1 /*
2 * Copyright (c) 2006-2019 Red Hat, Inc.
3 *
4 * All rights reserved.
5 *
6 * Author: Christine Caulfield (ccaulfie@redhat.com)
7 * Author: Jan Friesse (jfriesse@redhat.com)
8 *
9 * This software licensed under BSD license, the text of which follows:
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions are met:
13 *
14 * - Redistributions of source code must retain the above copyright notice,
15 * this list of conditions and the following disclaimer.
16 * - Redistributions in binary form must reproduce the above copyright notice,
17 * this list of conditions and the following disclaimer in the documentation
18 * and/or other materials provided with the distribution.
19 * - Neither the name of the MontaVista Software, Inc. nor the names of its
20 * contributors may be used to endorse or promote products derived from this
21 * software without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTIBUTORS "AS IS"
24 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
27 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
33 * THE POSSIBILITY OF SUCH DAMAGE.
34 */
35
36 #include <config.h>
37
38 #ifdef HAVE_ALLOCA_H
39 #include <alloca.h>
40 #endif
41 #include <sys/types.h>
42 #include <sys/socket.h>
43 #include <sys/un.h>
44 #include <sys/ioctl.h>
45 #include <netinet/in.h>
46 #include <sys/uio.h>
47 #include <unistd.h>
48 #include <fcntl.h>
49 #include <stdlib.h>
50 #include <stdio.h>
51 #include <errno.h>
52 #include <time.h>
53 #include <assert.h>
54 #include <arpa/inet.h>
55 #include <sys/mman.h>
56
57 #include <qb/qblist.h>
58 #include <qb/qbmap.h>
59
60 #include <corosync/corotypes.h>
61 #include <qb/qbipc_common.h>
62 #include <corosync/corodefs.h>
63 #include <corosync/logsys.h>
64 #include <corosync/coroapi.h>
65
66 #include <corosync/cpg.h>
67 #include <corosync/ipc_cpg.h>
68
69 #ifndef MAP_ANONYMOUS
70 #define MAP_ANONYMOUS MAP_ANON
71 #endif
72
73 #include "service.h"
74
75 LOGSYS_DECLARE_SUBSYS ("CPG");
76
77 #define GROUP_HASH_SIZE 32
78
79 enum cpg_message_req_types {
80 MESSAGE_REQ_EXEC_CPG_PROCJOIN = 0,
81 MESSAGE_REQ_EXEC_CPG_PROCLEAVE = 1,
82 MESSAGE_REQ_EXEC_CPG_JOINLIST = 2,
83 MESSAGE_REQ_EXEC_CPG_MCAST = 3,
84 MESSAGE_REQ_EXEC_CPG_DOWNLIST_OLD = 4,
85 MESSAGE_REQ_EXEC_CPG_DOWNLIST = 5,
86 MESSAGE_REQ_EXEC_CPG_PARTIAL_MCAST = 6,
87 };
88
89 struct zcb_mapped {
90 struct qb_list_head list;
91 void *addr;
92 size_t size;
93 };
94 /*
95 * state` exec deliver
96 * match group name, pid -> if matched deliver for YES:
97 * XXX indicates impossible state
98 *
99 * join leave mcast
100 * UNJOINED XXX XXX NO
101 * LEAVE_STARTED XXX YES(unjoined_enter) YES
102 * JOIN_STARTED YES(join_started_enter) XXX NO
103 * JOIN_COMPLETED XXX NO YES
104 *
105 * join_started_enter
106 * set JOIN_COMPLETED
107 * add entry to process_info list
108 * unjoined_enter
109 * set UNJOINED
110 * delete entry from process_info list
111 *
112 *
113 * library accept join error codes
114 * UNJOINED YES(CS_OK) set JOIN_STARTED
115 * LEAVE_STARTED NO(CS_ERR_BUSY)
116 * JOIN_STARTED NO(CS_ERR_EXIST)
117 * JOIN_COMPlETED NO(CS_ERR_EXIST)
118 *
119 * library accept leave error codes
120 * UNJOINED NO(CS_ERR_NOT_EXIST)
121 * LEAVE_STARTED NO(CS_ERR_NOT_EXIST)
122 * JOIN_STARTED NO(CS_ERR_BUSY)
123 * JOIN_COMPLETED YES(CS_OK) set LEAVE_STARTED
124 *
125 * library accept mcast
126 * UNJOINED NO(CS_ERR_NOT_EXIST)
127 * LEAVE_STARTED NO(CS_ERR_NOT_EXIST)
128 * JOIN_STARTED YES(CS_OK)
129 * JOIN_COMPLETED YES(CS_OK)
130 */
131 enum cpd_state {
132 CPD_STATE_UNJOINED,
133 CPD_STATE_LEAVE_STARTED,
134 CPD_STATE_JOIN_STARTED,
135 CPD_STATE_JOIN_COMPLETED
136 };
137
138 enum cpg_sync_state {
139 CPGSYNC_DOWNLIST,
140 CPGSYNC_JOINLIST
141 };
142
143 static struct qb_list_head joinlist_messages_head;
144
145 struct cpg_pd {
146 void *conn;
147 mar_cpg_name_t group_name;
148 uint32_t pid;
149 enum cpd_state cpd_state;
150 unsigned int flags;
151 int initial_totem_conf_sent;
152 uint64_t transition_counter; /* These two are used when sending fragmented messages */
153 uint64_t initial_transition_counter;
154 struct qb_list_head list;
155 struct qb_list_head iteration_instance_list_head;
156 struct qb_list_head zcb_mapped_list_head;
157 };
158
159 struct cpg_iteration_instance {
160 hdb_handle_t handle;
161 struct qb_list_head list;
162 struct qb_list_head items_list_head; /* List of process_info */
163 struct qb_list_head *current_pointer;
164 };
165
166 DECLARE_HDB_DATABASE(cpg_iteration_handle_t_db,NULL);
167
168 QB_LIST_DECLARE (cpg_pd_list_head);
169
170 static unsigned int my_member_list[PROCESSOR_COUNT_MAX];
171
172 static unsigned int my_member_list_entries;
173
174 static unsigned int my_old_member_list[PROCESSOR_COUNT_MAX];
175
176 static unsigned int my_old_member_list_entries = 0;
177
178 static struct corosync_api_v1 *api = NULL;
179
180 static enum cpg_sync_state my_sync_state = CPGSYNC_DOWNLIST;
181
182 static mar_cpg_ring_id_t last_sync_ring_id;
183
184 struct process_info {
185 unsigned int nodeid;
186 uint32_t pid;
187 mar_cpg_name_t group;
188 struct qb_list_head list; /* on the group_info members list */
189 };
190 QB_LIST_DECLARE (process_info_list_head);
191
192 struct join_list_entry {
193 uint32_t pid;
194 mar_cpg_name_t group_name;
195 };
196
197 struct join_list_confchg_data {
198 mar_cpg_name_t cpg_group;
199 mar_cpg_address_t join_list[CPG_MEMBERS_MAX];
200 int join_list_entries;
201 };
202
203 /*
204 * Service Interfaces required by service_message_handler struct
205 */
206 static char *cpg_exec_init_fn (struct corosync_api_v1 *);
207
208 static int cpg_lib_init_fn (void *conn);
209
210 static int cpg_lib_exit_fn (void *conn);
211
212 static void message_handler_req_exec_cpg_procjoin (
213 const void *message,
214 unsigned int nodeid);
215
216 static void message_handler_req_exec_cpg_procleave (
217 const void *message,
218 unsigned int nodeid);
219
220 static void message_handler_req_exec_cpg_joinlist (
221 const void *message,
222 unsigned int nodeid);
223
224 static void message_handler_req_exec_cpg_mcast (
225 const void *message,
226 unsigned int nodeid);
227
228 static void message_handler_req_exec_cpg_partial_mcast (
229 const void *message,
230 unsigned int nodeid);
231
232 static void message_handler_req_exec_cpg_downlist_old (
233 const void *message,
234 unsigned int nodeid);
235
236 static void message_handler_req_exec_cpg_downlist (
237 const void *message,
238 unsigned int nodeid);
239
240 static void exec_cpg_procjoin_endian_convert (void *msg);
241
242 static void exec_cpg_joinlist_endian_convert (void *msg);
243
244 static void exec_cpg_mcast_endian_convert (void *msg);
245
246 static void exec_cpg_partial_mcast_endian_convert (void *msg);
247
248 static void exec_cpg_downlist_endian_convert_old (void *msg);
249
250 static void exec_cpg_downlist_endian_convert (void *msg);
251
252 static void message_handler_req_lib_cpg_join (void *conn, const void *message);
253
254 static void message_handler_req_lib_cpg_leave (void *conn, const void *message);
255
256 static void message_handler_req_lib_cpg_finalize (void *conn, const void *message);
257
258 static void message_handler_req_lib_cpg_mcast (void *conn, const void *message);
259
260 static void message_handler_req_lib_cpg_partial_mcast (void *conn, const void *message);
261
262 static void message_handler_req_lib_cpg_membership (void *conn,
263 const void *message);
264
265 static void message_handler_req_lib_cpg_local_get (void *conn,
266 const void *message);
267
268 static void message_handler_req_lib_cpg_iteration_initialize (
269 void *conn,
270 const void *message);
271
272 static void message_handler_req_lib_cpg_iteration_next (
273 void *conn,
274 const void *message);
275
276 static void message_handler_req_lib_cpg_iteration_finalize (
277 void *conn,
278 const void *message);
279
280 static void message_handler_req_lib_cpg_zc_alloc (
281 void *conn,
282 const void *message);
283
284 static void message_handler_req_lib_cpg_zc_free (
285 void *conn,
286 const void *message);
287
288 static void message_handler_req_lib_cpg_zc_execute (
289 void *conn,
290 const void *message);
291
292 static int cpg_node_joinleave_send (unsigned int pid, const mar_cpg_name_t *group_name, int fn, int reason);
293
294 static int cpg_exec_send_downlist(void);
295
296 static int cpg_exec_send_joinlist(void);
297
298 static void downlist_inform_clients (void);
299
300 static void joinlist_inform_clients (void);
301
302 static void joinlist_messages_delete (void);
303
304 static void cpg_sync_init (
305 const unsigned int *trans_list,
306 size_t trans_list_entries,
307 const unsigned int *member_list,
308 size_t member_list_entries,
309 const struct memb_ring_id *ring_id);
310
311 static int cpg_sync_process (void);
312
313 static void cpg_sync_activate (void);
314
315 static void cpg_sync_abort (void);
316
317 static void do_proc_join(
318 const mar_cpg_name_t *name,
319 uint32_t pid,
320 unsigned int nodeid,
321 int reason,
322 qb_map_t *group_notify_map);
323
324 static void do_proc_leave(
325 const mar_cpg_name_t *name,
326 uint32_t pid,
327 unsigned int nodeid,
328 int reason);
329
330 static int notify_lib_totem_membership (
331 void *conn,
332 int member_list_entries,
333 const unsigned int *member_list);
334
335 static inline int zcb_all_free (
336 struct cpg_pd *cpd);
337
338 static char *cpg_print_group_name (
339 const mar_cpg_name_t *group);
340
341 /*
342 * Library Handler Definition
343 */
344 static struct corosync_lib_handler cpg_lib_engine[] =
345 {
346 { /* 0 - MESSAGE_REQ_CPG_JOIN */
347 .lib_handler_fn = message_handler_req_lib_cpg_join,
348 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
349 },
350 { /* 1 - MESSAGE_REQ_CPG_LEAVE */
351 .lib_handler_fn = message_handler_req_lib_cpg_leave,
352 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
353 },
354 { /* 2 - MESSAGE_REQ_CPG_MCAST */
355 .lib_handler_fn = message_handler_req_lib_cpg_mcast,
356 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
357 },
358 { /* 3 - MESSAGE_REQ_CPG_MEMBERSHIP */
359 .lib_handler_fn = message_handler_req_lib_cpg_membership,
360 .flow_control = CS_LIB_FLOW_CONTROL_NOT_REQUIRED
361 },
362 { /* 4 - MESSAGE_REQ_CPG_LOCAL_GET */
363 .lib_handler_fn = message_handler_req_lib_cpg_local_get,
364 .flow_control = CS_LIB_FLOW_CONTROL_NOT_REQUIRED
365 },
366 { /* 5 - MESSAGE_REQ_CPG_ITERATIONINITIALIZE */
367 .lib_handler_fn = message_handler_req_lib_cpg_iteration_initialize,
368 .flow_control = CS_LIB_FLOW_CONTROL_NOT_REQUIRED
369 },
370 { /* 6 - MESSAGE_REQ_CPG_ITERATIONNEXT */
371 .lib_handler_fn = message_handler_req_lib_cpg_iteration_next,
372 .flow_control = CS_LIB_FLOW_CONTROL_NOT_REQUIRED
373 },
374 { /* 7 - MESSAGE_REQ_CPG_ITERATIONFINALIZE */
375 .lib_handler_fn = message_handler_req_lib_cpg_iteration_finalize,
376 .flow_control = CS_LIB_FLOW_CONTROL_NOT_REQUIRED
377 },
378 { /* 8 - MESSAGE_REQ_CPG_FINALIZE */
379 .lib_handler_fn = message_handler_req_lib_cpg_finalize,
380 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
381 },
382 { /* 9 */
383 .lib_handler_fn = message_handler_req_lib_cpg_zc_alloc,
384 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
385 },
386 { /* 10 */
387 .lib_handler_fn = message_handler_req_lib_cpg_zc_free,
388 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
389 },
390 { /* 11 */
391 .lib_handler_fn = message_handler_req_lib_cpg_zc_execute,
392 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
393 },
394 { /* 12 */
395 .lib_handler_fn = message_handler_req_lib_cpg_partial_mcast,
396 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED
397 },
398
399 };
400
401 static struct corosync_exec_handler cpg_exec_engine[] =
402 {
403 { /* 0 - MESSAGE_REQ_EXEC_CPG_PROCJOIN */
404 .exec_handler_fn = message_handler_req_exec_cpg_procjoin,
405 .exec_endian_convert_fn = exec_cpg_procjoin_endian_convert
406 },
407 { /* 1 - MESSAGE_REQ_EXEC_CPG_PROCLEAVE */
408 .exec_handler_fn = message_handler_req_exec_cpg_procleave,
409 .exec_endian_convert_fn = exec_cpg_procjoin_endian_convert
410 },
411 { /* 2 - MESSAGE_REQ_EXEC_CPG_JOINLIST */
412 .exec_handler_fn = message_handler_req_exec_cpg_joinlist,
413 .exec_endian_convert_fn = exec_cpg_joinlist_endian_convert
414 },
415 { /* 3 - MESSAGE_REQ_EXEC_CPG_MCAST */
416 .exec_handler_fn = message_handler_req_exec_cpg_mcast,
417 .exec_endian_convert_fn = exec_cpg_mcast_endian_convert
418 },
419 { /* 4 - MESSAGE_REQ_EXEC_CPG_DOWNLIST_OLD */
420 .exec_handler_fn = message_handler_req_exec_cpg_downlist_old,
421 .exec_endian_convert_fn = exec_cpg_downlist_endian_convert_old
422 },
423 { /* 5 - MESSAGE_REQ_EXEC_CPG_DOWNLIST */
424 .exec_handler_fn = message_handler_req_exec_cpg_downlist,
425 .exec_endian_convert_fn = exec_cpg_downlist_endian_convert
426 },
427 { /* 6 - MESSAGE_REQ_EXEC_CPG_PARTIAL_MCAST */
428 .exec_handler_fn = message_handler_req_exec_cpg_partial_mcast,
429 .exec_endian_convert_fn = exec_cpg_partial_mcast_endian_convert
430 },
431 };
432
433 struct corosync_service_engine cpg_service_engine = {
434 .name = "corosync cluster closed process group service v1.01",
435 .id = CPG_SERVICE,
436 .priority = 1,
437 .private_data_size = sizeof (struct cpg_pd),
438 .flow_control = CS_LIB_FLOW_CONTROL_REQUIRED,
439 .allow_inquorate = CS_LIB_ALLOW_INQUORATE,
440 .lib_init_fn = cpg_lib_init_fn,
441 .lib_exit_fn = cpg_lib_exit_fn,
442 .lib_engine = cpg_lib_engine,
443 .lib_engine_count = sizeof (cpg_lib_engine) / sizeof (struct corosync_lib_handler),
444 .exec_init_fn = cpg_exec_init_fn,
445 .exec_dump_fn = NULL,
446 .exec_engine = cpg_exec_engine,
447 .exec_engine_count = sizeof (cpg_exec_engine) / sizeof (struct corosync_exec_handler),
448 .sync_init = cpg_sync_init,
449 .sync_process = cpg_sync_process,
450 .sync_activate = cpg_sync_activate,
451 .sync_abort = cpg_sync_abort
452 };
453
454 struct corosync_service_engine *cpg_get_service_engine_ver0 (void)
455 {
456 return (&cpg_service_engine);
457 }
458
459 struct req_exec_cpg_procjoin {
460 struct qb_ipc_request_header header __attribute__((aligned(8)));
461 mar_cpg_name_t group_name __attribute__((aligned(8)));
462 mar_uint32_t pid __attribute__((aligned(8)));
463 mar_uint32_t reason __attribute__((aligned(8)));
464 };
465
466 struct req_exec_cpg_mcast {
467 struct qb_ipc_request_header header __attribute__((aligned(8)));
468 mar_cpg_name_t group_name __attribute__((aligned(8)));
469 mar_uint32_t msglen __attribute__((aligned(8)));
470 mar_uint32_t pid __attribute__((aligned(8)));
471 mar_message_source_t source __attribute__((aligned(8)));
472 mar_uint8_t message[] __attribute__((aligned(8)));
473 };
474
475 struct req_exec_cpg_partial_mcast {
476 struct qb_ipc_request_header header __attribute__((aligned(8)));
477 mar_cpg_name_t group_name __attribute__((aligned(8)));
478 mar_uint32_t msglen __attribute__((aligned(8)));
479 mar_uint32_t fraglen __attribute__((aligned(8)));
480 mar_uint32_t pid __attribute__((aligned(8)));
481 mar_uint32_t type __attribute__((aligned(8)));
482 mar_message_source_t source __attribute__((aligned(8)));
483 mar_uint8_t message[] __attribute__((aligned(8)));
484 };
485
486 struct req_exec_cpg_downlist_old {
487 struct qb_ipc_request_header header __attribute__((aligned(8)));
488 mar_uint32_t left_nodes __attribute__((aligned(8)));
489 mar_uint32_t nodeids[PROCESSOR_COUNT_MAX] __attribute__((aligned(8)));
490 };
491
492 struct req_exec_cpg_downlist {
493 struct qb_ipc_request_header header __attribute__((aligned(8)));
494 /* merge decisions */
495 mar_uint32_t old_members __attribute__((aligned(8)));
496 /* downlist below */
497 mar_uint32_t left_nodes __attribute__((aligned(8)));
498 mar_uint32_t nodeids[PROCESSOR_COUNT_MAX] __attribute__((aligned(8)));
499 };
500
501 struct joinlist_msg {
502 mar_uint32_t sender_nodeid;
503 uint32_t pid;
504 mar_cpg_name_t group_name;
505 struct qb_list_head list;
506 };
507
508 static struct req_exec_cpg_downlist g_req_exec_cpg_downlist;
509
510 /*
511 * Function print group name. It's not reentrant
512 */
513 static char *cpg_print_group_name(const mar_cpg_name_t *group)
514 {
515 static char res[CPG_MAX_NAME_LENGTH * 4 + 1];
516 int dest_pos = 0;
517 char c;
518 int i;
519
520 for (i = 0; i < group->length; i++) {
521 c = group->value[i];
522
523 if (c >= ' ' && c < 0x7f && c != '\\') {
524 res[dest_pos++] = c;
525 } else {
526 if (c == '\\') {
527 res[dest_pos++] = '\\';
528 res[dest_pos++] = '\\';
529 } else {
530 snprintf(res + dest_pos, sizeof(res) - dest_pos, "\\x%02X", c);
531 dest_pos += 4;
532 }
533 }
534 }
535 res[dest_pos] = 0;
536
537 return (res);
538 }
539
540 static void cpg_sync_init (
541 const unsigned int *trans_list,
542 size_t trans_list_entries,
543 const unsigned int *member_list,
544 size_t member_list_entries,
545 const struct memb_ring_id *ring_id)
546 {
547 int entries;
548 int i, j;
549 int found;
550
551 my_sync_state = CPGSYNC_DOWNLIST;
552
553 /*
554 * Drop joinlist messages left over from an aborted sync round. A
555 * stale message would replay a join for a node that may no longer be
556 * a member, with no leave ever following, and would also shield the
557 * entry from joinlist_remove_zombie_pi_entries(). Messages of the
558 * current round can only arrive after this init ran on all members.
559 */
560 joinlist_messages_delete ();
561
562 memcpy (my_member_list, member_list, member_list_entries *
563 sizeof (unsigned int));
564 my_member_list_entries = member_list_entries;
565
566 last_sync_ring_id.nodeid = ring_id->nodeid;
567 last_sync_ring_id.seq = ring_id->seq;
568
569 entries = 0;
570 /*
571 * Determine list of nodeids for downlist message
572 */
573 for (i = 0; i < my_old_member_list_entries; i++) {
574 found = 0;
575 for (j = 0; j < trans_list_entries; j++) {
576 if (my_old_member_list[i] == trans_list[j]) {
577 found = 1;
578 break;
579 }
580 }
581 if (found == 0) {
582 g_req_exec_cpg_downlist.nodeids[entries++] =
583 my_old_member_list[i];
584 }
585 }
586 g_req_exec_cpg_downlist.left_nodes = entries;
587 }
588
589 static int cpg_sync_process (void)
590 {
591 int res = -1;
592
593 if (my_sync_state == CPGSYNC_DOWNLIST) {
594 res = cpg_exec_send_downlist();
595 if (res == -1) {
596 return (-1);
597 }
598 my_sync_state = CPGSYNC_JOINLIST;
599 }
600 if (my_sync_state == CPGSYNC_JOINLIST) {
601 res = cpg_exec_send_joinlist();
602 }
603 return (res);
604 }
605
606 static void cpg_sync_activate (void)
607 {
608 memcpy (my_old_member_list, my_member_list,
609 my_member_list_entries * sizeof (unsigned int));
610 my_old_member_list_entries = my_member_list_entries;
611
612 downlist_inform_clients ();
613
614 joinlist_inform_clients ();
615
616 joinlist_messages_delete ();
617
618 notify_lib_totem_membership (NULL, my_member_list_entries, my_member_list);
619 }
620
621 static void cpg_sync_abort (void)
622 {
623
624 joinlist_messages_delete ();
625 }
626
627 static int notify_lib_totem_membership (
628 void *conn,
629 int member_list_entries,
630 const unsigned int *member_list)
631 {
632 struct qb_list_head *iter;
633 char *buf;
634 int size;
635 struct res_lib_cpg_totem_confchg_callback *res;
636
637 size = sizeof(struct res_lib_cpg_totem_confchg_callback) +
638 sizeof(mar_uint32_t) * (member_list_entries);
639 buf = alloca(size);
640 if (!buf)
641 return CS_ERR_LIBRARY;
642
643 res = (struct res_lib_cpg_totem_confchg_callback *)buf;
644 res->member_list_entries = member_list_entries;
645 res->header.size = size;
646 res->header.id = MESSAGE_RES_CPG_TOTEM_CONFCHG_CALLBACK;
647 res->header.error = CS_OK;
648
649 memcpy (&res->ring_id, &last_sync_ring_id, sizeof (mar_cpg_ring_id_t));
650 memcpy (res->member_list, member_list, res->member_list_entries * sizeof (mar_uint32_t));
651
652 if (conn == NULL) {
653 qb_list_for_each(iter, &cpg_pd_list_head) {
654 struct cpg_pd *cpg_pd = qb_list_entry (iter, struct cpg_pd, list);
655 api->ipc_dispatch_send (cpg_pd->conn, buf, size);
656 }
657 } else {
658 api->ipc_dispatch_send (conn, buf, size);
659 }
660
661 return CS_OK;
662 }
663
664 /*
665 * Helper function for notify_lib_joinlist which prepares member_list using
666 * process_info_list with removed left_list items.
667 * member_list_entries - When not NULL it contains number of member_list entries
668 * member_list - When not NULL it is used as pointer to start of preallocated
669 * array of members. Pointer is adjusted to the end of array on
670 * exit.
671 */
672 static void notify_lib_joinlist_fill_member_list(
673 const mar_cpg_name_t *group_name,
674 int left_list_entries,
675 const mar_cpg_address_t *left_list,
676 int *member_list_entries,
677 mar_cpg_address_t **member_list)
678 {
679 struct qb_list_head *iter;
680 int i;
681
682 if (member_list_entries != NULL) {
683 *member_list_entries = 0;
684 }
685
686 qb_list_for_each(iter, &process_info_list_head) {
687 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
688
689 if (mar_name_compare (&pi->group, group_name) == 0) {
690 int in_left_list = 0;
691
692 for (i = 0; i < left_list_entries; i++) {
693 if (left_list[i].nodeid == pi->nodeid && left_list[i].pid == pi->pid) {
694 in_left_list = 1;
695 break ;
696 }
697 }
698
699 if (!in_left_list) {
700 if (member_list_entries != NULL) {
701 (*member_list_entries)++;
702 }
703
704 if (member_list != NULL) {
705 (*member_list)->nodeid = pi->nodeid;
706 (*member_list)->pid = pi->pid;
707 (*member_list)->reason = CPG_REASON_UNDEFINED;
708 (*member_list)++;
709 }
710 }
711 }
712 }
713 }
714
715 static int notify_lib_joinlist(
716 const mar_cpg_name_t *group_name,
717 int joined_list_entries,
718 mar_cpg_address_t *joined_list,
719 int left_list_entries,
720 mar_cpg_address_t *left_list,
721 int id)
722 {
723 int size;
724 char *buf;
725 struct qb_list_head *iter;
726 int member_list_entries;
727 struct res_lib_cpg_confchg_callback *res;
728 mar_cpg_address_t *retgi;
729 int i;
730
731 /*
732 * Find size of member_list (use process_info_list but remove items in left_list)
733 */
734 notify_lib_joinlist_fill_member_list(group_name, left_list_entries, left_list,
735 &member_list_entries, NULL);
736
737 size = sizeof(struct res_lib_cpg_confchg_callback) +
738 sizeof(mar_cpg_address_t) * (member_list_entries + left_list_entries + joined_list_entries);
739 buf = alloca(size);
740 if (!buf)
741 return CS_ERR_LIBRARY;
742
743 res = (struct res_lib_cpg_confchg_callback *)buf;
744 res->joined_list_entries = joined_list_entries;
745 res->left_list_entries = left_list_entries;
746 res->member_list_entries = member_list_entries;
747 retgi = res->member_list;
748 res->header.size = size;
749 res->header.id = id;
750 res->header.error = CS_OK;
751 memcpy(&res->group_name, group_name, sizeof(mar_cpg_name_t));
752
753 /*
754 * Fill res->memberlist. Use process_info_list but remove items in left_list.
755 */
756 notify_lib_joinlist_fill_member_list(group_name, left_list_entries, left_list,
757 NULL, &retgi);
758
759 /*
760 * Fill res->left_list
761 */
762 if (left_list_entries) {
763 memcpy (retgi, left_list, left_list_entries * sizeof(mar_cpg_address_t));
764 retgi += left_list_entries;
765 }
766
767 if (joined_list_entries) {
768 /*
769 * Fill res->joined_list
770 */
771 memcpy (retgi, joined_list, joined_list_entries * sizeof(mar_cpg_address_t));
772 retgi += joined_list_entries;
773
774 /*
775 * Update cpd_state for all local joined processes in group
776 */
777 for (i = 0; i < joined_list_entries; i++) {
778 if (joined_list[i].nodeid == api->totem_nodeid_get()) {
779 qb_list_for_each(iter, &cpg_pd_list_head) {
780 struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
781 if (joined_list[i].pid == cpd->pid &&
782 mar_name_compare (&cpd->group_name, group_name) == 0) {
783 cpd->cpd_state = CPD_STATE_JOIN_COMPLETED;
784 }
785 }
786 }
787 }
788 }
789
790 /*
791 * Send notification to all ipc clients joined in group_name
792 */
793 qb_list_for_each(iter, &cpg_pd_list_head) {
794 struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
795 if (mar_name_compare (&cpd->group_name, group_name) == 0) {
796 if (cpd->cpd_state == CPD_STATE_JOIN_COMPLETED ||
797 cpd->cpd_state == CPD_STATE_LEAVE_STARTED) {
798
799 api->ipc_dispatch_send (cpd->conn, buf, size);
800 cpd->transition_counter++;
801 }
802 }
803 }
804
805 if (left_list_entries) {
806 /*
807 * Zero internal cpd state for all local processes leaving group
808 * (this loop is not strictly needed because left_list always either
809 * contains exactly one process running on local node or more items
810 * but none of them is running on local node)
811 */
812 for (i = 0; i < joined_list_entries; i++) {
813 if (left_list[i].nodeid == api->totem_nodeid_get() &&
814 left_list[i].reason == CONFCHG_CPG_REASON_LEAVE) {
815 qb_list_for_each(iter, &cpg_pd_list_head) {
816 struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
817 if (left_list[i].pid == cpd->pid &&
818 mar_name_compare (&cpd->group_name, group_name) == 0) {
819 cpd->pid = 0;
820 memset (&cpd->group_name, 0, sizeof(cpd->group_name));
821 cpd->cpd_state = CPD_STATE_UNJOINED;
822 }
823 }
824 }
825 }
826 }
827
828 /*
829 * Traverse thru cpds and send totem membership for cpd, where it is not send yet
830 */
831 qb_list_for_each(iter, &cpg_pd_list_head) {
832 struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
833
834 if ((cpd->flags & CPG_MODEL_V1_DELIVER_INITIAL_TOTEM_CONF) && (cpd->initial_totem_conf_sent == 0)) {
835 cpd->initial_totem_conf_sent = 1;
836
837 notify_lib_totem_membership (cpd->conn, my_old_member_list_entries, my_old_member_list);
838 }
839 }
840
841 return CS_OK;
842 }
843
844 static void downlist_log(const char *msg, struct req_exec_cpg_downlist *dl)
845 {
846 log_printf (LOG_DEBUG,
847 "%s: members(old:%d left:%d)",
848 msg,
849 dl->old_members,
850 dl->left_nodes);
851 }
852
853 static void downlist_inform_clients (void)
854 {
855 struct qb_list_head *iter, *tmp_iter;
856 struct process_info *left_pi;
857 qb_map_t *group_map;
858 struct cpg_name cpg_group;
859 mar_cpg_name_t group;
860 struct confchg_data{
861 struct cpg_name cpg_group;
862 mar_cpg_address_t left_list[CPG_MEMBERS_MAX];
863 int left_list_entries;
864 struct qb_list_head list;
865 } *pcd;
866 qb_map_iter_t *miter;
867 int i, size;
868
869 downlist_log("my downlist", &g_req_exec_cpg_downlist);
870
871 group_map = qb_skiplist_create();
872
873 /*
874 * only the cpg groups included in left nodes should receive
875 * confchg event, so we will collect these cpg groups and
876 * relative left_lists here.
877 */
878 qb_list_for_each_safe(iter, tmp_iter, &process_info_list_head) {
879 struct process_info *pi = qb_list_entry(iter, struct process_info, list);
880
881 left_pi = NULL;
882 for (i = 0; i < g_req_exec_cpg_downlist.left_nodes; i++) {
883
884 if (pi->nodeid == g_req_exec_cpg_downlist.nodeids[i]) {
885 left_pi = pi;
886 break;
887 }
888 }
889
890 if (left_pi) {
891 marshall_from_mar_cpg_name_t(&cpg_group, &left_pi->group);
892 cpg_group.value[cpg_group.length] = 0;
893
894 pcd = (struct confchg_data *)qb_map_get(group_map, cpg_group.value);
895 if (pcd == NULL) {
896 pcd = (struct confchg_data *)calloc(1, sizeof(struct confchg_data));
897 memcpy(&pcd->cpg_group, &cpg_group, sizeof(struct cpg_name));
898 qb_map_put(group_map, pcd->cpg_group.value, pcd);
899 }
900 size = pcd->left_list_entries;
901 pcd->left_list[size].nodeid = left_pi->nodeid;
902 pcd->left_list[size].pid = left_pi->pid;
903 pcd->left_list[size].reason = CONFCHG_CPG_REASON_NODEDOWN;
904 pcd->left_list_entries++;
905 qb_list_del (&left_pi->list);
906 free (left_pi);
907 }
908 }
909
910 /* send only one confchg event per cpg group */
911 miter = qb_map_iter_create(group_map);
912 while (qb_map_iter_next(miter, (void **)&pcd)) {
913 marshall_to_mar_cpg_name_t(&group, &pcd->cpg_group);
914
915 log_printf (LOG_DEBUG, "left_list_entries:%d", pcd->left_list_entries);
916 for (i=0; i<pcd->left_list_entries; i++) {
917 log_printf (LOG_DEBUG, "left_list[%d] group:%s, ip:%s, pid:%d",
918 i, cpg_print_group_name(&group),
919 (char*)api->totem_ifaces_print(pcd->left_list[i].nodeid),
920 pcd->left_list[i].pid);
921 }
922
923 /* send confchg event */
924 notify_lib_joinlist(&group,
925 0, NULL,
926 pcd->left_list_entries,
927 pcd->left_list,
928 MESSAGE_RES_CPG_CONFCHG_CALLBACK);
929
930 free(pcd);
931 }
932 qb_map_iter_free(miter);
933 qb_map_destroy(group_map);
934 }
935
936 /*
937 * Remove processes that might have left the group while we were suspended.
938 */
939 static void joinlist_remove_zombie_pi_entries (void)
940 {
941 struct qb_list_head *pi_iter, *tmp_iter;
942 struct qb_list_head *jl_iter;
943 struct process_info *pi;
944 struct joinlist_msg *stored_msg;
945 int found;
946
947 qb_list_for_each_safe(pi_iter, tmp_iter, &process_info_list_head) {
948 pi = qb_list_entry (pi_iter, struct process_info, list);
949
950 /*
951 * Ignore local node
952 */
953 if (pi->nodeid == api->totem_nodeid_get()) {
954 continue ;
955 }
956
957 /*
958 * Try to find message in joinlist messages
959 */
960 found = 0;
961 qb_list_for_each(jl_iter, &joinlist_messages_head) {
962 stored_msg = qb_list_entry(jl_iter, struct joinlist_msg, list);
963
964 if (stored_msg->sender_nodeid == api->totem_nodeid_get()) {
965 continue ;
966 }
967
968 if (pi->nodeid == stored_msg->sender_nodeid &&
969 pi->pid == stored_msg->pid &&
970 mar_name_compare (&pi->group, &stored_msg->group_name) == 0) {
971 found = 1;
972 break ;
973 }
974 }
975
976 if (!found) {
977 do_proc_leave(&pi->group, pi->pid, pi->nodeid, CONFCHG_CPG_REASON_PROCDOWN);
978 }
979 }
980 }
981
982 static void joinlist_inform_clients (void)
983 {
984 struct joinlist_msg *stored_msg;
985 struct qb_list_head *iter;
986 unsigned int i;
987 qb_map_t *group_notify_map;
988 qb_map_iter_t *miter;
989 struct join_list_confchg_data *jld;
990
991 group_notify_map = qb_skiplist_create();
992
993 i = 0;
994 qb_list_for_each(iter, &joinlist_messages_head) {
995 stored_msg = qb_list_entry(iter, struct joinlist_msg, list);
996
997 log_printf (LOG_DEBUG, "joinlist_messages[%u] group:%s, ip:%s, pid:%d",
998 i++, cpg_print_group_name(&stored_msg->group_name),
999 (char*)api->totem_ifaces_print(stored_msg->sender_nodeid),
1000 stored_msg->pid);
1001
1002 /* Ignore our own messages */
1003 if (stored_msg->sender_nodeid == api->totem_nodeid_get()) {
1004 continue ;
1005 }
1006
1007 do_proc_join (&stored_msg->group_name, stored_msg->pid, stored_msg->sender_nodeid,
1008 CONFCHG_CPG_REASON_NODEUP, group_notify_map);
1009 }
1010
1011 miter = qb_map_iter_create(group_notify_map);
1012 while (qb_map_iter_next(miter, (void **)&jld)) {
1013 notify_lib_joinlist(&jld->cpg_group,
1014 jld->join_list_entries, jld->join_list,
1015 0, NULL,
1016 MESSAGE_RES_CPG_CONFCHG_CALLBACK);
1017 free(jld);
1018 }
1019 qb_map_iter_free(miter);
1020 qb_map_destroy(group_notify_map);
1021
1022 joinlist_remove_zombie_pi_entries ();
1023 }
1024
1025 static void joinlist_messages_delete (void)
1026 {
1027 struct joinlist_msg *stored_msg;
1028 struct qb_list_head *iter, *tmp_iter;
1029
1030 qb_list_for_each_safe(iter, tmp_iter, &joinlist_messages_head) {
1031 stored_msg = qb_list_entry(iter, struct joinlist_msg, list);
1032 qb_list_del (&stored_msg->list);
1033 free (stored_msg);
1034 }
1035 qb_list_init (&joinlist_messages_head);
1036 }
1037
1038 static char *cpg_exec_init_fn (struct corosync_api_v1 *corosync_api)
1039 {
1040 qb_list_init (&joinlist_messages_head);
1041 api = corosync_api;
1042 return (NULL);
1043 }
1044
1045 static void cpg_iteration_instance_finalize (struct cpg_iteration_instance *cpg_iteration_instance)
1046 {
1047 struct qb_list_head *iter, *tmp_iter;
1048 struct process_info *pi;
1049
1050 qb_list_for_each_safe(iter, tmp_iter, &(cpg_iteration_instance->items_list_head)) {
1051 pi = qb_list_entry (iter, struct process_info, list);
1052 qb_list_del (&pi->list);
1053 free (pi);
1054 }
1055
1056 qb_list_del (&cpg_iteration_instance->list);
1057 hdb_handle_destroy (&cpg_iteration_handle_t_db, cpg_iteration_instance->handle);
1058 }
1059
1060 static void cpg_pd_finalize (struct cpg_pd *cpd)
1061 {
1062 struct qb_list_head *iter, *tmp_iter;
1063 struct cpg_iteration_instance *cpii;
1064
1065 zcb_all_free(cpd);
1066 qb_list_for_each_safe(iter, tmp_iter, &(cpd->iteration_instance_list_head)) {
1067 cpii = qb_list_entry (iter, struct cpg_iteration_instance, list);
1068
1069 cpg_iteration_instance_finalize (cpii);
1070 }
1071
1072 qb_list_del (&cpd->list);
1073 }
1074
1075 static int cpg_lib_exit_fn (void *conn)
1076 {
1077 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1078
1079 log_printf(LOGSYS_LEVEL_DEBUG, "exit_fn for conn=%p", conn);
1080
1081 if (cpd->group_name.length > 0 && cpd->cpd_state != CPD_STATE_LEAVE_STARTED) {
1082 cpg_node_joinleave_send (cpd->pid, &cpd->group_name,
1083 MESSAGE_REQ_EXEC_CPG_PROCLEAVE, CONFCHG_CPG_REASON_PROCDOWN);
1084 }
1085
1086 cpg_pd_finalize (cpd);
1087
1088 api->ipc_refcnt_dec (conn);
1089 return (0);
1090 }
1091
1092 static int cpg_node_joinleave_send (unsigned int pid, const mar_cpg_name_t *group_name, int fn, int reason)
1093 {
1094 struct req_exec_cpg_procjoin req_exec_cpg_procjoin;
1095 struct iovec req_exec_cpg_iovec;
1096 int result;
1097
1098 memset(&req_exec_cpg_procjoin, 0, sizeof(req_exec_cpg_procjoin));
1099
1100 memcpy(&req_exec_cpg_procjoin.group_name, group_name, sizeof(mar_cpg_name_t));
1101 req_exec_cpg_procjoin.pid = pid;
1102 req_exec_cpg_procjoin.reason = reason;
1103
1104 req_exec_cpg_procjoin.header.size = sizeof(req_exec_cpg_procjoin);
1105 req_exec_cpg_procjoin.header.id = SERVICE_ID_MAKE(CPG_SERVICE, fn);
1106
1107 req_exec_cpg_iovec.iov_base = (char *)&req_exec_cpg_procjoin;
1108 req_exec_cpg_iovec.iov_len = sizeof(req_exec_cpg_procjoin);
1109
1110 result = api->totem_mcast (&req_exec_cpg_iovec, 1, TOTEM_AGREED);
1111
1112 return (result);
1113 }
1114
1115 /* Can byteswap join & leave messages */
1116 static void exec_cpg_procjoin_endian_convert (void *msg)
1117 {
1118 struct req_exec_cpg_procjoin *req_exec_cpg_procjoin = msg;
1119
1120 req_exec_cpg_procjoin->pid = swab32(req_exec_cpg_procjoin->pid);
1121 swab_mar_cpg_name_t (&req_exec_cpg_procjoin->group_name);
1122 req_exec_cpg_procjoin->reason = swab32(req_exec_cpg_procjoin->reason);
1123 }
1124
1125 static void exec_cpg_joinlist_endian_convert (void *msg_v)
1126 {
1127 char *msg = msg_v;
1128 struct qb_ipc_response_header *res = (struct qb_ipc_response_header *)msg;
1129 struct join_list_entry *jle = (struct join_list_entry *)(msg + sizeof(struct qb_ipc_response_header));
1130
1131 swab_mar_int32_t (&res->size);
1132
1133 while ((const char*)jle < msg + res->size) {
1134 jle->pid = swab32(jle->pid);
1135 swab_mar_cpg_name_t (&jle->group_name);
1136 jle++;
1137 }
1138 }
1139
1140 static void exec_cpg_downlist_endian_convert_old (void *msg)
1141 {
1142 }
1143
1144 static void exec_cpg_downlist_endian_convert (void *msg)
1145 {
1146 struct req_exec_cpg_downlist *req_exec_cpg_downlist = msg;
1147 unsigned int i;
1148
1149 req_exec_cpg_downlist->left_nodes = swab32(req_exec_cpg_downlist->left_nodes);
1150 req_exec_cpg_downlist->old_members = swab32(req_exec_cpg_downlist->old_members);
1151
1152 for (i = 0; i < req_exec_cpg_downlist->left_nodes; i++) {
1153 req_exec_cpg_downlist->nodeids[i] = swab32(req_exec_cpg_downlist->nodeids[i]);
1154 }
1155 }
1156
1157
1158 static void exec_cpg_mcast_endian_convert (void *msg)
1159 {
1160 struct req_exec_cpg_mcast *req_exec_cpg_mcast = msg;
1161
1162 swab_coroipc_request_header_t (&req_exec_cpg_mcast->header);
1163 swab_mar_cpg_name_t (&req_exec_cpg_mcast->group_name);
1164 req_exec_cpg_mcast->pid = swab32(req_exec_cpg_mcast->pid);
1165 req_exec_cpg_mcast->msglen = swab32(req_exec_cpg_mcast->msglen);
1166 swab_mar_message_source_t (&req_exec_cpg_mcast->source);
1167 }
1168
1169 static void exec_cpg_partial_mcast_endian_convert (void *msg)
1170 {
1171 struct req_exec_cpg_partial_mcast *req_exec_cpg_mcast = msg;
1172
1173 swab_coroipc_request_header_t (&req_exec_cpg_mcast->header);
1174 swab_mar_cpg_name_t (&req_exec_cpg_mcast->group_name);
1175 req_exec_cpg_mcast->pid = swab32(req_exec_cpg_mcast->pid);
1176 req_exec_cpg_mcast->msglen = swab32(req_exec_cpg_mcast->msglen);
1177 req_exec_cpg_mcast->fraglen = swab32(req_exec_cpg_mcast->fraglen);
1178 req_exec_cpg_mcast->type = swab32(req_exec_cpg_mcast->type);
1179 swab_mar_message_source_t (&req_exec_cpg_mcast->source);
1180 }
1181
1182 static struct process_info *process_info_find(const mar_cpg_name_t *group_name, uint32_t pid, unsigned int nodeid) {
1183 struct qb_list_head *iter;
1184
1185 qb_list_for_each(iter, &process_info_list_head) {
1186 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
1187
1188 if (pi->pid == pid && pi->nodeid == nodeid &&
1189 mar_name_compare (&pi->group, group_name) == 0) {
1190 return pi;
1191 }
1192 }
1193
1194 return NULL;
1195 }
1196
1197 static void do_proc_join(
1198 const mar_cpg_name_t *name,
1199 uint32_t pid,
1200 unsigned int nodeid,
1201 int reason,
1202 qb_map_t *group_notify_map)
1203 {
1204 struct process_info *pi;
1205 struct process_info *pi_entry;
1206 mar_cpg_address_t notify_info;
1207 struct qb_list_head *list;
1208 struct qb_list_head *list_to_add = NULL;
1209 int size;
1210
1211 if (process_info_find (name, pid, nodeid) != NULL) {
1212 return ;
1213 }
1214 pi = malloc (sizeof (struct process_info));
1215 if (!pi) {
1216 log_printf(LOGSYS_LEVEL_WARNING, "Unable to allocate process_info struct");
1217 return;
1218 }
1219 pi->nodeid = nodeid;
1220 pi->pid = pid;
1221 memcpy(&pi->group, name, sizeof(*name));
1222 qb_list_init(&pi->list);
1223
1224 /*
1225 * Insert new process in sorted order so synchronization works properly
1226 */
1227 list_to_add = &process_info_list_head;
1228 qb_list_for_each(list, &process_info_list_head) {
1229 pi_entry = qb_list_entry(list, struct process_info, list);
1230 if (pi_entry->nodeid > pi->nodeid ||
1231 (pi_entry->nodeid == pi->nodeid && pi_entry->pid > pi->pid)) {
1232
1233 break;
1234 }
1235 list_to_add = list;
1236 }
1237 qb_list_add (&pi->list, list_to_add);
1238
1239 notify_info.pid = pi->pid;
1240 notify_info.nodeid = nodeid;
1241 notify_info.reason = reason;
1242
1243 if (group_notify_map == NULL) {
1244 notify_lib_joinlist(&pi->group,
1245 1, ¬ify_info,
1246 0, NULL,
1247 MESSAGE_RES_CPG_CONFCHG_CALLBACK);
1248 } else {
1249 struct join_list_confchg_data *jld = qb_map_get(group_notify_map, pi->group.value);
1250 if (jld == NULL) {
1251 jld = (struct join_list_confchg_data *)calloc(1, sizeof(struct join_list_confchg_data));
1252 memcpy(&jld->cpg_group, &pi->group, sizeof(mar_cpg_name_t));
1253 qb_map_put(group_notify_map, jld->cpg_group.value, jld);
1254 }
1255 size = jld->join_list_entries;
1256 jld->join_list[size].nodeid = notify_info.nodeid;
1257 jld->join_list[size].pid = notify_info.pid;
1258 jld->join_list[size].reason = notify_info.reason;
1259 jld->join_list_entries++;
1260 }
1261 }
1262
1263 static void do_proc_leave(
1264 const mar_cpg_name_t *name,
1265 uint32_t pid,
1266 unsigned int nodeid,
1267 int reason)
1268 {
1269 struct process_info *pi;
1270 struct qb_list_head *iter, *tmp_iter;
1271 mar_cpg_address_t notify_info;
1272
1273 notify_info.pid = pid;
1274 notify_info.nodeid = nodeid;
1275 notify_info.reason = reason;
1276
1277 notify_lib_joinlist(name,
1278 0, NULL,
1279 1, ¬ify_info,
1280 MESSAGE_RES_CPG_CONFCHG_CALLBACK);
1281
1282 qb_list_for_each_safe(iter, tmp_iter, &process_info_list_head) {
1283 pi = qb_list_entry(iter, struct process_info, list);
1284
1285 if (pi->pid == pid && pi->nodeid == nodeid &&
1286 mar_name_compare (&pi->group, name)==0) {
1287 qb_list_del (&pi->list);
1288 free (pi);
1289 }
1290 }
1291 }
1292
1293 static void message_handler_req_exec_cpg_downlist_old (
1294 const void *message,
1295 unsigned int nodeid)
1296 {
1297 log_printf (LOGSYS_LEVEL_DEBUG, "downlist OLD from node " CS_PRI_NODE_ID,
1298 nodeid);
1299 }
1300
1301 static void message_handler_req_exec_cpg_downlist(
1302 const void *message,
1303 unsigned int nodeid)
1304 {
1305 const struct req_exec_cpg_downlist *req_exec_cpg_downlist = message;
1306
1307 log_printf (LOGSYS_LEVEL_DEBUG, "downlist left_list: %d received",
1308 req_exec_cpg_downlist->left_nodes);
1309 }
1310
1311
1312 static void message_handler_req_exec_cpg_procjoin (
1313 const void *message,
1314 unsigned int nodeid)
1315 {
1316 const struct req_exec_cpg_procjoin *req_exec_cpg_procjoin = message;
1317
1318 log_printf(LOGSYS_LEVEL_DEBUG, "got procjoin message from cluster node " CS_PRI_NODE_ID " (%s) for pid %u",
1319 nodeid,
1320 api->totem_ifaces_print(nodeid),
1321 (unsigned int)req_exec_cpg_procjoin->pid);
1322
1323 do_proc_join (&req_exec_cpg_procjoin->group_name,
1324 req_exec_cpg_procjoin->pid, nodeid,
1325 CONFCHG_CPG_REASON_JOIN, NULL);
1326 }
1327
1328 static void message_handler_req_exec_cpg_procleave (
1329 const void *message,
1330 unsigned int nodeid)
1331 {
1332 const struct req_exec_cpg_procjoin *req_exec_cpg_procjoin = message;
1333
1334 log_printf(LOGSYS_LEVEL_DEBUG, "got procleave message from cluster node " CS_PRI_NODE_ID " (%s) for pid %u",
1335 nodeid,
1336 api->totem_ifaces_print(nodeid),
1337 (unsigned int)req_exec_cpg_procjoin->pid);
1338
1339 do_proc_leave (&req_exec_cpg_procjoin->group_name,
1340 req_exec_cpg_procjoin->pid, nodeid,
1341 req_exec_cpg_procjoin->reason);
1342 }
1343
1344
1345 /* Got a proclist from another node */
1346 static void message_handler_req_exec_cpg_joinlist (
1347 const void *message_v,
1348 unsigned int nodeid)
1349 {
1350 const char *message = message_v;
1351 const struct qb_ipc_response_header *res = (const struct qb_ipc_response_header *)message;
1352 const struct join_list_entry *jle = (const struct join_list_entry *)(message + sizeof(struct qb_ipc_response_header));
1353 struct joinlist_msg *stored_msg;
1354
1355 log_printf(LOGSYS_LEVEL_DEBUG, "got joinlist message from node " CS_PRI_NODE_ID,
1356 nodeid);
1357
1358 while ((const char*)jle < message + res->size) {
1359 stored_msg = malloc (sizeof (struct joinlist_msg));
1360 memset(stored_msg, 0, sizeof (struct joinlist_msg));
1361 stored_msg->sender_nodeid = nodeid;
1362 stored_msg->pid = jle->pid;
1363 memcpy(&stored_msg->group_name, &jle->group_name, sizeof(mar_cpg_name_t));
1364 qb_list_init (&stored_msg->list);
1365 qb_list_add (&stored_msg->list, &joinlist_messages_head);
1366 jle++;
1367 }
1368 }
1369
1370 static void message_handler_req_exec_cpg_mcast (
1371 const void *message,
1372 unsigned int nodeid)
1373 {
1374 const struct req_exec_cpg_mcast *req_exec_cpg_mcast = message;
1375 struct res_lib_cpg_deliver_callback res_lib_cpg_mcast;
1376 int msglen = req_exec_cpg_mcast->msglen;
1377 struct qb_list_head *iter, *pi_iter, *tmp_iter;
1378 struct cpg_pd *cpd;
1379 struct iovec iovec[2];
1380 int known_node = 0;
1381
1382 res_lib_cpg_mcast.header.id = MESSAGE_RES_CPG_DELIVER_CALLBACK;
1383 res_lib_cpg_mcast.header.size = sizeof(res_lib_cpg_mcast) + msglen;
1384 res_lib_cpg_mcast.msglen = msglen;
1385 res_lib_cpg_mcast.pid = req_exec_cpg_mcast->pid;
1386 res_lib_cpg_mcast.nodeid = nodeid;
1387
1388 memcpy(&res_lib_cpg_mcast.group_name, &req_exec_cpg_mcast->group_name,
1389 sizeof(mar_cpg_name_t));
1390 iovec[0].iov_base = (void *)&res_lib_cpg_mcast;
1391 iovec[0].iov_len = sizeof (res_lib_cpg_mcast);
1392
1393 iovec[1].iov_base = (char*)message+sizeof(*req_exec_cpg_mcast);
1394 iovec[1].iov_len = msglen;
1395
1396 qb_list_for_each_safe(iter, tmp_iter, &cpg_pd_list_head) {
1397 cpd = qb_list_entry(iter, struct cpg_pd, list);
1398 if ((cpd->cpd_state == CPD_STATE_LEAVE_STARTED || cpd->cpd_state == CPD_STATE_JOIN_COMPLETED)
1399 && (mar_name_compare (&cpd->group_name, &req_exec_cpg_mcast->group_name) == 0)) {
1400
1401 if (!known_node) {
1402 /* Try to find, if we know the node */
1403 qb_list_for_each(pi_iter, &process_info_list_head) {
1404 struct process_info *pi = qb_list_entry (pi_iter, struct process_info, list);
1405
1406 if (pi->nodeid == nodeid &&
1407 mar_name_compare (&pi->group, &req_exec_cpg_mcast->group_name) == 0) {
1408 known_node = 1;
1409 break;
1410 }
1411 }
1412 }
1413
1414 if (!known_node) {
1415 log_printf(LOGSYS_LEVEL_WARNING, "Unknown node -> we will not deliver message");
1416 return ;
1417 }
1418
1419 api->ipc_dispatch_iov_send (cpd->conn, iovec, 2);
1420 }
1421 }
1422 }
1423
1424 static void message_handler_req_exec_cpg_partial_mcast (
1425 const void *message,
1426 unsigned int nodeid)
1427 {
1428 const struct req_exec_cpg_partial_mcast *req_exec_cpg_mcast = message;
1429 struct res_lib_cpg_partial_deliver_callback res_lib_cpg_mcast;
1430 int msglen = req_exec_cpg_mcast->fraglen;
1431 struct qb_list_head *iter, *pi_iter, *tmp_iter;
1432 struct cpg_pd *cpd;
1433 struct iovec iovec[2];
1434 int known_node = 0;
1435
1436 log_printf(LOGSYS_LEVEL_DEBUG, "Got fragmented message from node " CS_PRI_NODE_ID ", size = %d bytes\n", nodeid, msglen);
1437
1438 res_lib_cpg_mcast.header.id = MESSAGE_RES_CPG_PARTIAL_DELIVER_CALLBACK;
1439 res_lib_cpg_mcast.header.size = sizeof(res_lib_cpg_mcast) + msglen;
1440 res_lib_cpg_mcast.fraglen = msglen;
1441 res_lib_cpg_mcast.msglen = req_exec_cpg_mcast->msglen;
1442 res_lib_cpg_mcast.pid = req_exec_cpg_mcast->pid;
1443 res_lib_cpg_mcast.type = req_exec_cpg_mcast->type;
1444 res_lib_cpg_mcast.nodeid = nodeid;
1445
1446 memcpy(&res_lib_cpg_mcast.group_name, &req_exec_cpg_mcast->group_name,
1447 sizeof(mar_cpg_name_t));
1448 iovec[0].iov_base = (void *)&res_lib_cpg_mcast;
1449 iovec[0].iov_len = sizeof (res_lib_cpg_mcast);
1450
1451 iovec[1].iov_base = (char*)message+sizeof(*req_exec_cpg_mcast);
1452 iovec[1].iov_len = msglen;
1453
1454 qb_list_for_each_safe(iter, tmp_iter, &cpg_pd_list_head) {
1455 cpd = qb_list_entry(iter, struct cpg_pd, list);
1456
1457 if ((cpd->cpd_state == CPD_STATE_LEAVE_STARTED || cpd->cpd_state == CPD_STATE_JOIN_COMPLETED)
1458 && (mar_name_compare (&cpd->group_name, &req_exec_cpg_mcast->group_name) == 0)) {
1459
1460 if (!known_node) {
1461 /* Try to find, if we know the node */
1462 qb_list_for_each(pi_iter, &process_info_list_head) {
1463 struct process_info *pi = qb_list_entry (pi_iter, struct process_info, list);
1464
1465 if (pi->nodeid == nodeid &&
1466 mar_name_compare (&pi->group, &req_exec_cpg_mcast->group_name) == 0) {
1467 known_node = 1;
1468 break;
1469 }
1470 }
1471 }
1472
1473 if (!known_node) {
1474 log_printf(LOGSYS_LEVEL_WARNING, "Unknown node -> we will not deliver message");
1475 return ;
1476 }
1477
1478 api->ipc_dispatch_iov_send (cpd->conn, iovec, 2);
1479 }
1480 }
1481 }
1482
1483
1484 static int cpg_exec_send_downlist(void)
1485 {
1486 struct iovec iov;
1487
1488 g_req_exec_cpg_downlist.header.id = SERVICE_ID_MAKE(CPG_SERVICE, MESSAGE_REQ_EXEC_CPG_DOWNLIST);
1489 g_req_exec_cpg_downlist.header.size = sizeof(struct req_exec_cpg_downlist);
1490
1491 g_req_exec_cpg_downlist.old_members = my_old_member_list_entries;
1492
1493 iov.iov_base = (void *)&g_req_exec_cpg_downlist;
1494 iov.iov_len = g_req_exec_cpg_downlist.header.size;
1495
1496 return (api->totem_mcast (&iov, 1, TOTEM_AGREED));
1497 }
1498
1499 static int cpg_exec_send_joinlist(void)
1500 {
1501 int count = 0;
1502 struct qb_list_head *iter;
1503 struct qb_ipc_response_header *res;
1504 char *buf;
1505 size_t buf_size;
1506 struct join_list_entry *jle;
1507 struct iovec req_exec_cpg_iovec;
1508
1509 qb_list_for_each(iter, &process_info_list_head) {
1510 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
1511
1512 if (pi->nodeid == api->totem_nodeid_get ()) {
1513 count++;
1514 }
1515 }
1516
1517 /* Nothing to send */
1518 if (!count)
1519 return 0;
1520
1521 buf_size = sizeof(struct qb_ipc_response_header) + sizeof(struct join_list_entry) * count;
1522 buf = alloca(buf_size);
1523 if (!buf) {
1524 log_printf(LOGSYS_LEVEL_WARNING, "Unable to allocate joinlist buffer");
1525 return -1;
1526 }
1527 memset(buf, 0, buf_size);
1528
1529 jle = (struct join_list_entry *)(buf + sizeof(struct qb_ipc_response_header));
1530 res = (struct qb_ipc_response_header *)buf;
1531
1532 qb_list_for_each(iter, &process_info_list_head) {
1533 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
1534
1535 if (pi->nodeid == api->totem_nodeid_get ()) {
1536 memcpy (&jle->group_name, &pi->group, sizeof (mar_cpg_name_t));
1537 jle->pid = pi->pid;
1538 jle++;
1539 }
1540 }
1541
1542 res->id = SERVICE_ID_MAKE(CPG_SERVICE, MESSAGE_REQ_EXEC_CPG_JOINLIST);
1543 res->size = sizeof(struct qb_ipc_response_header)+sizeof(struct join_list_entry) * count;
1544
1545 req_exec_cpg_iovec.iov_base = buf;
1546 req_exec_cpg_iovec.iov_len = res->size;
1547
1548 return (api->totem_mcast (&req_exec_cpg_iovec, 1, TOTEM_AGREED));
1549 }
1550
1551 static int cpg_lib_init_fn (void *conn)
1552 {
1553 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1554 memset (cpd, 0, sizeof(struct cpg_pd));
1555 cpd->conn = conn;
1556 qb_list_add (&cpd->list, &cpg_pd_list_head);
1557
1558 qb_list_init (&cpd->iteration_instance_list_head);
1559 qb_list_init (&cpd->zcb_mapped_list_head);
1560
1561 api->ipc_refcnt_inc (conn);
1562 log_printf(LOGSYS_LEVEL_DEBUG, "lib_init_fn: conn=%p, cpd=%p", conn, cpd);
1563 return (0);
1564 }
1565
1566 /* Join message from the library */
1567 static void message_handler_req_lib_cpg_join (void *conn, const void *message)
1568 {
1569 const struct req_lib_cpg_join *req_lib_cpg_join = message;
1570 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1571 struct res_lib_cpg_join res_lib_cpg_join;
1572 cs_error_t error = CS_OK;
1573 struct qb_list_head *iter;
1574
1575 /* Test, if we don't have same pid and group name joined */
1576 qb_list_for_each(iter, &cpg_pd_list_head) {
1577 struct cpg_pd *cpd_item = qb_list_entry (iter, struct cpg_pd, list);
1578
1579 if (cpd_item->pid == req_lib_cpg_join->pid &&
1580 mar_name_compare(&req_lib_cpg_join->group_name, &cpd_item->group_name) == 0) {
1581
1582 /* We have same pid and group name joined -> return error */
1583 error = CS_ERR_EXIST;
1584 goto response_send;
1585 }
1586 }
1587
1588 /*
1589 * Same check must be done in process info list, because there may be not yet delivered
1590 * leave of client.
1591 */
1592 qb_list_for_each(iter, &process_info_list_head) {
1593 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
1594
1595 if (pi->nodeid == api->totem_nodeid_get () && pi->pid == req_lib_cpg_join->pid &&
1596 mar_name_compare(&req_lib_cpg_join->group_name, &pi->group) == 0) {
1597 /* We have same pid and group name joined -> return error */
1598 error = CS_ERR_TRY_AGAIN;
1599 goto response_send;
1600 }
1601 }
1602
1603 if (req_lib_cpg_join->group_name.length > CPG_MAX_NAME_LENGTH) {
1604 error = CS_ERR_NAME_TOO_LONG;
1605 goto response_send;
1606 }
1607
1608 switch (cpd->cpd_state) {
1609 case CPD_STATE_UNJOINED:
1610 error = CS_OK;
1611 cpd->cpd_state = CPD_STATE_JOIN_STARTED;
1612 cpd->pid = req_lib_cpg_join->pid;
1613 cpd->flags = req_lib_cpg_join->flags;
1614 memcpy (&cpd->group_name, &req_lib_cpg_join->group_name,
1615 sizeof (cpd->group_name));
1616
1617 cpg_node_joinleave_send (req_lib_cpg_join->pid,
1618 &req_lib_cpg_join->group_name,
1619 MESSAGE_REQ_EXEC_CPG_PROCJOIN, CONFCHG_CPG_REASON_JOIN);
1620 break;
1621 case CPD_STATE_LEAVE_STARTED:
1622 error = CS_ERR_BUSY;
1623 break;
1624 case CPD_STATE_JOIN_STARTED:
1625 error = CS_ERR_EXIST;
1626 break;
1627 case CPD_STATE_JOIN_COMPLETED:
1628 error = CS_ERR_EXIST;
1629 break;
1630 }
1631
1632 response_send:
1633 res_lib_cpg_join.header.size = sizeof(res_lib_cpg_join);
1634 res_lib_cpg_join.header.id = MESSAGE_RES_CPG_JOIN;
1635 res_lib_cpg_join.header.error = error;
1636 api->ipc_response_send (conn, &res_lib_cpg_join, sizeof(res_lib_cpg_join));
1637 }
1638
1639 /* Leave message from the library */
1640 static void message_handler_req_lib_cpg_leave (void *conn, const void *message)
1641 {
1642 struct res_lib_cpg_leave res_lib_cpg_leave;
1643 cs_error_t error = CS_OK;
1644 struct req_lib_cpg_leave *req_lib_cpg_leave = (struct req_lib_cpg_leave *)message;
1645 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1646
1647 log_printf(LOGSYS_LEVEL_DEBUG, "got leave request on %p", conn);
1648
1649 switch (cpd->cpd_state) {
1650 case CPD_STATE_UNJOINED:
1651 error = CS_ERR_NOT_EXIST;
1652 break;
1653 case CPD_STATE_LEAVE_STARTED:
1654 error = CS_ERR_NOT_EXIST;
1655 break;
1656 case CPD_STATE_JOIN_STARTED:
1657 error = CS_ERR_BUSY;
1658 break;
1659 case CPD_STATE_JOIN_COMPLETED:
1660 error = CS_OK;
1661 cpd->cpd_state = CPD_STATE_LEAVE_STARTED;
1662 cpg_node_joinleave_send (req_lib_cpg_leave->pid,
1663 &req_lib_cpg_leave->group_name,
1664 MESSAGE_REQ_EXEC_CPG_PROCLEAVE,
1665 CONFCHG_CPG_REASON_LEAVE);
1666 break;
1667 }
1668
1669 /* send return */
1670 res_lib_cpg_leave.header.size = sizeof(res_lib_cpg_leave);
1671 res_lib_cpg_leave.header.id = MESSAGE_RES_CPG_LEAVE;
1672 res_lib_cpg_leave.header.error = error;
1673 api->ipc_response_send(conn, &res_lib_cpg_leave, sizeof(res_lib_cpg_leave));
1674 }
1675
1676 /* Finalize message from library */
1677 static void message_handler_req_lib_cpg_finalize (
1678 void *conn,
1679 const void *message)
1680 {
1681 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1682 struct res_lib_cpg_finalize res_lib_cpg_finalize;
1683 cs_error_t error = CS_OK;
1684
1685 log_printf (LOGSYS_LEVEL_DEBUG, "cpg finalize for conn=%p", conn);
1686
1687 /*
1688 * We will just remove cpd from list. After this call, connection will be
1689 * closed on lib side, and cpg_lib_exit_fn will be called
1690 */
1691 qb_list_del (&cpd->list);
1692 qb_list_init (&cpd->list);
1693
1694 res_lib_cpg_finalize.header.size = sizeof (res_lib_cpg_finalize);
1695 res_lib_cpg_finalize.header.id = MESSAGE_RES_CPG_FINALIZE;
1696 res_lib_cpg_finalize.header.error = error;
1697
1698 api->ipc_response_send (conn, &res_lib_cpg_finalize,
1699 sizeof (res_lib_cpg_finalize));
1700 }
1701
1702 static int
1703 memory_map (
1704 const char *path,
1705 size_t bytes,
1706 void **buf)
1707 {
1708 int32_t fd;
1709 void *addr;
1710 int32_t res;
1711
1712 fd = open (path, O_RDWR, 0600);
1713
1714 unlink (path);
1715
1716 if (fd == -1) {
1717 return (-1);
1718 }
1719
1720 res = ftruncate (fd, bytes);
1721 if (res == -1) {
1722 goto error_close_unlink;
1723 }
1724
1725 addr = mmap (NULL, bytes, PROT_READ | PROT_WRITE,
1726 MAP_SHARED, fd, 0);
1727
1728 if (addr == MAP_FAILED) {
1729 goto error_close_unlink;
1730 }
1731 #ifdef MADV_NOSYNC
1732 madvise(addr, bytes, MADV_NOSYNC);
1733 #endif
1734
1735 res = close (fd);
1736 if (res) {
1737 munmap (addr, bytes);
1738 return (-1);
1739 }
1740 *buf = addr;
1741 return (0);
1742
1743 error_close_unlink:
1744 close (fd);
1745 unlink(path);
1746 return -1;
1747 }
1748
1749 static inline int zcb_alloc (
1750 struct cpg_pd *cpd,
1751 const char *path_to_file,
1752 size_t size,
1753 void **addr)
1754 {
1755 struct zcb_mapped *zcb_mapped;
1756 unsigned int res;
1757
1758 zcb_mapped = malloc (sizeof (struct zcb_mapped));
1759 if (zcb_mapped == NULL) {
1760 return (-1);
1761 }
1762
1763 res = memory_map (
1764 path_to_file,
1765 size,
1766 addr);
1767 if (res == -1) {
1768 free (zcb_mapped);
1769 return (-1);
1770 }
1771
1772 qb_list_init (&zcb_mapped->list);
1773 zcb_mapped->addr = *addr;
1774 zcb_mapped->size = size;
1775 qb_list_add_tail (&zcb_mapped->list, &cpd->zcb_mapped_list_head);
1776 return (0);
1777 }
1778
1779
1780 static inline int zcb_free (struct zcb_mapped *zcb_mapped)
1781 {
1782 int res;
1783
1784 res = munmap (zcb_mapped->addr, zcb_mapped->size);
1785 qb_list_del (&zcb_mapped->list);
1786 free (zcb_mapped);
1787 return (res);
1788 }
1789
1790 static inline int zcb_by_addr_free (struct cpg_pd *cpd, void *addr)
1791 {
1792 struct qb_list_head *list, *tmp_iter;
1793 struct zcb_mapped *zcb_mapped;
1794 int res = 0;
1795
1796 qb_list_for_each_safe(list, tmp_iter, &(cpd->zcb_mapped_list_head)) {
1797 zcb_mapped = qb_list_entry (list, struct zcb_mapped, list);
1798
1799 if (zcb_mapped->addr == addr) {
1800 res = zcb_free (zcb_mapped);
1801 break;
1802 }
1803
1804 }
1805 return (res);
1806 }
1807
1808 static inline int zcb_all_free (
1809 struct cpg_pd *cpd)
1810 {
1811 struct qb_list_head *list, *tmp_iter;
1812 struct zcb_mapped *zcb_mapped;
1813
1814 qb_list_for_each_safe(list, tmp_iter, &(cpd->zcb_mapped_list_head)) {
1815 zcb_mapped = qb_list_entry (list, struct zcb_mapped, list);
1816
1817 zcb_free (zcb_mapped);
1818 }
1819 return (0);
1820 }
1821
1822 union u {
1823 uint64_t server_addr;
1824 void *server_ptr;
1825 };
1826
1827 static uint64_t void2serveraddr (void *server_ptr)
1828 {
1829 union u u;
1830
1831 u.server_ptr = server_ptr;
|
CID (unavailable; MK=373a2bfb8d356c141d16a1aa1a84e9fe) (#1 of 1): Inconsistent C union access (INCONSISTENT_UNION_ACCESS): |
|
(2) Event inconsistent_union_field_access: |
In "u.server_addr", the union field used: "server_addr" is inconsistent with the field most recently stored: "server_ptr". |
| Also see events: |
[assign_union_field] |
1832 return (u.server_addr);
1833 }
1834
1835 static void *serveraddr2void (uint64_t server_addr)
1836 {
1837 union u u;
1838
1839 u.server_addr = server_addr;
1840 return (u.server_ptr);
1841 };
1842
1843 static void message_handler_req_lib_cpg_zc_alloc (
1844 void *conn,
1845 const void *message)
1846 {
1847 mar_req_coroipcc_zc_alloc_t *hdr = (mar_req_coroipcc_zc_alloc_t *)message;
1848 struct qb_ipc_response_header res_header;
1849 void *addr = NULL;
1850 struct coroipcs_zc_header *zc_header;
1851 unsigned int res;
1852 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1853
1854 log_printf(LOGSYS_LEVEL_DEBUG, "path: %s", hdr->path_to_file);
1855
1856 res = zcb_alloc (cpd, hdr->path_to_file, hdr->map_size,
1857 &addr);
1858 assert(res == 0);
1859
1860 zc_header = (struct coroipcs_zc_header *)addr;
1861 zc_header->server_address = void2serveraddr(addr);
1862
1863 res_header.size = sizeof (struct qb_ipc_response_header);
1864 res_header.id = 0;
1865 api->ipc_response_send (conn,
1866 &res_header,
1867 res_header.size);
1868 }
1869
1870 static void message_handler_req_lib_cpg_zc_free (
1871 void *conn,
1872 const void *message)
1873 {
1874 mar_req_coroipcc_zc_free_t *hdr = (mar_req_coroipcc_zc_free_t *)message;
1875 struct qb_ipc_response_header res_header;
1876 void *addr = NULL;
1877 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1878
1879 log_printf(LOGSYS_LEVEL_DEBUG, " free'ing");
1880
1881 addr = serveraddr2void (hdr->server_address);
1882
1883 zcb_by_addr_free (cpd, addr);
1884
1885 res_header.size = sizeof (struct qb_ipc_response_header);
1886 res_header.id = 0;
1887 api->ipc_response_send (
1888 conn, &res_header,
1889 res_header.size);
1890 }
1891
1892 /* Fragmented mcast message from the library */
1893 static void message_handler_req_lib_cpg_partial_mcast (void *conn, const void *message)
1894 {
1895 const struct req_lib_cpg_partial_mcast *req_lib_cpg_mcast = message;
1896 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1897 mar_cpg_name_t group_name = cpd->group_name;
1898
1899 struct iovec req_exec_cpg_iovec[2];
1900 struct req_exec_cpg_partial_mcast req_exec_cpg_mcast;
1901 struct res_lib_cpg_partial_send res_lib_cpg_partial_send;
1902 int msglen = req_lib_cpg_mcast->fraglen;
1903 int result;
1904 cs_error_t error = CS_ERR_NOT_EXIST;
1905
1906 log_printf(LOGSYS_LEVEL_TRACE, "got fragmented mcast request on %p", conn);
1907 log_printf(LOGSYS_LEVEL_DEBUG, "Sending fragmented message size = %d bytes\n", msglen);
1908
1909 switch (cpd->cpd_state) {
1910 case CPD_STATE_UNJOINED:
1911 error = CS_ERR_NOT_EXIST;
1912 break;
1913 case CPD_STATE_LEAVE_STARTED:
1914 error = CS_ERR_NOT_EXIST;
1915 break;
1916 case CPD_STATE_JOIN_STARTED:
1917 error = CS_OK;
1918 break;
1919 case CPD_STATE_JOIN_COMPLETED:
1920 error = CS_OK;
1921 break;
1922 }
1923
1924 res_lib_cpg_partial_send.header.size = sizeof(res_lib_cpg_partial_send);
1925 res_lib_cpg_partial_send.header.id = MESSAGE_RES_CPG_PARTIAL_SEND;
1926
1927 if (req_lib_cpg_mcast->type == LIBCPG_PARTIAL_FIRST) {
1928 cpd->initial_transition_counter = cpd->transition_counter;
1929 }
1930 if (cpd->transition_counter != cpd->initial_transition_counter) {
1931 error = CS_ERR_INTERRUPT;
1932 }
1933
1934 if (error == CS_OK) {
1935 req_exec_cpg_mcast.header.size = sizeof(req_exec_cpg_mcast) + msglen;
1936 req_exec_cpg_mcast.header.id = SERVICE_ID_MAKE(CPG_SERVICE,
1937 MESSAGE_REQ_EXEC_CPG_PARTIAL_MCAST);
1938 req_exec_cpg_mcast.pid = cpd->pid;
1939 req_exec_cpg_mcast.msglen = req_lib_cpg_mcast->msglen;
1940 req_exec_cpg_mcast.type = req_lib_cpg_mcast->type;
1941 req_exec_cpg_mcast.fraglen = req_lib_cpg_mcast->fraglen;
1942 api->ipc_source_set (&req_exec_cpg_mcast.source, conn);
1943 memcpy(&req_exec_cpg_mcast.group_name, &group_name,
1944 sizeof(mar_cpg_name_t));
1945
1946 req_exec_cpg_iovec[0].iov_base = (char *)&req_exec_cpg_mcast;
1947 req_exec_cpg_iovec[0].iov_len = sizeof(req_exec_cpg_mcast);
1948 req_exec_cpg_iovec[1].iov_base = (char *)&req_lib_cpg_mcast->message;
1949 req_exec_cpg_iovec[1].iov_len = msglen;
1950
1951 result = api->totem_mcast (req_exec_cpg_iovec, 2, TOTEM_AGREED);
1952 assert(result == 0);
1953 } else {
1954 log_printf(LOGSYS_LEVEL_ERROR, "*** %p can't mcast to group %s state:%d, error:%d",
1955 conn, group_name.value, cpd->cpd_state, error);
1956 }
1957
1958 res_lib_cpg_partial_send.header.error = error;
1959 api->ipc_response_send (conn, &res_lib_cpg_partial_send,
1960 sizeof (res_lib_cpg_partial_send));
1961 }
1962
1963 /* Mcast message from the library */
1964 static void message_handler_req_lib_cpg_mcast (void *conn, const void *message)
1965 {
1966 const struct req_lib_cpg_mcast *req_lib_cpg_mcast = message;
1967 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
1968 mar_cpg_name_t group_name = cpd->group_name;
1969
1970 struct iovec req_exec_cpg_iovec[2];
1971 struct req_exec_cpg_mcast req_exec_cpg_mcast;
1972 int msglen = req_lib_cpg_mcast->msglen;
1973 int result;
1974 cs_error_t error = CS_ERR_NOT_EXIST;
1975
1976 log_printf(LOGSYS_LEVEL_TRACE, "got mcast request on %p", conn);
1977
1978 switch (cpd->cpd_state) {
1979 case CPD_STATE_UNJOINED:
1980 error = CS_ERR_NOT_EXIST;
1981 break;
1982 case CPD_STATE_LEAVE_STARTED:
1983 error = CS_ERR_NOT_EXIST;
1984 break;
1985 case CPD_STATE_JOIN_STARTED:
1986 error = CS_OK;
1987 break;
1988 case CPD_STATE_JOIN_COMPLETED:
1989 error = CS_OK;
1990 break;
1991 }
1992
1993 if (error == CS_OK) {
1994 memset(&req_exec_cpg_mcast, 0, sizeof(req_exec_cpg_mcast));
1995
1996 req_exec_cpg_mcast.header.size = sizeof(req_exec_cpg_mcast) + msglen;
1997 req_exec_cpg_mcast.header.id = SERVICE_ID_MAKE(CPG_SERVICE,
1998 MESSAGE_REQ_EXEC_CPG_MCAST);
1999 req_exec_cpg_mcast.pid = cpd->pid;
2000 req_exec_cpg_mcast.msglen = msglen;
2001 api->ipc_source_set (&req_exec_cpg_mcast.source, conn);
2002 memcpy(&req_exec_cpg_mcast.group_name, &group_name,
2003 sizeof(mar_cpg_name_t));
2004
2005 req_exec_cpg_iovec[0].iov_base = (char *)&req_exec_cpg_mcast;
2006 req_exec_cpg_iovec[0].iov_len = sizeof(req_exec_cpg_mcast);
2007 req_exec_cpg_iovec[1].iov_base = (char *)&req_lib_cpg_mcast->message;
2008 req_exec_cpg_iovec[1].iov_len = msglen;
2009
2010 result = api->totem_mcast (req_exec_cpg_iovec, 2, TOTEM_AGREED);
2011 assert(result == 0);
2012 } else {
2013 log_printf(LOGSYS_LEVEL_ERROR, "*** %p can't mcast to group %s state:%d, error:%d",
2014 conn, group_name.value, cpd->cpd_state, error);
2015 }
2016 }
2017
2018 static void message_handler_req_lib_cpg_zc_execute (
2019 void *conn,
2020 const void *message)
2021 {
2022 mar_req_coroipcc_zc_execute_t *hdr = (mar_req_coroipcc_zc_execute_t *)message;
2023 struct qb_ipc_request_header *header;
2024 struct res_lib_cpg_mcast res_lib_cpg_mcast;
2025 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
2026 struct iovec req_exec_cpg_iovec[2];
2027 struct req_exec_cpg_mcast req_exec_cpg_mcast;
2028 struct req_lib_cpg_mcast *req_lib_cpg_mcast;
2029 int result;
2030 cs_error_t error = CS_ERR_NOT_EXIST;
2031
2032 log_printf(LOGSYS_LEVEL_TRACE, "got ZC mcast request on %p", conn);
2033
2034 header = (struct qb_ipc_request_header *)(((char *)serveraddr2void(hdr->server_address) + sizeof (struct coroipcs_zc_header)));
2035 req_lib_cpg_mcast = (struct req_lib_cpg_mcast *)header;
2036
2037 switch (cpd->cpd_state) {
2038 case CPD_STATE_UNJOINED:
2039 error = CS_ERR_NOT_EXIST;
2040 break;
2041 case CPD_STATE_LEAVE_STARTED:
2042 error = CS_ERR_NOT_EXIST;
2043 break;
2044 case CPD_STATE_JOIN_STARTED:
2045 error = CS_OK;
2046 break;
2047 case CPD_STATE_JOIN_COMPLETED:
2048 error = CS_OK;
2049 break;
2050 }
2051
2052 res_lib_cpg_mcast.header.size = sizeof(res_lib_cpg_mcast);
2053 res_lib_cpg_mcast.header.id = MESSAGE_RES_CPG_MCAST;
2054 if (error == CS_OK) {
2055 req_exec_cpg_mcast.header.size = sizeof(req_exec_cpg_mcast) + req_lib_cpg_mcast->msglen;
2056 req_exec_cpg_mcast.header.id = SERVICE_ID_MAKE(CPG_SERVICE,
2057 MESSAGE_REQ_EXEC_CPG_MCAST);
2058 req_exec_cpg_mcast.pid = cpd->pid;
2059 req_exec_cpg_mcast.msglen = req_lib_cpg_mcast->msglen;
2060 api->ipc_source_set (&req_exec_cpg_mcast.source, conn);
2061 memcpy(&req_exec_cpg_mcast.group_name, &cpd->group_name,
2062 sizeof(mar_cpg_name_t));
2063
2064 req_exec_cpg_iovec[0].iov_base = (char *)&req_exec_cpg_mcast;
2065 req_exec_cpg_iovec[0].iov_len = sizeof(req_exec_cpg_mcast);
2066 req_exec_cpg_iovec[1].iov_base = (char *)header + sizeof(struct req_lib_cpg_mcast);
2067 req_exec_cpg_iovec[1].iov_len = req_exec_cpg_mcast.msglen;
2068
2069 result = api->totem_mcast (req_exec_cpg_iovec, 2, TOTEM_AGREED);
2070 if (result == 0) {
2071 res_lib_cpg_mcast.header.error = CS_OK;
2072 } else {
2073 res_lib_cpg_mcast.header.error = CS_ERR_TRY_AGAIN;
2074 }
2075 } else {
2076 res_lib_cpg_mcast.header.error = error;
2077 }
2078
2079 api->ipc_response_send (conn, &res_lib_cpg_mcast,
2080 sizeof (res_lib_cpg_mcast));
2081
2082 }
2083
2084 static void message_handler_req_lib_cpg_membership (void *conn,
2085 const void *message)
2086 {
2087 struct req_lib_cpg_membership_get *req_lib_cpg_membership_get =
2088 (struct req_lib_cpg_membership_get *)message;
2089 struct res_lib_cpg_membership_get res_lib_cpg_membership_get;
2090 struct qb_list_head *iter;
2091 int member_count = 0;
2092
2093 res_lib_cpg_membership_get.header.id = MESSAGE_RES_CPG_MEMBERSHIP;
2094 res_lib_cpg_membership_get.header.error = CS_OK;
2095 res_lib_cpg_membership_get.header.size =
2096 sizeof (struct res_lib_cpg_membership_get);
2097
2098 qb_list_for_each(iter, &process_info_list_head) {
2099 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
2100 if (mar_name_compare (&pi->group, &req_lib_cpg_membership_get->group_name) == 0) {
2101 res_lib_cpg_membership_get.member_list[member_count].nodeid = pi->nodeid;
2102 res_lib_cpg_membership_get.member_list[member_count].pid = pi->pid;
2103 member_count += 1;
2104 }
2105 }
2106 res_lib_cpg_membership_get.member_count = member_count;
2107
2108 api->ipc_response_send (conn, &res_lib_cpg_membership_get,
2109 sizeof (res_lib_cpg_membership_get));
2110 }
2111
2112 static void message_handler_req_lib_cpg_local_get (void *conn,
2113 const void *message)
2114 {
2115 struct res_lib_cpg_local_get res_lib_cpg_local_get;
2116
2117 res_lib_cpg_local_get.header.size = sizeof (res_lib_cpg_local_get);
2118 res_lib_cpg_local_get.header.id = MESSAGE_RES_CPG_LOCAL_GET;
2119 res_lib_cpg_local_get.header.error = CS_OK;
2120 res_lib_cpg_local_get.local_nodeid = api->totem_nodeid_get ();
2121
2122 api->ipc_response_send (conn, &res_lib_cpg_local_get,
2123 sizeof (res_lib_cpg_local_get));
2124 }
2125
2126 static void message_handler_req_lib_cpg_iteration_initialize (
2127 void *conn,
2128 const void *message)
2129 {
2130 const struct req_lib_cpg_iterationinitialize *req_lib_cpg_iterationinitialize = message;
2131 struct cpg_pd *cpd = (struct cpg_pd *)api->ipc_private_data_get (conn);
2132 hdb_handle_t cpg_iteration_handle = 0;
2133 struct res_lib_cpg_iterationinitialize res_lib_cpg_iterationinitialize;
2134 struct qb_list_head *iter, *iter2;
2135 struct cpg_iteration_instance *cpg_iteration_instance;
2136 cs_error_t error = CS_OK;
2137 int res;
2138
2139 log_printf (LOGSYS_LEVEL_DEBUG, "cpg iteration initialize");
2140
2141 /* Because between calling this function and *next can be some operations which will
2142 * change list, we must do full copy.
2143 */
2144
2145 /*
2146 * Create new iteration instance
2147 */
2148 res = hdb_handle_create (&cpg_iteration_handle_t_db, sizeof (struct cpg_iteration_instance),
2149 &cpg_iteration_handle);
2150
2151 if (res != 0) {
2152 error = CS_ERR_NO_MEMORY;
2153 goto response_send;
2154 }
2155
2156 res = hdb_handle_get (&cpg_iteration_handle_t_db, cpg_iteration_handle, (void *)&cpg_iteration_instance);
2157
2158 if (res != 0) {
2159 error = CS_ERR_BAD_HANDLE;
2160 goto error_destroy;
2161 }
2162
2163 qb_list_init (&cpg_iteration_instance->items_list_head);
2164 cpg_iteration_instance->handle = cpg_iteration_handle;
2165
2166 /*
2167 * Create copy of process_info list "grouped by" group name
2168 */
2169 qb_list_for_each(iter, &process_info_list_head) {
2170 struct process_info *pi = qb_list_entry (iter, struct process_info, list);
2171 struct process_info *new_pi;
2172
2173 if (req_lib_cpg_iterationinitialize->iteration_type == CPG_ITERATION_NAME_ONLY) {
2174 /*
2175 * Try to find processed group name in our list new list
2176 */
2177 int found = 0;
2178
2179 qb_list_for_each(iter2, &(cpg_iteration_instance->items_list_head)) {
2180 struct process_info *pi2 = qb_list_entry (iter2, struct process_info, list);
2181
2182 if (mar_name_compare (&pi2->group, &pi->group) == 0) {
2183 found = 1;
2184 break;
2185 }
2186 }
2187
2188 if (found) {
2189 /*
2190 * We have this name in list -> don't add
2191 */
2192 continue ;
2193 }
2194 } else if (req_lib_cpg_iterationinitialize->iteration_type == CPG_ITERATION_ONE_GROUP) {
2195 /*
2196 * Test pi group name with request
2197 */
2198 if (mar_name_compare (&pi->group, &req_lib_cpg_iterationinitialize->group_name) != 0)
2199 /*
2200 * Not same -> don't add
2201 */
2202 continue ;
2203 }
2204
2205 new_pi = malloc (sizeof (struct process_info));
2206 if (!new_pi) {
2207 log_printf(LOGSYS_LEVEL_WARNING, "Unable to allocate process_info struct");
2208
2209 error = CS_ERR_NO_MEMORY;
2210
2211 goto error_put_destroy;
2212 }
2213
2214 memcpy (new_pi, pi, sizeof (struct process_info));
2215 qb_list_init (&new_pi->list);
2216
2217 if (req_lib_cpg_iterationinitialize->iteration_type == CPG_ITERATION_NAME_ONLY) {
2218 /*
2219 * pid and nodeid -> undefined
2220 */
2221 new_pi->pid = new_pi->nodeid = 0;
2222 }
2223
2224 /*
2225 * We will return list "grouped" by "group name", so try to find right place to add
2226 */
2227 qb_list_for_each(iter2, &(cpg_iteration_instance->items_list_head)) {
2228 struct process_info *pi2 = qb_list_entry (iter2, struct process_info, list);
2229
2230 if (mar_name_compare (&pi2->group, &pi->group) == 0) {
2231 break;
2232 }
2233 }
2234
2235 qb_list_add (&new_pi->list, iter2);
2236 }
2237
2238 /*
2239 * Now we have a full "grouped by" copy of process_info list
2240 */
2241
2242 /*
2243 * Add instance to current cpd list
2244 */
2245 qb_list_init (&cpg_iteration_instance->list);
2246 qb_list_add (&cpg_iteration_instance->list, &cpd->iteration_instance_list_head);
2247
2248 cpg_iteration_instance->current_pointer = &cpg_iteration_instance->items_list_head;
2249
2250 error_put_destroy:
2251 hdb_handle_put (&cpg_iteration_handle_t_db, cpg_iteration_handle);
2252 error_destroy:
2253 if (error != CS_OK) {
2254 hdb_handle_destroy (&cpg_iteration_handle_t_db, cpg_iteration_handle);
2255 }
2256
2257 response_send:
2258 res_lib_cpg_iterationinitialize.header.size = sizeof (res_lib_cpg_iterationinitialize);
2259 res_lib_cpg_iterationinitialize.header.id = MESSAGE_RES_CPG_ITERATIONINITIALIZE;
2260 res_lib_cpg_iterationinitialize.header.error = error;
2261 res_lib_cpg_iterationinitialize.iteration_handle = cpg_iteration_handle;
2262
2263 api->ipc_response_send (conn, &res_lib_cpg_iterationinitialize,
2264 sizeof (res_lib_cpg_iterationinitialize));
2265 }
2266
2267 static void message_handler_req_lib_cpg_iteration_next (
2268 void *conn,
2269 const void *message)
2270 {
2271 const struct req_lib_cpg_iterationnext *req_lib_cpg_iterationnext = message;
2272 struct res_lib_cpg_iterationnext res_lib_cpg_iterationnext;
2273 struct cpg_iteration_instance *cpg_iteration_instance;
2274 cs_error_t error = CS_OK;
2275 int res;
2276 struct process_info *pi;
2277
2278 log_printf (LOGSYS_LEVEL_DEBUG, "cpg iteration next");
2279
2280 res = hdb_handle_get (&cpg_iteration_handle_t_db,
2281 req_lib_cpg_iterationnext->iteration_handle,
2282 (void *)&cpg_iteration_instance);
2283
2284 if (res != 0) {
2285 error = CS_ERR_LIBRARY;
2286 goto error_exit;
2287 }
2288
2289 assert (cpg_iteration_instance);
2290
2291 cpg_iteration_instance->current_pointer = cpg_iteration_instance->current_pointer->next;
2292
2293 if (cpg_iteration_instance->current_pointer == &cpg_iteration_instance->items_list_head) {
2294 error = CS_ERR_NO_SECTIONS;
2295 goto error_put;
2296 }
2297
2298 pi = qb_list_entry (cpg_iteration_instance->current_pointer, struct process_info, list);
2299
2300 /*
2301 * Copy iteration data
2302 */
2303 res_lib_cpg_iterationnext.description.nodeid = pi->nodeid;
2304 res_lib_cpg_iterationnext.description.pid = pi->pid;
2305 memcpy (&res_lib_cpg_iterationnext.description.group,
2306 &pi->group,
2307 sizeof (mar_cpg_name_t));
2308
2309 error_put:
2310 hdb_handle_put (&cpg_iteration_handle_t_db, req_lib_cpg_iterationnext->iteration_handle);
2311 error_exit:
2312 res_lib_cpg_iterationnext.header.size = sizeof (res_lib_cpg_iterationnext);
2313 res_lib_cpg_iterationnext.header.id = MESSAGE_RES_CPG_ITERATIONNEXT;
2314 res_lib_cpg_iterationnext.header.error = error;
2315
2316 api->ipc_response_send (conn, &res_lib_cpg_iterationnext,
2317 sizeof (res_lib_cpg_iterationnext));
2318 }
2319
2320 static void message_handler_req_lib_cpg_iteration_finalize (
2321 void *conn,
2322 const void *message)
2323 {
2324 const struct req_lib_cpg_iterationfinalize *req_lib_cpg_iterationfinalize = message;
2325 struct res_lib_cpg_iterationfinalize res_lib_cpg_iterationfinalize;
2326 struct cpg_iteration_instance *cpg_iteration_instance;
2327 cs_error_t error = CS_OK;
2328 int res;
2329
2330 log_printf (LOGSYS_LEVEL_DEBUG, "cpg iteration finalize");
2331
2332 res = hdb_handle_get (&cpg_iteration_handle_t_db,
2333 req_lib_cpg_iterationfinalize->iteration_handle,
2334 (void *)&cpg_iteration_instance);
2335
2336 if (res != 0) {
2337 error = CS_ERR_LIBRARY;
2338 goto error_exit;
2339 }
2340
2341 assert (cpg_iteration_instance);
2342
2343 cpg_iteration_instance_finalize (cpg_iteration_instance);
2344 hdb_handle_put (&cpg_iteration_handle_t_db, cpg_iteration_instance->handle);
2345
2346 error_exit:
2347 res_lib_cpg_iterationfinalize.header.size = sizeof (res_lib_cpg_iterationfinalize);
2348 res_lib_cpg_iterationfinalize.header.id = MESSAGE_RES_CPG_ITERATIONFINALIZE;
2349 res_lib_cpg_iterationfinalize.header.error = error;
2350
2351 api->ipc_response_send (conn, &res_lib_cpg_iterationfinalize,
2352 sizeof (res_lib_cpg_iterationfinalize));
2353 }
2354