
217                          * the same object */
218                         if ((trans->tas_opc & transop) == LST_TRANS_PRIVATE)
219                                 return -EPERM;
220                 }
221         }
222
223         /* create a trans group */
224         LIBCFS_ALLOC(trans, sizeof(*trans));
225         if (trans == NULL)
226                 return -ENOMEM;
227         
228         trans->tas_opc = transop;
229
230         if (translist == NULL)       
231                 CFS_INIT_LIST_HEAD(&trans->tas_olink);
232         else
233                 list_add_tail(&trans->tas_olink, translist);
234
235         list_add_tail(&trans->tas_link, &console_session.ses_trans_list);
236
237         CFS_INIT_LIST_HEAD(&trans->tas_rpcs_list);
238         atomic_set(&trans->tas_remaining, 0);
239         cfs_waitq_init(&trans->tas_waitq);
240
241         *transpp = trans;
242
243         return 0;
244 }
245
246 void
247 lstcon_rpc_trans_addreq(lstcon_rpc_trans_t *trans, lstcon_rpc_t *crpc)
248 {
249         list_add_tail(&crpc->crp_link, &trans->tas_rpcs_list);
250         crpc->crp_trans = trans;
251 }
252
253 void
254 lstcon_rpc_trans_abort(lstcon_rpc_trans_t *trans, int error)
255 {
256         srpc_client_rpc_t *rpc;
257         lstcon_rpc_t      *crpc;
258         lstcon_node_t     *nd;
259
260         list_for_each_entry (crpc, &trans->tas_rpcs_list, crp_link) {
261                 rpc = crpc->crp_rpc;
262
263                 spin_lock(&rpc->crpc_lock);
264
265                 if (!crpc->crp_posted || crpc->crp_stamp != 0) {
266                         /* rpc done or aborted already */
267                         spin_unlock(&rpc->crpc_lock);
268                         continue;
269                 }
270
271                 crpc->crp_stamp  = cfs_time_current();
272                 crpc->crp_status = error;
273
274                 spin_unlock(&rpc->crpc_lock);
275
276                 sfw_abort_rpc(rpc);
277
278                 if  (error != ETIMEDOUT)
279                         continue;
280
281                 nd = crpc->crp_node;
282                 if (cfs_time_after(nd->nd_stamp, crpc->crp_stamp))
283                         continue;
284
285                 nd->nd_stamp = crpc->crp_stamp;
286                 nd->nd_state = LST_NODE_DOWN;
287         }
288 }
289
290 static int
291 lstcon_rpc_trans_check(lstcon_rpc_trans_t *trans)
292 {
293         if (console_session.ses_shutdown &&
294             !list_empty(&trans->tas_olink)) /* It's not an end session RPC */
295                 return 1;
296
297         return (atomic_read(&trans->tas_remaining) == 0) ? 1: 0;
298 }
299
300 int
301 lstcon_rpc_trans_postwait(lstcon_rpc_trans_t *trans, int timeout)
302 {
303         lstcon_rpc_t  *crpc;
304         int            rc;
305
306         if (list_empty(&trans->tas_rpcs_list))
307                 return 0;
308
309         if (timeout < LST_TRANS_MIN_TIMEOUT)
310                 timeout = LST_TRANS_MIN_TIMEOUT;
311
312         CDEBUG(D_NET, "Transaction %s started\n",
313                lstcon_rpc_trans_name(trans->tas_opc));
314
315         /* post all requests */
316         list_for_each_entry (crpc, &trans->tas_rpcs_list, crp_link) {
317                 LASSERT (!crpc->crp_posted);
318
319                 lstcon_rpc_post(crpc);
320         }
321
322         mutex_up(&console_session.ses_mutex);
323
324         rc = wait_event_interruptible_timeout(trans->tas_waitq,
325                                               lstcon_rpc_trans_check(trans),
326                                               timeout * HZ);
327
328         rc = (rc > 0)? 0: ((rc < 0)? -EINTR: -ETIMEDOUT);
329
330         mutex_down(&console_session.ses_mutex);
331
332         if (console_session.ses_shutdown)
333                 rc = -ESHUTDOWN;
334
335         if (rc != 0) {
336                 /* treat short timeout as canceled */
337                 if (rc == -ETIMEDOUT && timeout < LST_TRANS_MIN_TIMEOUT * 2)
338                         rc = -EINTR;
339
340                 lstcon_rpc_trans_abort(trans, rc);
341         }
342
343         CDEBUG(D_NET, "Transaction %s stopped: %d\n",
344                lstcon_rpc_trans_name(trans->tas_opc), rc);
345
346         lstcon_rpc_trans_stat(trans, lstcon_trans_stat());
347
348         return rc;
349 }
350
351 int
352 lstcon_rpc_get_reply(lstcon_rpc_t *crpc, srpc_msg_t **msgpp)
353 {
354         lstcon_node_t        *nd  = crpc->crp_node;
355         srpc_client_rpc_t    *rpc = crpc->crp_rpc;
356         srpc_generic_reply_t *rep;
357
358         LASSERT (nd != NULL && rpc != NULL);
359         LASSERT (crpc->crp_stamp != 0);
360
361         if (crpc->crp_status != 0) {
362                 *msgpp = NULL;
363                 return crpc->crp_status;
364         }
365
366         *msgpp = &rpc->crpc_replymsg;
367         if (!crpc->crp_unpacked) {
368                 sfw_unpack_message(*msgpp);
369                 crpc->crp_unpacked = 1;
370         }
371        
372         if (cfs_time_after(nd->nd_stamp, crpc->crp_stamp))
373                 return 0;
374
375         nd->nd_stamp = crpc->crp_stamp;
376         rep = &(*msgpp)->msg_body.reply;
377
378         if (rep->sid.ses_nid == LNET_NID_ANY)
379                 nd->nd_state = LST_NODE_UNKNOWN;
380         else if (lstcon_session_match(rep->sid))
381                 nd->nd_state = LST_NODE_ACTIVE;
382         else
383                 nd->nd_state = LST_NODE_BUSY;
384
385         return 0;
386 }
387
388 void
389 lstcon_rpc_trans_stat(lstcon_rpc_trans_t *trans, lstcon_trans_stat_t *stat)
390 {
391         lstcon_rpc_t      *crpc;
392         srpc_client_rpc_t *rpc;
393         srpc_msg_t        *rep;
394         int                error;
395
396         LASSERT (stat != NULL);
397
398         memset(stat, 0, sizeof(*stat));
399
400         list_for_each_entry(crpc, &trans->tas_rpcs_list, crp_link) {
401                 lstcon_rpc_stat_total(stat, 1);
402
403                 rpc = crpc->crp_rpc;
404
405                 LASSERT (crpc->crp_stamp != 0);
406
407                 error = lstcon_rpc_get_reply(crpc, &rep);
408                 if (error != 0) {
409                         lstcon_rpc_stat_failure(stat, 1);
410                         if (stat->trs_rpc_errno == 0)
411                                 stat->trs_rpc_errno = -error;
412
413                         continue;
414                 }
415
416                 lstcon_rpc_stat_success(stat, 1);
417
418                 lstcon_rpc_stat_reply(trans->tas_opc, rep,
419                                       crpc->crp_node, stat);
420         }
421
422         CDEBUG(D_NET, "transaction %s success, %d failure, %d total %d, "
423                       "RPC error(%d), Framework error(%d)\n",
424                lstcon_rpc_trans_name(trans->tas_opc),
425                lstcon_rpc_stat_success(stat, 0),
426                lstcon_rpc_stat_failure(stat, 0),
427                lstcon_rpc_stat_total(stat, 0),
428                stat->trs_rpc_errno, stat->trs_fwk_errno);
429
430         return;
431 }
432
433 int
434 lstcon_rpc_trans_interpreter(lstcon_rpc_trans_t *trans,
435                              struct list_head *head_up,
436                              lstcon_rpc_readent_func_t readent)
437 {
438         struct list_head      tmp;
439         struct list_head     *next;
440         lstcon_rpc_ent_t     *ent;
441         srpc_generic_reply_t *rep;
442         srpc_client_rpc_t    *rpc;
443         lstcon_rpc_t         *crpc;
444         srpc_msg_t           *msg;
445         lstcon_node_t        *nd;
446         cfs_duration_t        dur;
447         struct timeval        tv;
448         int                   error;
449
450         LASSERT (head_up != NULL);
451
452         next = head_up;
453
454         list_for_each_entry(crpc, &trans->tas_rpcs_list, crp_link) {
455                 if (copy_from_user(&tmp, next, sizeof(struct list_head)))
456                         return -EFAULT;
457
458                 if (tmp.next == head_up)
459                         return 0;
460
461                 next = tmp.next;
462
463                 ent = list_entry(next, lstcon_rpc_ent_t, rpe_link);
464
465                 rpc = crpc->crp_rpc;
466
467                 LASSERT (crpc->crp_stamp != 0);
468
469                 error = lstcon_rpc_get_reply(crpc, &msg);
470
471                 nd = crpc->crp_node;
472
473                 dur = cfs_time_sub(crpc->crp_stamp,
474                                    console_session.ses_id.ses_stamp);
475                 cfs_duration_usec(dur, &tv);
476
477                 if (copy_to_user(&ent->rpe_peer,
478                                  &nd->nd_id, sizeof(lnet_process_id_t)) ||
479                     copy_to_user(&ent->rpe_stamp, &tv, sizeof(tv)) ||
480                     copy_to_user(&ent->rpe_state,
481                                  &nd->nd_state, sizeof(nd->nd_state)) ||
482                     copy_to_user(&ent->rpe_rpc_errno, &error, sizeof(error)))
483                         return -EFAULT;
484
485                 if (error != 0)
486                         continue;
487
488                 /* RPC is done */
489                 rep = (srpc_generic_reply_t *)&msg->msg_body.reply;
490
491                 if (copy_to_user(&ent->rpe_sid,
492                                  &rep->sid, sizeof(lst_sid_t)) ||
493                     copy_to_user(&ent->rpe_fwk_errno,
494                                  &rep->status, sizeof(rep->status)))
495                         return -EFAULT;
496
497                 if (readent == NULL)
498                         continue;
499
500                 if ((error = readent(trans->tas_opc, msg, ent)) != 0)
501                         return error;
502         }
503
504         return 0;
505 }
506
507 void
508 lstcon_rpc_trans_destroy(lstcon_rpc_trans_t *trans)
509 {
510         srpc_client_rpc_t *rpc;
511         lstcon_rpc_t      *crpc;
512         lstcon_rpc_t      *tmp;
513         int                count = 0;
514         
515         list_for_each_entry_safe(crpc, tmp,
516                                  &trans->tas_rpcs_list, crp_link) {
517                 rpc = crpc->crp_rpc;
518
519                 spin_lock(&rpc->crpc_lock);
520
521                 /* free it if not posted or finished already */
522                 if (!crpc->crp_posted || crpc->crp_finished) {
523                         spin_unlock(&rpc->crpc_lock);
524
525                         list_del_init(&crpc->crp_link);
526                         lstcon_rpc_put(crpc);
527
528                         continue;
529                 }
530
531                 /* rpcs can be still not callbacked (even LNetMDUnlink is called)
532                  * because huge timeout for inaccessible network, don't make
533                  * user wait for them, just put rpcs in orphan list */
534
535                 LASSERT (crpc->crp_status != 0);
536
537                 crpc->crp_node  = NULL;
538                 crpc->crp_trans = NULL;
539                 list_del(&crpc->crp_link);
540
541                 spin_lock(&console_session.ses_rpc_lock);
542
543                 count ++;
544                 /* add to orphan list */
545                 console_session.ses_rpc_pending ++;
546                 list_add_tail(&crpc->crp_link, &console_session.ses_rpc_list);
547
548                 spin_unlock(&console_session.ses_rpc_lock);
549
550                 spin_unlock(&rpc->crpc_lock);
551
552                 atomic_dec(&trans->tas_remaining);
553         }
554
555         LASSERT (atomic_read(&trans->tas_remaining) == 0);
556
557         list_del(&trans->tas_link);
558         if (!list_empty(&trans->tas_olink))
559                 list_del(&trans->tas_olink);
560
561         CDEBUG(D_NET, "Transaction %s destroyed with %d pending RPCs\n",
562                lstcon_rpc_trans_name(trans->tas_opc), count);
563
564         LIBCFS_FREE(trans, sizeof(*trans));
565
566         return;
567 }
568
569 int
570 lstcon_sesrpc_prep(lstcon_node_t *nd, int transop, lstcon_rpc_t **crpc)
571 {
572         srpc_mksn_reqst_t *msrq;
573         srpc_rmsn_reqst_t *rsrq;
574         int                rc;
575
576         switch (transop) {
577         case LST_TRANS_SESNEW:
578                 rc = lstcon_rpc_prep(nd, SRPC_SERVICE_MAKE_SESSION, 0, crpc);
579                 if (rc != 0)
580                         return rc;
581
582                 msrq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.mksn_reqst;
583                 msrq->mksn_sid     = console_session.ses_id;
584                 msrq->mksn_force   = console_session.ses_force;
585                 strncpy(msrq->mksn_name, console_session.ses_name,
586                         strlen(console_session.ses_name));
587                 break;
588
589         case LST_TRANS_SESEND:
590                 rc = lstcon_rpc_prep(nd, SRPC_SERVICE_REMOVE_SESSION, 0, crpc);
591                 if (rc != 0)
592                         return rc;
593
594                 rsrq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.rmsn_reqst;
595                 rsrq->rmsn_sid = console_session.ses_id;
596                 break;
597
598         default:
599                 LBUG();
600         }
601
602         return 0;
603 }
604
605 int
606 lstcon_dbgrpc_prep(lstcon_node_t *nd, lstcon_rpc_t **crpc)
607 {
608         srpc_debug_reqst_t *drq;
609         int                 rc;
610
611         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_DEBUG, 0, crpc);
612         if (rc != 0)
613                 return rc;
614
615         drq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.dbg_reqst;
616
617         drq->dbg_sid   = console_session.ses_id;
618         drq->dbg_flags = 0;
619         
620         return rc;
621 }
622
623 int
624 lstcon_batrpc_prep(lstcon_node_t *nd, int transop,
625                    lstcon_tsb_hdr_t *tsb, lstcon_rpc_t **crpc)
626 {
627         lstcon_batch_t     *batch;
628         srpc_batch_reqst_t *brq;
629         int                 rc;
630
631         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_BATCH, 0, crpc);
632         if (rc != 0)
633                 return rc;
634
635         brq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.bat_reqst;
636
637         brq->bar_sid     = console_session.ses_id;
638         brq->bar_bid     = tsb->tsb_id;
639         brq->bar_testidx = tsb->tsb_index;
640         brq->bar_opc     = transop == LST_TRANS_TSBRUN ? SRPC_BATCH_OPC_RUN :
641                            (transop == LST_TRANS_TSBSTOP ? SRPC_BATCH_OPC_STOP:
642                             SRPC_BATCH_OPC_QUERY);
643
644         if (transop != LST_TRANS_TSBRUN &&
645             transop != LST_TRANS_TSBSTOP)
646                 return 0;
647
648         LASSERT (tsb->tsb_index == 0);
649
650         batch = (lstcon_batch_t *)tsb;
651         brq->bar_arg = batch->bat_arg;
652         
653         return 0;
654 }
655
656 int
657 lstcon_statrpc_prep(lstcon_node_t *nd, lstcon_rpc_t **crpc)
658 {
659         srpc_stat_reqst_t *srq;
660         int                rc;
661
662         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_QUERY_STAT, 0, crpc);
663         if (rc != 0)
664                 return rc;
665
666         srq = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.stat_reqst;
667
668         srq->str_sid  = console_session.ses_id;
669         srq->str_type = 0; /* XXX remove it */
670
671         return 0;
672 }
673
674 lnet_process_id_t *
675 lstcon_next_id(int idx, int nkiov, lnet_kiov_t *kiov)
676 {
677         lnet_process_id_t *pid;
678         int                i;
679
680         i = idx / (CFS_PAGE_SIZE / sizeof(lnet_process_id_t));
681         
682         LASSERT (i < nkiov);
683
684         pid = (lnet_process_id_t *)cfs_page_address(kiov[i].kiov_page);
685
686         return &pid[idx % (CFS_PAGE_SIZE / sizeof(lnet_process_id_t))];
687 }
688
689 int
690 lstcon_dstnodes_prep(lstcon_group_t *grp, int idx,
691                      int dist, int span, int nkiov, lnet_kiov_t *kiov)
692 {
693         lnet_process_id_t *pid;
694         lstcon_ndlink_t   *ndl;
695         lstcon_node_t     *nd;
696         int                start;
697         int                end;
698         int                i = 0;
699
700         LASSERT (dist >= 1);
701         LASSERT (span >= 1);
702         LASSERT (grp->grp_nnode >= 1);
703
704         if (span > grp->grp_nnode)
705                 return -EINVAL;
706
707         start = ((idx / dist) * span) % grp->grp_nnode;
708         end   = ((idx / dist) * span + span - 1) % grp->grp_nnode;
709
710         list_for_each_entry(ndl, &grp->grp_ndl_list, ndl_link) {
711                 nd = ndl->ndl_node;
712                 if (i < start) {
713                         i ++;
714                         continue;
715                 }
716
717                 if (i > (end >= start ? end: grp->grp_nnode))
718                         break;
719
720                 pid = lstcon_next_id((i - start), nkiov, kiov);
721                 *pid = nd->nd_id;
722                 i++;
723         }
724
725         if (start <= end) /* done */
726                 return 0;
727
728         list_for_each_entry(ndl, &grp->grp_ndl_list, ndl_link) {
729                 if (i > grp->grp_nnode + end)
730                         break;
731
732                 nd = ndl->ndl_node;
733                 pid = lstcon_next_id((i - start), nkiov, kiov);
734                 *pid = nd->nd_id;
735                 i++;
736         }
737
738         return 0;
739 }
740
741 int
742 lstcon_pingrpc_prep(lst_test_ping_param_t *param, srpc_test_reqst_t *req)
743 {
744         test_ping_req_t *prq = &req->tsr_u.ping;
745         
746         prq->png_size   = param->png_size;
747         prq->png_flags  = param->png_flags;
748         /* TODO dest */
749         return 0;
750 }
751
752 int
753 lstcon_bulkrpc_prep(lst_test_bulk_param_t *param, srpc_test_reqst_t *req)
754 {
755         test_bulk_req_t *brq = &req->tsr_u.bulk;
756
757         brq->blk_opc    = param->blk_opc;
758         brq->blk_npg    = (param->blk_size + CFS_PAGE_SIZE - 1) / CFS_PAGE_SIZE;
759         brq->blk_flags  = param->blk_flags;
760
761         return 0;
762 }
763
764 int
765 lstcon_testrpc_prep(lstcon_node_t *nd, int transop,
766                     lstcon_test_t *test, lstcon_rpc_t **crpc)
767 {
768         lstcon_group_t    *sgrp = test->tes_src_grp;
769         lstcon_group_t    *dgrp = test->tes_dst_grp;
770         srpc_test_reqst_t *trq;
771         srpc_bulk_t       *bulk;
772         int                i;
773         int                n  = 0;
774         int                rc = 0;
775
776         if (transop == LST_TRANS_TSBCLIADD)
777                 n = sfw_id_pages(test->tes_span);
778
779         rc = lstcon_rpc_prep(nd, SRPC_SERVICE_TEST, n, crpc);
780         if (rc != 0) 
781                 return rc;
782
783         trq  = &(*crpc)->crp_rpc->crpc_reqstmsg.msg_body.tes_reqst;
784
785         if (transop == LST_TRANS_TSBSRVADD) {
786                 int ndist = (sgrp->grp_nnode + test->tes_dist - 1) / test->tes_dist;
787                 int nspan = (dgrp->grp_nnode + test->tes_span - 1) / test->tes_span;
788                 int nmax = (ndist + nspan - 1) / nspan;
789
790                 trq->tsr_ndest = 0;
791                 trq->tsr_loop  = nmax * test->tes_dist * test->tes_concur;
792
793         } else {
794                 bulk = &(*crpc)->crp_rpc->crpc_bulk;
795
796                 for (i = 0; i < n; i++) {
797                         bulk->bk_iovs[i].kiov_offset = 0;
798                         bulk->bk_iovs[i].kiov_len    = CFS_PAGE_SIZE;
799                         bulk->bk_iovs[i].kiov_page   = cfs_alloc_page(CFS_ALLOC_STD);
800
801                         if (bulk->bk_iovs[i].kiov_page != NULL) 
802                                 continue;
803
804                         lstcon_rpc_put(*crpc);
805                         return -ENOMEM;
806                 }
807
808                 bulk->bk_sink = 0;
809
810                 LASSERT (transop == LST_TRANS_TSBCLIADD);
811
812                 rc = lstcon_dstnodes_prep(test->tes_dst_grp,
813                                           test->tes_cliidx++, test->tes_dist,
814                                           test->tes_span, n, &bulk->bk_iovs[0]);
815                 if (rc != 0) {
816                         lstcon_rpc_put(*crpc);
817                         return rc;
818                 }
819
820                 trq->tsr_ndest = test->tes_span;
821                 trq->tsr_loop  = test->tes_loop;
822         } 
823
824         trq->tsr_sid        = console_session.ses_id;
825         trq->tsr_bid        = test->tes_hdr.tsb_id;
826         trq->tsr_concur     = test->tes_concur;
827         trq->tsr_is_client  = (transop == LST_TRANS_TSBCLIADD) ? 1 : 0;
828         trq->tsr_stop_onerr = test->tes_stop_onerr;
829
830         switch (test->tes_type) {
831         case LST_TEST_PING:
832                 trq->tsr_service = SRPC_SERVICE_PING;
833                 rc = lstcon_pingrpc_prep((lst_test_ping_param_t *)&test->tes_param[0], trq);
834                 break;
835         case LST_TEST_BULK:
836                 trq->tsr_service = SRPC_SERVICE_BRW;
837                 rc = lstcon_bulkrpc_prep((lst_test_bulk_param_t *)&test->tes_param[0], trq);
838                 break;
839         default:
840                 LBUG();
841                 break;
842         }
843
844         return rc;
845 }
846
847 void
848 lstcon_rpc_stat_reply(int transop, srpc_msg_t *msg,
849                       lstcon_node_t *nd, lstcon_trans_stat_t *stat)
850 {
851         srpc_mksn_reply_t  *mksn_rep;
852         srpc_rmsn_reply_t  *rmsn_rep;
853         srpc_debug_reply_t *dbg_rep;
854         srpc_batch_reply_t *bat_rep;
855         srpc_test_reply_t  *test_rep;
856         srpc_stat_reply_t  *stat_rep;
857         int                 errno = 0;
858
859         switch (transop) {
860         case LST_TRANS_SESNEW:
861                 mksn_rep = &msg->msg_body.mksn_reply;
862
863                 if (mksn_rep->mksn_status == 0) {
864                         lstcon_sesop_stat_success(stat, 1);
865                         /* session timeout on remote node */
866                         nd->nd_timeout = mksn_rep->mksn_timeout;
867                         return;
868                 }
869
870                 LASSERT (mksn_rep->mksn_status == EBUSY ||
871                          mksn_rep->mksn_status == EINVAL);
872
873                 lstcon_sesop_stat_failure(stat, 1);
874                 errno = mksn_rep->mksn_status;
875                 break;
876
877         case LST_TRANS_SESEND:
878                 rmsn_rep = &msg->msg_body.rmsn_reply;
879                 /* ESRCH is not an error for end session */
880                 if (rmsn_rep->rmsn_status == 0 ||
881                     rmsn_rep->rmsn_status == ESRCH) {
882                         lstcon_sesop_stat_success(stat, 1);
883                         return;
884                 }
885
886                 LASSERT (rmsn_rep->rmsn_status == EBUSY ||
887                          rmsn_rep->rmsn_status == EINVAL);
888
889                 lstcon_sesop_stat_failure(stat, 1);
890                 errno = rmsn_rep->rmsn_status;
891                 break;
892
893         case LST_TRANS_SESQRY:
894         case LST_TRANS_SESPING:
895                 dbg_rep = &msg->msg_body.dbg_reply;
896
897                 if (dbg_rep->dbg_status == ESRCH) {
898                         lstcon_sesqry_stat_unknown(stat, 1);
899                         return;
900                 } 
901
902                 LASSERT (dbg_rep->dbg_status == 0);
903
904                 if (lstcon_session_match(dbg_rep->dbg_sid))
905                         lstcon_sesqry_stat_active(stat, 1);
906                 else
907                         lstcon_sesqry_stat_busy(stat, 1);
908                 return;
909
910         case LST_TRANS_TSBRUN:
911         case LST_TRANS_TSBSTOP:
912                 bat_rep = &msg->msg_body.bat_reply;
913
914                 if (bat_rep->bar_status == 0) {
915                         lstcon_tsbop_stat_success(stat, 1);
916                         return;
917                 }
918
919                 if (bat_rep->bar_status == EPERM && 
920                     transop == LST_TRANS_TSBSTOP) {
921                         lstcon_tsbop_stat_success(stat, 1);
922                         return;
923                 }
924
925                 lstcon_tsbop_stat_failure(stat, 1);
926                 errno = bat_rep->bar_status;
927                 break;
928
929         case LST_TRANS_TSBCLIQRY:
930         case LST_TRANS_TSBSRVQRY:
931                 bat_rep = &msg->msg_body.bat_reply;
932
933                 if (bat_rep->bar_active != 0) 
934                         lstcon_tsbqry_stat_run(stat, 1);
935                 else
936                         lstcon_tsbqry_stat_idle(stat, 1);
937
938                 if (bat_rep->bar_status == 0) 
939                         return;
940
941                 lstcon_tsbqry_stat_failure(stat, 1);
942                 errno = bat_rep->bar_status;
943                 break;
944
945         case LST_TRANS_TSBCLIADD:
946         case LST_TRANS_TSBSRVADD:
947                 test_rep = &msg->msg_body.tes_reply;
948
949                 if (test_rep->tsr_status == 0) {
950                         lstcon_tsbop_stat_success(stat, 1);
951                         return;
952                 }
953
954                 lstcon_tsbop_stat_failure(stat, 1);
955                 errno = test_rep->tsr_status;
956                 break;
957
958         case LST_TRANS_STATQRY:
959                 stat_rep = &msg->msg_body.stat_reply;
960
961                 if (stat_rep->str_status == 0) {
962                         lstcon_statqry_stat_success(stat, 1);
963                         return;
964                 }
965
966                 lstcon_statqry_stat_failure(stat, 1);
967                 errno = stat_rep->str_status;
968                 break;
969
970         default:
971                 LBUG();
972         }
973
974         if (stat->trs_fwk_errno == 0)
975                 stat->trs_fwk_errno = errno;
976
977         return;
978 }
979
980 int
981 lstcon_rpc_trans_ndlist(struct list_head *ndlist,
982                         struct list_head *translist, int transop, void *arg,
983                         lstcon_rpc_cond_func_t condition, lstcon_rpc_trans_t **transpp)
984 {
985         lstcon_rpc_trans_t *trans;
986         lstcon_ndlink_t    *ndl;
987         lstcon_node_t      *nd;
988         lstcon_rpc_t       *rpc;
989         int                 rc;
990
991         /* Creating session RPG for list of nodes */
992
993         rc = lstcon_rpc_trans_prep(translist, transop, &trans);
994         if (rc != 0) {
995                 CERROR("Can't create transaction %d: %d\n", transop, rc);
996                 return rc;
997         }
998
999         list_for_each_entry(ndl, ndlist, ndl_link) {
1000                 rc = condition == NULL ? 1 :
1001                      condition(transop, ndl->ndl_node, arg);
1002
1003                 if (rc == 0)
1004                         continue;
1005
1006                 if (rc < 0) {
1007                         CDEBUG(D_NET, "Condition error while creating RPC "
1008                                       " for transaction %d: %d\n", transop, rc);
1009                         break;
1010                 }
1011
1012                 nd = ndl->ndl_node;
1013
1014                 switch (transop) {
1015                 case LST_TRANS_SESNEW:
1016                 case LST_TRANS_SESEND:
1017                         rc = lstcon_sesrpc_prep(nd, transop, &rpc);
1018                         break;
1019                 case LST_TRANS_SESQRY:
1020                 case LST_TRANS_SESPING:
1021                         rc = lstcon_dbgrpc_prep(nd, &rpc);
1022                         break;
1023                 case LST_TRANS_TSBCLIADD:
1024                 case LST_TRANS_TSBSRVADD:
1025                         rc = lstcon_testrpc_prep(nd, transop,
1026                                                  (lstcon_test_t *)arg, &rpc);
1027                         break;
1028                 case LST_TRANS_TSBRUN:
1029                 case LST_TRANS_TSBSTOP:
1030                 case LST_TRANS_TSBCLIQRY:
1031                 case LST_TRANS_TSBSRVQRY:
1032                         rc = lstcon_batrpc_prep(nd, transop,
1033                                                 (lstcon_tsb_hdr_t *)arg, &rpc);
1034                         break;
1035                 case LST_TRANS_STATQRY:
1036                         rc = lstcon_statrpc_prep(nd, &rpc);
1037                         break;
1038                 default:
1039                         rc = -EINVAL;
1040                         break;
1041                 }
1042
1043                 if (rc != 0) {
1044                         CERROR("Failed to create RPC for transaction %s: %d\n",
1045                                lstcon_rpc_trans_name(transop), rc);
1046                         break;
1047                 }
1048                                 
1049                 lstcon_rpc_trans_addreq(trans, rpc);
1050         }
1051
1052         if (rc == 0) {
1053                 *transpp = trans;
1054                 return 0;
1055         }
1056
1057         lstcon_rpc_trans_destroy(trans);
1058
1059         return rc;
1060 }
1061
1062 void
1063 lstcon_rpc_pinger(void *arg)
1064 {
1065         stt_timer_t        *ptimer = (stt_timer_t *)arg;
1066         lstcon_rpc_trans_t *trans;
1067         lstcon_rpc_t       *crpc;
1068         srpc_msg_t         *rep;
1069         srpc_debug_reqst_t *drq;
1070         lstcon_ndlink_t    *ndl;
1071         lstcon_node_t      *nd;
1072         time_t              intv;
1073         int                 count = 0;
1074         int                 rc;
1075
1076         /* RPC pinger is a special case of transaction,
1077          * it's called by timer at 8 seconds interval.
1078          */
1079         mutex_down(&console_session.ses_mutex);
1080
1081         if (console_session.ses_shutdown || console_session.ses_expired) {
1082                 mutex_up(&console_session.ses_mutex);
1083                 return;
1084         }
1085
1086         if (!console_session.ses_expired &&
1087             cfs_time_current_sec() - console_session.ses_laststamp >
1088             console_session.ses_timeout)
1089                 console_session.ses_expired = 1;
1090
1091         trans = console_session.ses_ping;
1092
1093         LASSERT (trans != NULL);
1094
1095         list_for_each_entry(ndl, &console_session.ses_ndl_list, ndl_link) {
1096                 nd = ndl->ndl_node;
1097
1098                 if (console_session.ses_expired) {
1099                         /* idle console, end session on all nodes */
1100                         if (nd->nd_state != LST_NODE_ACTIVE)
1101                                 continue;
1102
1103                         rc = lstcon_sesrpc_prep(nd, LST_TRANS_SESEND, &crpc);
1104                         if (rc != 0) {
1105                                 CERROR("Out of memory\n");
1106                                 break;
1107                         }
1108
1109                         lstcon_rpc_trans_addreq(trans, crpc);
1110                         lstcon_rpc_post(crpc);
1111
1112                         continue;
1113                 }
1114
1115                 crpc = &nd->nd_ping;
1116
1117                 if (crpc->crp_rpc != NULL) {
1118                         LASSERT (crpc->crp_trans == trans);
1119                         LASSERT (!list_empty(&crpc->crp_link));
1120
1121                         spin_lock(&crpc->crp_rpc->crpc_lock);
1122
1123                         LASSERT (crpc->crp_posted);
1124
1125                         if (!crpc->crp_finished) {
1126                                 /* in flight */
1127                                 spin_unlock(&crpc->crp_rpc->crpc_lock);
1128                                 continue;
1129                         }
1130
1131                         spin_unlock(&crpc->crp_rpc->crpc_lock);
1132
1133                         lstcon_rpc_get_reply(crpc, &rep);
1134
1135                         list_del_init(&crpc->crp_link);
1136                 
1137                         lstcon_rpc_put(crpc);
1138                 }
1139
1140                 if (nd->nd_state != LST_NODE_ACTIVE)
1141                         continue;
1142
1143                 intv = cfs_duration_sec(cfs_time_sub(cfs_time_current(),
1144                                                      nd->nd_stamp));
1145                 if (intv < nd->nd_timeout / 2)
1146                         continue;
1147
1148                 rc = lstcon_rpc_init(nd, SRPC_SERVICE_DEBUG, 0, 0, crpc);
1149                 if (rc != 0) {
1150                         CERROR("Out of memory\n");
1151                         break;
1152                 }
1153
1154                 drq = &crpc->crp_rpc->crpc_reqstmsg.msg_body.dbg_reqst;
1155
1156                 drq->dbg_sid   = console_session.ses_id;
1157                 drq->dbg_flags = 0;
1158
1159                 lstcon_rpc_trans_addreq(trans, crpc);
1160                 lstcon_rpc_post(crpc);
1161
1162                 count ++;
1163         }
1164
1165         if (console_session.ses_expired) {
1166                 mutex_up(&console_session.ses_mutex);
1167                 return;
1168         }
1169
1170         CDEBUG(D_NET, "Ping %d nodes in session\n", count);
1171
1172         ptimer->stt_expires = cfs_time_current_sec() + LST_PING_INTERVAL;
1173         stt_add_timer(ptimer);
1174
1175         mutex_up(&console_session.ses_mutex);
1176 }
1177
1178 int
1179 lstcon_rpc_pinger_start(void)
1180 {
1181         stt_timer_t    *ptimer;
1182         int             rc;
1183
1184         LASSERT (console_session.ses_rpc_pending == 0);
1185         LASSERT (list_empty(&console_session.ses_rpc_list));
1186         LASSERT (list_empty(&console_session.ses_rpc_freelist));
1187
1188         rc = lstcon_rpc_trans_prep(NULL, LST_TRANS_SESPING,
1189                                    &console_session.ses_ping);
1190         if (rc != 0) {
1191                 CERROR("Failed to create console pinger\n");
1192                 return rc;
1193         }
1194
1195         ptimer = &console_session.ses_ping_timer;
1196         ptimer->stt_expires = cfs_time_current_sec() + LST_PING_INTERVAL;
1197
1198         stt_add_timer(ptimer);
1199
1200         return 0;
1201 }
1202
1203 void
1204 lstcon_rpc_pinger_stop(void)
1205 {
1206         LASSERT (console_session.ses_shutdown);
1207
1208         stt_del_timer(&console_session.ses_ping_timer);
1209
1210         lstcon_rpc_trans_abort(console_session.ses_ping, -ESHUTDOWN);
1211         lstcon_rpc_trans_stat(console_session.ses_ping, lstcon_trans_stat());
1212         lstcon_rpc_trans_destroy(console_session.ses_ping);
1213
1214         memset(lstcon_trans_stat(), 0, sizeof(lstcon_trans_stat_t));
1215
1216         console_session.ses_ping = NULL;
1217 }
1218
1219 void
1220 lstcon_rpc_cleanup_wait(void)
1221 {
1222         lstcon_rpc_trans_t *trans;
1223         lstcon_rpc_t       *crpc;
1224         struct list_head   *pacer;
1225         struct list_head    zlist;
1226
1227         LASSERT (console_session.ses_shutdown);
1228
1229         while (!list_empty(&console_session.ses_trans_list)) { 
1230                 list_for_each(pacer, &console_session.ses_trans_list) {
1231                         trans = list_entry(pacer, lstcon_rpc_trans_t, tas_link);
1232                         cfs_waitq_signal(&trans->tas_waitq);
1233
1234                         CDEBUG(D_NET, "Session closed, wakeup transaction %s\n",
1235                                lstcon_rpc_trans_name(trans->tas_opc));
1236                 }
1237
1238                 mutex_up(&console_session.ses_mutex);
1239
1240                 CWARN("Session is shutting down, close all transactions\n");
1241                 cfs_pause(cfs_time_seconds(1));
1242
1243                 mutex_down(&console_session.ses_mutex);
1244         }
1245
1246         spin_lock(&console_session.ses_rpc_lock);
1247
1248         lst_wait_until(list_empty(&console_session.ses_rpc_list),
1249                        console_session.ses_rpc_lock,
1250                        "Network is not accessable or target is down, "
1251                        "waiting for %d console rpcs to die\n",
1252                        console_session.ses_rpc_pending);
1253
1254         list_add(&zlist, &console_session.ses_rpc_freelist);
1255         list_del_init(&console_session.ses_rpc_freelist);
1256
1257         spin_unlock(&console_session.ses_rpc_lock);
1258
1259         LASSERT (console_session.ses_rpc_pending == 0);
1260
1261         while (!list_empty(&zlist)) {
1262                 crpc = list_entry(zlist.next, lstcon_rpc_t, crp_link);
1263
1264                 list_del(&crpc->crp_link);
1265                 LIBCFS_FREE(crpc, sizeof(lstcon_rpc_t));
1266         }
1267 }
1268
1269 int
1270 lstcon_rpc_module_init(void)
1271 {
1272         CFS_INIT_LIST_HEAD(&console_session.ses_ping_timer.stt_list);
1273         console_session.ses_ping_timer.stt_func = lstcon_rpc_pinger;
1274         console_session.ses_ping_timer.stt_data = &console_session.ses_ping_timer;
1275
1276         console_session.ses_ping = NULL;
1277         console_session.ses_rpc_pending = 0;
1278         spin_lock_init(&console_session.ses_rpc_lock);
1279         CFS_INIT_LIST_HEAD(&console_session.ses_rpc_list);
1280         CFS_INIT_LIST_HEAD(&console_session.ses_rpc_freelist);
1281
1282         return 0;
1283 }
1284
1285 void
1286 lstcon_rpc_module_fini(void)
1287 {
1288         LASSERT (console_session.ses_rpc_pending == 0);
1289         LASSERT (list_empty(&console_session.ses_rpc_list));
1290         LASSERT (list_empty(&console_session.ses_rpc_freelist));
1291 }
1292
1293 #endif