1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) 2005-2011 Red Hat, Inc. All rights reserved.
6 ** This copyrighted material is made available to anyone wishing to use,
7 ** modify, copy, or redistribute it subject to the terms and conditions
8 ** of the GNU General Public License v.2.
10 *******************************************************************************
11 ******************************************************************************/
13 #include "dlm_internal.h"
14 #include "lockspace.h"
22 int dlm_slots_version(struct dlm_header
*h
)
24 if ((h
->h_version
& 0x0000FFFF) < DLM_HEADER_SLOTS
)
29 void dlm_slot_save(struct dlm_ls
*ls
, struct dlm_rcom
*rc
,
30 struct dlm_member
*memb
)
32 struct rcom_config
*rf
= (struct rcom_config
*)rc
->rc_buf
;
34 if (!dlm_slots_version(&rc
->rc_header
))
37 memb
->slot
= le16_to_cpu(rf
->rf_our_slot
);
38 memb
->generation
= le32_to_cpu(rf
->rf_generation
);
41 void dlm_slots_copy_out(struct dlm_ls
*ls
, struct dlm_rcom
*rc
)
43 struct dlm_slot
*slot
;
47 ro
= (struct rcom_slot
*)(rc
->rc_buf
+ sizeof(struct rcom_config
));
49 /* ls_slots array is sparse, but not rcom_slots */
51 for (i
= 0; i
< ls
->ls_slots_size
; i
++) {
52 slot
= &ls
->ls_slots
[i
];
55 ro
->ro_nodeid
= cpu_to_le32(slot
->nodeid
);
56 ro
->ro_slot
= cpu_to_le16(slot
->slot
);
61 #define SLOT_DEBUG_LINE 128
63 static void log_slots(struct dlm_ls
*ls
, uint32_t gen
, int num_slots
,
64 struct rcom_slot
*ro0
, struct dlm_slot
*array
,
67 char line
[SLOT_DEBUG_LINE
];
68 int len
= SLOT_DEBUG_LINE
- 1;
72 memset(line
, 0, sizeof(line
));
75 for (i
= 0; i
< array_size
; i
++) {
79 ret
= snprintf(line
+ pos
, len
- pos
, " %d:%d",
80 array
[i
].slot
, array
[i
].nodeid
);
86 for (i
= 0; i
< num_slots
; i
++) {
87 ret
= snprintf(line
+ pos
, len
- pos
, " %d:%d",
88 ro0
[i
].ro_slot
, ro0
[i
].ro_nodeid
);
95 log_rinfo(ls
, "generation %u slots %d%s", gen
, num_slots
, line
);
98 int dlm_slots_copy_in(struct dlm_ls
*ls
)
100 struct dlm_member
*memb
;
101 struct dlm_rcom
*rc
= ls
->ls_recover_buf
;
102 struct rcom_config
*rf
= (struct rcom_config
*)rc
->rc_buf
;
103 struct rcom_slot
*ro0
, *ro
;
104 int our_nodeid
= dlm_our_nodeid();
108 if (!dlm_slots_version(&rc
->rc_header
))
111 gen
= le32_to_cpu(rf
->rf_generation
);
112 if (gen
<= ls
->ls_generation
) {
113 log_error(ls
, "dlm_slots_copy_in gen %u old %u",
114 gen
, ls
->ls_generation
);
116 ls
->ls_generation
= gen
;
118 num_slots
= le16_to_cpu(rf
->rf_num_slots
);
122 ro0
= (struct rcom_slot
*)(rc
->rc_buf
+ sizeof(struct rcom_config
));
124 for (i
= 0, ro
= ro0
; i
< num_slots
; i
++, ro
++) {
125 ro
->ro_nodeid
= le32_to_cpu(ro
->ro_nodeid
);
126 ro
->ro_slot
= le16_to_cpu(ro
->ro_slot
);
129 log_slots(ls
, gen
, num_slots
, ro0
, NULL
, 0);
131 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
132 for (i
= 0, ro
= ro0
; i
< num_slots
; i
++, ro
++) {
133 if (ro
->ro_nodeid
!= memb
->nodeid
)
135 memb
->slot
= ro
->ro_slot
;
136 memb
->slot_prev
= memb
->slot
;
140 if (memb
->nodeid
== our_nodeid
) {
141 if (ls
->ls_slot
&& ls
->ls_slot
!= memb
->slot
) {
142 log_error(ls
, "dlm_slots_copy_in our slot "
143 "changed %d %d", ls
->ls_slot
,
149 ls
->ls_slot
= memb
->slot
;
153 log_error(ls
, "dlm_slots_copy_in nodeid %d no slot",
162 /* for any nodes that do not support slots, we will not have set memb->slot
163 in wait_status_all(), so memb->slot will remain -1, and we will not
164 assign slots or set ls_num_slots here */
166 int dlm_slots_assign(struct dlm_ls
*ls
, int *num_slots
, int *slots_size
,
167 struct dlm_slot
**slots_out
, uint32_t *gen_out
)
169 struct dlm_member
*memb
;
170 struct dlm_slot
*array
;
171 int our_nodeid
= dlm_our_nodeid();
172 int array_size
, max_slots
, i
;
178 /* our own memb struct will have slot -1 gen 0 */
180 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
181 if (memb
->nodeid
== our_nodeid
) {
182 memb
->slot
= ls
->ls_slot
;
183 memb
->generation
= ls
->ls_generation
;
188 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
189 if (memb
->generation
> gen
)
190 gen
= memb
->generation
;
192 /* node doesn't support slots */
194 if (memb
->slot
== -1)
197 /* node needs a slot assigned */
202 /* node has a slot assigned */
206 if (!max
|| max
< memb
->slot
)
209 /* sanity check, once slot is assigned it shouldn't change */
211 if (memb
->slot_prev
&& memb
->slot
&& memb
->slot_prev
!= memb
->slot
) {
212 log_error(ls
, "nodeid %d slot changed %d %d",
213 memb
->nodeid
, memb
->slot_prev
, memb
->slot
);
216 memb
->slot_prev
= memb
->slot
;
219 array_size
= max
+ need
;
221 array
= kzalloc(array_size
* sizeof(struct dlm_slot
), GFP_NOFS
);
227 /* fill in slots (offsets) that are used */
229 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
233 if (memb
->slot
> array_size
) {
234 log_error(ls
, "invalid slot number %d", memb
->slot
);
239 array
[memb
->slot
- 1].nodeid
= memb
->nodeid
;
240 array
[memb
->slot
- 1].slot
= memb
->slot
;
244 /* assign new slots from unused offsets */
246 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
250 for (i
= 0; i
< array_size
; i
++) {
255 memb
->slot_prev
= memb
->slot
;
256 array
[i
].nodeid
= memb
->nodeid
;
257 array
[i
].slot
= memb
->slot
;
260 if (!ls
->ls_slot
&& memb
->nodeid
== our_nodeid
)
261 ls
->ls_slot
= memb
->slot
;
266 log_error(ls
, "no free slot found");
274 log_slots(ls
, gen
, num
, NULL
, array
, array_size
);
276 max_slots
= (dlm_config
.ci_buffer_size
- sizeof(struct dlm_rcom
) -
277 sizeof(struct rcom_config
)) / sizeof(struct rcom_slot
);
279 if (num
> max_slots
) {
280 log_error(ls
, "num_slots %d exceeds max_slots %d",
288 *slots_size
= array_size
;
293 static void add_ordered_member(struct dlm_ls
*ls
, struct dlm_member
*new)
295 struct dlm_member
*memb
= NULL
;
296 struct list_head
*tmp
;
297 struct list_head
*newlist
= &new->list
;
298 struct list_head
*head
= &ls
->ls_nodes
;
300 list_for_each(tmp
, head
) {
301 memb
= list_entry(tmp
, struct dlm_member
, list
);
302 if (new->nodeid
< memb
->nodeid
)
307 list_add_tail(newlist
, head
);
309 /* FIXME: can use list macro here */
310 newlist
->prev
= tmp
->prev
;
312 tmp
->prev
->next
= newlist
;
317 static int dlm_add_member(struct dlm_ls
*ls
, struct dlm_config_node
*node
)
319 struct dlm_member
*memb
;
322 memb
= kzalloc(sizeof(struct dlm_member
), GFP_NOFS
);
326 error
= dlm_lowcomms_connect_node(node
->nodeid
);
332 memb
->nodeid
= node
->nodeid
;
333 memb
->weight
= node
->weight
;
334 memb
->comm_seq
= node
->comm_seq
;
335 add_ordered_member(ls
, memb
);
340 static struct dlm_member
*find_memb(struct list_head
*head
, int nodeid
)
342 struct dlm_member
*memb
;
344 list_for_each_entry(memb
, head
, list
) {
345 if (memb
->nodeid
== nodeid
)
351 int dlm_is_member(struct dlm_ls
*ls
, int nodeid
)
353 if (find_memb(&ls
->ls_nodes
, nodeid
))
358 int dlm_is_removed(struct dlm_ls
*ls
, int nodeid
)
360 if (find_memb(&ls
->ls_nodes_gone
, nodeid
))
365 static void clear_memb_list(struct list_head
*head
)
367 struct dlm_member
*memb
;
369 while (!list_empty(head
)) {
370 memb
= list_entry(head
->next
, struct dlm_member
, list
);
371 list_del(&memb
->list
);
376 void dlm_clear_members(struct dlm_ls
*ls
)
378 clear_memb_list(&ls
->ls_nodes
);
379 ls
->ls_num_nodes
= 0;
382 void dlm_clear_members_gone(struct dlm_ls
*ls
)
384 clear_memb_list(&ls
->ls_nodes_gone
);
387 static void make_member_array(struct dlm_ls
*ls
)
389 struct dlm_member
*memb
;
390 int i
, w
, x
= 0, total
= 0, all_zero
= 0, *array
;
392 kfree(ls
->ls_node_array
);
393 ls
->ls_node_array
= NULL
;
395 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
397 total
+= memb
->weight
;
400 /* all nodes revert to weight of 1 if all have weight 0 */
403 total
= ls
->ls_num_nodes
;
407 ls
->ls_total_weight
= total
;
409 array
= kmalloc(sizeof(int) * total
, GFP_NOFS
);
413 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
414 if (!all_zero
&& !memb
->weight
)
422 DLM_ASSERT(x
< total
, printk("total %d x %d\n", total
, x
););
424 for (i
= 0; i
< w
; i
++)
425 array
[x
++] = memb
->nodeid
;
428 ls
->ls_node_array
= array
;
431 /* send a status request to all members just to establish comms connections */
433 static int ping_members(struct dlm_ls
*ls
)
435 struct dlm_member
*memb
;
438 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
439 error
= dlm_recovery_stopped(ls
);
442 error
= dlm_rcom_status(ls
, memb
->nodeid
, 0);
447 log_rinfo(ls
, "ping_members aborted %d last nodeid %d",
448 error
, ls
->ls_recover_nodeid
);
452 static void dlm_lsop_recover_prep(struct dlm_ls
*ls
)
454 if (!ls
->ls_ops
|| !ls
->ls_ops
->recover_prep
)
456 ls
->ls_ops
->recover_prep(ls
->ls_ops_arg
);
459 static void dlm_lsop_recover_slot(struct dlm_ls
*ls
, struct dlm_member
*memb
)
461 struct dlm_slot slot
;
465 if (!ls
->ls_ops
|| !ls
->ls_ops
->recover_slot
)
468 /* if there is no comms connection with this node
469 or the present comms connection is newer
470 than the one when this member was added, then
471 we consider the node to have failed (versus
472 being removed due to dlm_release_lockspace) */
474 error
= dlm_comm_seq(memb
->nodeid
, &seq
);
476 if (!error
&& seq
== memb
->comm_seq
)
479 slot
.nodeid
= memb
->nodeid
;
480 slot
.slot
= memb
->slot
;
482 ls
->ls_ops
->recover_slot(ls
->ls_ops_arg
, &slot
);
485 void dlm_lsop_recover_done(struct dlm_ls
*ls
)
487 struct dlm_member
*memb
;
488 struct dlm_slot
*slots
;
491 if (!ls
->ls_ops
|| !ls
->ls_ops
->recover_done
)
494 num
= ls
->ls_num_nodes
;
496 slots
= kzalloc(num
* sizeof(struct dlm_slot
), GFP_KERNEL
);
501 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
503 log_error(ls
, "dlm_lsop_recover_done bad num %d", num
);
506 slots
[i
].nodeid
= memb
->nodeid
;
507 slots
[i
].slot
= memb
->slot
;
511 ls
->ls_ops
->recover_done(ls
->ls_ops_arg
, slots
, num
,
512 ls
->ls_slot
, ls
->ls_generation
);
517 static struct dlm_config_node
*find_config_node(struct dlm_recover
*rv
,
522 for (i
= 0; i
< rv
->nodes_count
; i
++) {
523 if (rv
->nodes
[i
].nodeid
== nodeid
)
524 return &rv
->nodes
[i
];
529 int dlm_recover_members(struct dlm_ls
*ls
, struct dlm_recover
*rv
, int *neg_out
)
531 struct dlm_member
*memb
, *safe
;
532 struct dlm_config_node
*node
;
533 int i
, error
, neg
= 0, low
= -1;
535 /* previously removed members that we've not finished removing need to
536 count as a negative change so the "neg" recovery steps will happen */
538 list_for_each_entry(memb
, &ls
->ls_nodes_gone
, list
) {
539 log_rinfo(ls
, "prev removed member %d", memb
->nodeid
);
543 /* move departed members from ls_nodes to ls_nodes_gone */
545 list_for_each_entry_safe(memb
, safe
, &ls
->ls_nodes
, list
) {
546 node
= find_config_node(rv
, memb
->nodeid
);
547 if (node
&& !node
->new)
551 log_rinfo(ls
, "remove member %d", memb
->nodeid
);
553 /* removed and re-added */
554 log_rinfo(ls
, "remove member %d comm_seq %u %u",
555 memb
->nodeid
, memb
->comm_seq
, node
->comm_seq
);
559 list_move(&memb
->list
, &ls
->ls_nodes_gone
);
561 dlm_lsop_recover_slot(ls
, memb
);
564 /* add new members to ls_nodes */
566 for (i
= 0; i
< rv
->nodes_count
; i
++) {
567 node
= &rv
->nodes
[i
];
568 if (dlm_is_member(ls
, node
->nodeid
))
570 dlm_add_member(ls
, node
);
571 log_rinfo(ls
, "add member %d", node
->nodeid
);
574 list_for_each_entry(memb
, &ls
->ls_nodes
, list
) {
575 if (low
== -1 || memb
->nodeid
< low
)
578 ls
->ls_low_nodeid
= low
;
580 make_member_array(ls
);
583 error
= ping_members(ls
);
584 if (!error
|| error
== -EPROTO
) {
585 /* new_lockspace() may be waiting to know if the config
587 ls
->ls_members_result
= error
;
588 complete(&ls
->ls_members_done
);
591 log_rinfo(ls
, "dlm_recover_members %d nodes", ls
->ls_num_nodes
);
595 /* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
596 dlm_ls_start() is called on any of them to start the new recovery. */
598 int dlm_ls_stop(struct dlm_ls
*ls
)
603 * Prevent dlm_recv from being in the middle of something when we do
604 * the stop. This includes ensuring dlm_recv isn't processing a
605 * recovery message (rcom), while dlm_recoverd is aborting and
606 * resetting things from an in-progress recovery. i.e. we want
607 * dlm_recoverd to abort its recovery without worrying about dlm_recv
608 * processing an rcom at the same time. Stopping dlm_recv also makes
609 * it easy for dlm_receive_message() to check locking stopped and add a
610 * message to the requestqueue without races.
613 down_write(&ls
->ls_recv_active
);
616 * Abort any recovery that's in progress (see RECOVER_STOP,
617 * dlm_recovery_stopped()) and tell any other threads running in the
618 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
621 spin_lock(&ls
->ls_recover_lock
);
622 set_bit(LSFL_RECOVER_STOP
, &ls
->ls_flags
);
623 new = test_and_clear_bit(LSFL_RUNNING
, &ls
->ls_flags
);
624 ls
->ls_recover_seq
++;
625 spin_unlock(&ls
->ls_recover_lock
);
628 * Let dlm_recv run again, now any normal messages will be saved on the
629 * requestqueue for later.
632 up_write(&ls
->ls_recv_active
);
635 * This in_recovery lock does two things:
636 * 1) Keeps this function from returning until all threads are out
637 * of locking routines and locking is truly stopped.
638 * 2) Keeps any new requests from being processed until it's unlocked
639 * when recovery is complete.
643 set_bit(LSFL_RECOVER_DOWN
, &ls
->ls_flags
);
644 wake_up_process(ls
->ls_recoverd_task
);
645 wait_event(ls
->ls_recover_lock_wait
,
646 test_bit(LSFL_RECOVER_LOCK
, &ls
->ls_flags
));
650 * The recoverd suspend/resume makes sure that dlm_recoverd (if
651 * running) has noticed RECOVER_STOP above and quit processing the
655 dlm_recoverd_suspend(ls
);
657 spin_lock(&ls
->ls_recover_lock
);
660 ls
->ls_num_slots
= 0;
661 ls
->ls_slots_size
= 0;
662 ls
->ls_recover_status
= 0;
663 spin_unlock(&ls
->ls_recover_lock
);
665 dlm_recoverd_resume(ls
);
667 if (!ls
->ls_recover_begin
)
668 ls
->ls_recover_begin
= jiffies
;
670 dlm_lsop_recover_prep(ls
);
674 int dlm_ls_start(struct dlm_ls
*ls
)
676 struct dlm_recover
*rv
= NULL
, *rv_old
;
677 struct dlm_config_node
*nodes
;
680 rv
= kzalloc(sizeof(struct dlm_recover
), GFP_NOFS
);
684 error
= dlm_config_nodes(ls
->ls_name
, &nodes
, &count
);
688 spin_lock(&ls
->ls_recover_lock
);
690 /* the lockspace needs to be stopped before it can be started */
692 if (!dlm_locking_stopped(ls
)) {
693 spin_unlock(&ls
->ls_recover_lock
);
694 log_error(ls
, "start ignored: lockspace running");
700 rv
->nodes_count
= count
;
701 rv
->seq
= ++ls
->ls_recover_seq
;
702 rv_old
= ls
->ls_recover_args
;
703 ls
->ls_recover_args
= rv
;
704 spin_unlock(&ls
->ls_recover_lock
);
707 log_error(ls
, "unused recovery %llx %d",
708 (unsigned long long)rv_old
->seq
, rv_old
->nodes_count
);
709 kfree(rv_old
->nodes
);
713 set_bit(LSFL_RECOVER_WORK
, &ls
->ls_flags
);
714 wake_up_process(ls
->ls_recoverd_task
);