blob: 4a7a76e42fc365e603e5ec9af26089ef9fe7d727 [file] [log] [blame]
David Teiglande7fd4172006-01-18 09:30:29 +00001/******************************************************************************
2*******************************************************************************
3**
4** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5** Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
6**
7** This copyrighted material is made available to anyone wishing to use,
8** modify, copy, or redistribute it subject to the terms and conditions
9** of the GNU General Public License v.2.
10**
11*******************************************************************************
12******************************************************************************/
13
14#include "dlm_internal.h"
15#include "lockspace.h"
16#include "dir.h"
17#include "config.h"
18#include "ast.h"
19#include "memory.h"
20#include "rcom.h"
21#include "lock.h"
22#include "lowcomms.h"
23#include "member.h"
24#include "recover.h"
25
26
27/*
28 * Recovery waiting routines: these functions wait for a particular reply from
29 * a remote node, or for the remote node to report a certain status. They need
30 * to abort if the lockspace is stopped indicating a node has failed (perhaps
31 * the one being waited for).
32 */
33
34/*
35 * Wait until given function returns non-zero or lockspace is stopped
36 * (LS_RECOVERY_STOP set due to failure of a node in ls_nodes). When another
37 * function thinks it could have completed the waited-on task, they should wake
38 * up ls_wait_general to get an immediate response rather than waiting for the
David Teigland6d768172012-06-05 11:23:21 -050039 * timeout. This uses a timeout so it can check periodically if the wait
40 * should abort due to node failure (which doesn't cause a wake_up).
41 * This should only be called by the dlm_recoverd thread.
David Teiglande7fd4172006-01-18 09:30:29 +000042 */
43
David Teiglande7fd4172006-01-18 09:30:29 +000044int dlm_wait_function(struct dlm_ls *ls, int (*testfn) (struct dlm_ls *ls))
45{
46 int error = 0;
David Teigland6d768172012-06-05 11:23:21 -050047 int rv;
David Teiglande7fd4172006-01-18 09:30:29 +000048
David Teigland6d768172012-06-05 11:23:21 -050049 while (1) {
50 rv = wait_event_timeout(ls->ls_wait_general,
51 testfn(ls) || dlm_recovery_stopped(ls),
52 dlm_config.ci_recover_timer * HZ);
53 if (rv)
54 break;
55 }
David Teiglande7fd4172006-01-18 09:30:29 +000056
57 if (dlm_recovery_stopped(ls)) {
58 log_debug(ls, "dlm_wait_function aborted");
59 error = -EINTR;
60 }
61 return error;
62}
63
64/*
65 * An efficient way for all nodes to wait for all others to have a certain
66 * status. The node with the lowest nodeid polls all the others for their
67 * status (wait_status_all) and all the others poll the node with the low id
68 * for its accumulated result (wait_status_low). When all nodes have set
69 * status flag X, then status flag X_ALL will be set on the low nodeid.
70 */
71
72uint32_t dlm_recover_status(struct dlm_ls *ls)
73{
74 uint32_t status;
75 spin_lock(&ls->ls_recover_lock);
76 status = ls->ls_recover_status;
77 spin_unlock(&ls->ls_recover_lock);
78 return status;
79}
80
David Teigland757a4272011-10-20 13:26:28 -050081static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
82{
83 ls->ls_recover_status |= status;
84}
85
David Teiglande7fd4172006-01-18 09:30:29 +000086void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status)
87{
88 spin_lock(&ls->ls_recover_lock);
David Teigland757a4272011-10-20 13:26:28 -050089 _set_recover_status(ls, status);
David Teiglande7fd4172006-01-18 09:30:29 +000090 spin_unlock(&ls->ls_recover_lock);
91}
92
David Teigland757a4272011-10-20 13:26:28 -050093static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status,
94 int save_slots)
David Teiglande7fd4172006-01-18 09:30:29 +000095{
Al Viro40076852008-01-25 03:01:51 -050096 struct dlm_rcom *rc = ls->ls_recover_buf;
David Teiglande7fd4172006-01-18 09:30:29 +000097 struct dlm_member *memb;
98 int error = 0, delay;
99
100 list_for_each_entry(memb, &ls->ls_nodes, list) {
101 delay = 0;
102 for (;;) {
103 if (dlm_recovery_stopped(ls)) {
104 error = -EINTR;
105 goto out;
106 }
107
David Teigland757a4272011-10-20 13:26:28 -0500108 error = dlm_rcom_status(ls, memb->nodeid, 0);
David Teiglande7fd4172006-01-18 09:30:29 +0000109 if (error)
110 goto out;
111
David Teigland757a4272011-10-20 13:26:28 -0500112 if (save_slots)
113 dlm_slot_save(ls, rc, memb);
114
David Teiglande7fd4172006-01-18 09:30:29 +0000115 if (rc->rc_result & wait_status)
116 break;
117 if (delay < 1000)
118 delay += 20;
119 msleep(delay);
120 }
121 }
122 out:
123 return error;
124}
125
David Teigland757a4272011-10-20 13:26:28 -0500126static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status,
127 uint32_t status_flags)
David Teiglande7fd4172006-01-18 09:30:29 +0000128{
Al Viro40076852008-01-25 03:01:51 -0500129 struct dlm_rcom *rc = ls->ls_recover_buf;
David Teiglande7fd4172006-01-18 09:30:29 +0000130 int error = 0, delay = 0, nodeid = ls->ls_low_nodeid;
131
132 for (;;) {
133 if (dlm_recovery_stopped(ls)) {
134 error = -EINTR;
135 goto out;
136 }
137
David Teigland757a4272011-10-20 13:26:28 -0500138 error = dlm_rcom_status(ls, nodeid, status_flags);
David Teiglande7fd4172006-01-18 09:30:29 +0000139 if (error)
140 break;
141
142 if (rc->rc_result & wait_status)
143 break;
144 if (delay < 1000)
145 delay += 20;
146 msleep(delay);
147 }
148 out:
149 return error;
150}
151
152static int wait_status(struct dlm_ls *ls, uint32_t status)
153{
154 uint32_t status_all = status << 1;
155 int error;
156
157 if (ls->ls_low_nodeid == dlm_our_nodeid()) {
David Teigland757a4272011-10-20 13:26:28 -0500158 error = wait_status_all(ls, status, 0);
David Teiglande7fd4172006-01-18 09:30:29 +0000159 if (!error)
160 dlm_set_recover_status(ls, status_all);
161 } else
David Teigland757a4272011-10-20 13:26:28 -0500162 error = wait_status_low(ls, status_all, 0);
David Teiglande7fd4172006-01-18 09:30:29 +0000163
164 return error;
165}
166
167int dlm_recover_members_wait(struct dlm_ls *ls)
168{
David Teigland757a4272011-10-20 13:26:28 -0500169 struct dlm_member *memb;
170 struct dlm_slot *slots;
171 int num_slots, slots_size;
172 int error, rv;
173 uint32_t gen;
174
175 list_for_each_entry(memb, &ls->ls_nodes, list) {
176 memb->slot = -1;
177 memb->generation = 0;
178 }
179
180 if (ls->ls_low_nodeid == dlm_our_nodeid()) {
181 error = wait_status_all(ls, DLM_RS_NODES, 1);
182 if (error)
183 goto out;
184
185 /* slots array is sparse, slots_size may be > num_slots */
186
187 rv = dlm_slots_assign(ls, &num_slots, &slots_size, &slots, &gen);
188 if (!rv) {
189 spin_lock(&ls->ls_recover_lock);
190 _set_recover_status(ls, DLM_RS_NODES_ALL);
191 ls->ls_num_slots = num_slots;
192 ls->ls_slots_size = slots_size;
193 ls->ls_slots = slots;
194 ls->ls_generation = gen;
195 spin_unlock(&ls->ls_recover_lock);
196 } else {
197 dlm_set_recover_status(ls, DLM_RS_NODES_ALL);
198 }
199 } else {
200 error = wait_status_low(ls, DLM_RS_NODES_ALL, DLM_RSF_NEED_SLOTS);
201 if (error)
202 goto out;
203
204 dlm_slots_copy_in(ls);
205 }
206 out:
207 return error;
David Teiglande7fd4172006-01-18 09:30:29 +0000208}
209
210int dlm_recover_directory_wait(struct dlm_ls *ls)
211{
212 return wait_status(ls, DLM_RS_DIR);
213}
214
215int dlm_recover_locks_wait(struct dlm_ls *ls)
216{
217 return wait_status(ls, DLM_RS_LOCKS);
218}
219
220int dlm_recover_done_wait(struct dlm_ls *ls)
221{
222 return wait_status(ls, DLM_RS_DONE);
223}
224
225/*
226 * The recover_list contains all the rsb's for which we've requested the new
227 * master nodeid. As replies are returned from the resource directories the
228 * rsb's are removed from the list. When the list is empty we're done.
229 *
230 * The recover_list is later similarly used for all rsb's for which we've sent
231 * new lkb's and need to receive new corresponding lkid's.
232 *
233 * We use the address of the rsb struct as a simple local identifier for the
234 * rsb so we can match an rcom reply with the rsb it was sent for.
235 */
236
237static int recover_list_empty(struct dlm_ls *ls)
238{
239 int empty;
240
241 spin_lock(&ls->ls_recover_list_lock);
242 empty = list_empty(&ls->ls_recover_list);
243 spin_unlock(&ls->ls_recover_list_lock);
244
245 return empty;
246}
247
248static void recover_list_add(struct dlm_rsb *r)
249{
250 struct dlm_ls *ls = r->res_ls;
251
252 spin_lock(&ls->ls_recover_list_lock);
253 if (list_empty(&r->res_recover_list)) {
254 list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
255 ls->ls_recover_list_count++;
256 dlm_hold_rsb(r);
257 }
258 spin_unlock(&ls->ls_recover_list_lock);
259}
260
261static void recover_list_del(struct dlm_rsb *r)
262{
263 struct dlm_ls *ls = r->res_ls;
264
265 spin_lock(&ls->ls_recover_list_lock);
266 list_del_init(&r->res_recover_list);
267 ls->ls_recover_list_count--;
268 spin_unlock(&ls->ls_recover_list_lock);
269
270 dlm_put_rsb(r);
271}
272
David Teiglande7fd4172006-01-18 09:30:29 +0000273static void recover_list_clear(struct dlm_ls *ls)
274{
275 struct dlm_rsb *r, *s;
276
277 spin_lock(&ls->ls_recover_list_lock);
278 list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
279 list_del_init(&r->res_recover_list);
David Teigland52069802006-11-02 09:49:02 -0600280 r->res_recover_locks_count = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000281 dlm_put_rsb(r);
282 ls->ls_recover_list_count--;
283 }
284
285 if (ls->ls_recover_list_count != 0) {
286 log_error(ls, "warning: recover_list_count %d",
287 ls->ls_recover_list_count);
288 ls->ls_recover_list_count = 0;
289 }
290 spin_unlock(&ls->ls_recover_list_lock);
291}
292
David Teigland1d7c4842012-05-15 16:07:49 -0500293static int recover_idr_empty(struct dlm_ls *ls)
294{
295 int empty = 1;
296
297 spin_lock(&ls->ls_recover_idr_lock);
298 if (ls->ls_recover_list_count)
299 empty = 0;
300 spin_unlock(&ls->ls_recover_idr_lock);
301
302 return empty;
303}
304
305static int recover_idr_add(struct dlm_rsb *r)
306{
307 struct dlm_ls *ls = r->res_ls;
308 int rv, id;
309
310 rv = idr_pre_get(&ls->ls_recover_idr, GFP_NOFS);
311 if (!rv)
312 return -ENOMEM;
313
314 spin_lock(&ls->ls_recover_idr_lock);
315 if (r->res_id) {
316 spin_unlock(&ls->ls_recover_idr_lock);
317 return -1;
318 }
319 rv = idr_get_new_above(&ls->ls_recover_idr, r, 1, &id);
320 if (rv) {
321 spin_unlock(&ls->ls_recover_idr_lock);
322 return rv;
323 }
324 r->res_id = id;
325 ls->ls_recover_list_count++;
326 dlm_hold_rsb(r);
327 spin_unlock(&ls->ls_recover_idr_lock);
328 return 0;
329}
330
331static void recover_idr_del(struct dlm_rsb *r)
332{
333 struct dlm_ls *ls = r->res_ls;
334
335 spin_lock(&ls->ls_recover_idr_lock);
336 idr_remove(&ls->ls_recover_idr, r->res_id);
337 r->res_id = 0;
338 ls->ls_recover_list_count--;
339 spin_unlock(&ls->ls_recover_idr_lock);
340
341 dlm_put_rsb(r);
342}
343
344static struct dlm_rsb *recover_idr_find(struct dlm_ls *ls, uint64_t id)
345{
346 struct dlm_rsb *r;
347
348 spin_lock(&ls->ls_recover_idr_lock);
349 r = idr_find(&ls->ls_recover_idr, (int)id);
350 spin_unlock(&ls->ls_recover_idr_lock);
351 return r;
352}
353
354static int recover_idr_clear_rsb(int id, void *p, void *data)
355{
356 struct dlm_ls *ls = data;
357 struct dlm_rsb *r = p;
358
359 r->res_id = 0;
360 r->res_recover_locks_count = 0;
361 ls->ls_recover_list_count--;
362
363 dlm_put_rsb(r);
364 return 0;
365}
366
367static void recover_idr_clear(struct dlm_ls *ls)
368{
369 spin_lock(&ls->ls_recover_idr_lock);
370 idr_for_each(&ls->ls_recover_idr, recover_idr_clear_rsb, ls);
371 idr_remove_all(&ls->ls_recover_idr);
372
373 if (ls->ls_recover_list_count != 0) {
374 log_error(ls, "warning: recover_list_count %d",
375 ls->ls_recover_list_count);
376 ls->ls_recover_list_count = 0;
377 }
378 spin_unlock(&ls->ls_recover_idr_lock);
379}
380
David Teiglande7fd4172006-01-18 09:30:29 +0000381
382/* Master recovery: find new master node for rsb's that were
383 mastered on nodes that have been removed.
384
385 dlm_recover_masters
386 recover_master
387 dlm_send_rcom_lookup -> receive_rcom_lookup
388 dlm_dir_lookup
389 receive_rcom_lookup_reply <-
390 dlm_recover_master_reply
391 set_new_master
392 set_master_lkbs
393 set_lock_master
394*/
395
396/*
397 * Set the lock master for all LKBs in a lock queue
398 * If we are the new master of the rsb, we may have received new
399 * MSTCPY locks from other nodes already which we need to ignore
400 * when setting the new nodeid.
401 */
402
403static void set_lock_master(struct list_head *queue, int nodeid)
404{
405 struct dlm_lkb *lkb;
406
David Teigland48756472012-04-26 15:54:29 -0500407 list_for_each_entry(lkb, queue, lkb_statequeue) {
408 if (!(lkb->lkb_flags & DLM_IFL_MSTCPY)) {
David Teiglande7fd4172006-01-18 09:30:29 +0000409 lkb->lkb_nodeid = nodeid;
David Teigland48756472012-04-26 15:54:29 -0500410 lkb->lkb_remid = 0;
411 }
412 }
David Teiglande7fd4172006-01-18 09:30:29 +0000413}
414
415static void set_master_lkbs(struct dlm_rsb *r)
416{
417 set_lock_master(&r->res_grantqueue, r->res_nodeid);
418 set_lock_master(&r->res_convertqueue, r->res_nodeid);
419 set_lock_master(&r->res_waitqueue, r->res_nodeid);
420}
421
422/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300423 * Propagate the new master nodeid to locks
David Teiglande7fd4172006-01-18 09:30:29 +0000424 * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider.
David Teigland48756472012-04-26 15:54:29 -0500425 * The NEW_MASTER2 flag tells recover_lvb() and recover_grant() which
David Teiglandf7da7902006-07-25 13:53:33 -0500426 * rsb's to consider.
David Teiglande7fd4172006-01-18 09:30:29 +0000427 */
428
David Teiglandc04fecb2012-05-10 10:18:07 -0500429static void set_new_master(struct dlm_rsb *r)
David Teiglande7fd4172006-01-18 09:30:29 +0000430{
David Teiglande7fd4172006-01-18 09:30:29 +0000431 set_master_lkbs(r);
432 rsb_set_flag(r, RSB_NEW_MASTER);
433 rsb_set_flag(r, RSB_NEW_MASTER2);
David Teiglande7fd4172006-01-18 09:30:29 +0000434}
435
436/*
437 * We do async lookups on rsb's that need new masters. The rsb's
438 * waiting for a lookup reply are kept on the recover_list.
David Teiglandc04fecb2012-05-10 10:18:07 -0500439 *
440 * Another node recovering the master may have sent us a rcom lookup,
441 * and our dlm_master_lookup() set it as the new master, along with
442 * NEW_MASTER so that we'll recover it here (this implies dir_nodeid
443 * equals our_nodeid below).
David Teiglande7fd4172006-01-18 09:30:29 +0000444 */
445
David Teiglandc04fecb2012-05-10 10:18:07 -0500446static int recover_master(struct dlm_rsb *r, unsigned int *count)
David Teiglande7fd4172006-01-18 09:30:29 +0000447{
448 struct dlm_ls *ls = r->res_ls;
David Teiglandc04fecb2012-05-10 10:18:07 -0500449 int our_nodeid, dir_nodeid;
450 int is_removed = 0;
451 int error;
452
453 if (is_master(r))
454 return 0;
455
456 is_removed = dlm_is_removed(ls, r->res_nodeid);
457
458 if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER))
459 return 0;
460
461 our_nodeid = dlm_our_nodeid();
462 dir_nodeid = dlm_dir_nodeid(r);
David Teiglande7fd4172006-01-18 09:30:29 +0000463
464 if (dir_nodeid == our_nodeid) {
David Teiglandc04fecb2012-05-10 10:18:07 -0500465 if (is_removed) {
466 r->res_master_nodeid = our_nodeid;
467 r->res_nodeid = 0;
468 }
David Teiglande7fd4172006-01-18 09:30:29 +0000469
David Teiglandc04fecb2012-05-10 10:18:07 -0500470 /* set master of lkbs to ourself when is_removed, or to
471 another new master which we set along with NEW_MASTER
472 in dlm_master_lookup */
473 set_new_master(r);
474 error = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000475 } else {
David Teigland1d7c4842012-05-15 16:07:49 -0500476 recover_idr_add(r);
David Teiglande7fd4172006-01-18 09:30:29 +0000477 error = dlm_send_rcom_lookup(r, dir_nodeid);
478 }
479
David Teiglandc04fecb2012-05-10 10:18:07 -0500480 (*count)++;
David Teiglande7fd4172006-01-18 09:30:29 +0000481 return error;
482}
483
484/*
David Teigland48756472012-04-26 15:54:29 -0500485 * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
486 * This is necessary because recovery can be started, aborted and restarted,
487 * causing the master nodeid to briefly change during the aborted recovery, and
488 * change back to the original value in the second recovery. The MSTCPY locks
489 * may or may not have been purged during the aborted recovery. Another node
490 * with an outstanding request in waiters list and a request reply saved in the
491 * requestqueue, cannot know whether it should ignore the reply and resend the
492 * request, or accept the reply and complete the request. It must do the
493 * former if the remote node purged MSTCPY locks, and it must do the later if
494 * the remote node did not. This is solved by always purging MSTCPY locks, in
495 * which case, the request reply would always be ignored and the request
496 * resent.
David Teiglande7fd4172006-01-18 09:30:29 +0000497 */
498
David Teiglandc04fecb2012-05-10 10:18:07 -0500499static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
David Teiglande7fd4172006-01-18 09:30:29 +0000500{
David Teigland48756472012-04-26 15:54:29 -0500501 int dir_nodeid = dlm_dir_nodeid(r);
502 int new_master = dir_nodeid;
David Teiglande7fd4172006-01-18 09:30:29 +0000503
David Teigland48756472012-04-26 15:54:29 -0500504 if (dir_nodeid == dlm_our_nodeid())
505 new_master = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000506
David Teigland48756472012-04-26 15:54:29 -0500507 dlm_purge_mstcpy_locks(r);
David Teiglandc04fecb2012-05-10 10:18:07 -0500508 r->res_master_nodeid = dir_nodeid;
509 r->res_nodeid = new_master;
510 set_new_master(r);
511 (*count)++;
512 return 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000513}
514
515/*
516 * Go through local root resources and for each rsb which has a master which
517 * has departed, get the new master nodeid from the directory. The dir will
518 * assign mastery to the first node to look up the new master. That means
519 * we'll discover in this lookup if we're the new master of any rsb's.
520 *
521 * We fire off all the dir lookup requests individually and asynchronously to
522 * the correct dir node.
523 */
524
525int dlm_recover_masters(struct dlm_ls *ls)
526{
527 struct dlm_rsb *r;
David Teiglandc04fecb2012-05-10 10:18:07 -0500528 unsigned int total = 0;
529 unsigned int count = 0;
530 int nodir = dlm_no_directory(ls);
531 int error;
David Teiglande7fd4172006-01-18 09:30:29 +0000532
533 log_debug(ls, "dlm_recover_masters");
534
535 down_read(&ls->ls_root_sem);
536 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
537 if (dlm_recovery_stopped(ls)) {
538 up_read(&ls->ls_root_sem);
539 error = -EINTR;
540 goto out;
541 }
542
David Teiglandc04fecb2012-05-10 10:18:07 -0500543 lock_rsb(r);
544 if (nodir)
545 error = recover_master_static(r, &count);
546 else
547 error = recover_master(r, &count);
548 unlock_rsb(r);
549 cond_resched();
550 total++;
David Teiglande7fd4172006-01-18 09:30:29 +0000551
David Teiglandc04fecb2012-05-10 10:18:07 -0500552 if (error) {
553 up_read(&ls->ls_root_sem);
554 goto out;
555 }
David Teiglande7fd4172006-01-18 09:30:29 +0000556 }
557 up_read(&ls->ls_root_sem);
558
David Teiglandc04fecb2012-05-10 10:18:07 -0500559 log_debug(ls, "dlm_recover_masters %u of %u", count, total);
David Teiglande7fd4172006-01-18 09:30:29 +0000560
David Teigland1d7c4842012-05-15 16:07:49 -0500561 error = dlm_wait_function(ls, &recover_idr_empty);
David Teiglande7fd4172006-01-18 09:30:29 +0000562 out:
563 if (error)
David Teigland1d7c4842012-05-15 16:07:49 -0500564 recover_idr_clear(ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000565 return error;
566}
567
568int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
569{
570 struct dlm_rsb *r;
David Teiglandc04fecb2012-05-10 10:18:07 -0500571 int ret_nodeid, new_master;
David Teiglande7fd4172006-01-18 09:30:29 +0000572
David Teigland1d7c4842012-05-15 16:07:49 -0500573 r = recover_idr_find(ls, rc->rc_id);
David Teiglande7fd4172006-01-18 09:30:29 +0000574 if (!r) {
David Teigland90135922006-01-20 08:47:07 +0000575 log_error(ls, "dlm_recover_master_reply no id %llx",
David Teigland9229f012006-05-24 09:21:30 -0400576 (unsigned long long)rc->rc_id);
David Teiglande7fd4172006-01-18 09:30:29 +0000577 goto out;
578 }
579
David Teiglandc04fecb2012-05-10 10:18:07 -0500580 ret_nodeid = rc->rc_result;
581
582 if (ret_nodeid == dlm_our_nodeid())
583 new_master = 0;
584 else
585 new_master = ret_nodeid;
David Teiglande7fd4172006-01-18 09:30:29 +0000586
David Teigland48756472012-04-26 15:54:29 -0500587 lock_rsb(r);
David Teiglandc04fecb2012-05-10 10:18:07 -0500588 r->res_master_nodeid = ret_nodeid;
589 r->res_nodeid = new_master;
590 set_new_master(r);
David Teigland48756472012-04-26 15:54:29 -0500591 unlock_rsb(r);
David Teigland1d7c4842012-05-15 16:07:49 -0500592 recover_idr_del(r);
David Teiglande7fd4172006-01-18 09:30:29 +0000593
David Teigland1d7c4842012-05-15 16:07:49 -0500594 if (recover_idr_empty(ls))
David Teiglande7fd4172006-01-18 09:30:29 +0000595 wake_up(&ls->ls_wait_general);
596 out:
597 return 0;
598}
599
600
601/* Lock recovery: rebuild the process-copy locks we hold on a
602 remastered rsb on the new rsb master.
603
604 dlm_recover_locks
605 recover_locks
606 recover_locks_queue
607 dlm_send_rcom_lock -> receive_rcom_lock
608 dlm_recover_master_copy
609 receive_rcom_lock_reply <-
610 dlm_recover_process_copy
611*/
612
613
614/*
615 * keep a count of the number of lkb's we send to the new master; when we get
616 * an equal number of replies then recovery for the rsb is done
617 */
618
619static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head)
620{
621 struct dlm_lkb *lkb;
622 int error = 0;
623
624 list_for_each_entry(lkb, head, lkb_statequeue) {
625 error = dlm_send_rcom_lock(r, lkb);
626 if (error)
627 break;
628 r->res_recover_locks_count++;
629 }
630
631 return error;
632}
633
David Teiglande7fd4172006-01-18 09:30:29 +0000634static int recover_locks(struct dlm_rsb *r)
635{
636 int error = 0;
637
638 lock_rsb(r);
David Teiglande7fd4172006-01-18 09:30:29 +0000639
David Teiglanda345da32006-08-18 11:54:25 -0500640 DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r););
David Teiglande7fd4172006-01-18 09:30:29 +0000641
642 error = recover_locks_queue(r, &r->res_grantqueue);
643 if (error)
644 goto out;
645 error = recover_locks_queue(r, &r->res_convertqueue);
646 if (error)
647 goto out;
648 error = recover_locks_queue(r, &r->res_waitqueue);
649 if (error)
650 goto out;
651
652 if (r->res_recover_locks_count)
653 recover_list_add(r);
654 else
655 rsb_clear_flag(r, RSB_NEW_MASTER);
656 out:
657 unlock_rsb(r);
658 return error;
659}
660
661int dlm_recover_locks(struct dlm_ls *ls)
662{
663 struct dlm_rsb *r;
664 int error, count = 0;
665
David Teiglande7fd4172006-01-18 09:30:29 +0000666 down_read(&ls->ls_root_sem);
667 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
668 if (is_master(r)) {
669 rsb_clear_flag(r, RSB_NEW_MASTER);
670 continue;
671 }
672
673 if (!rsb_flag(r, RSB_NEW_MASTER))
674 continue;
675
676 if (dlm_recovery_stopped(ls)) {
677 error = -EINTR;
678 up_read(&ls->ls_root_sem);
679 goto out;
680 }
681
682 error = recover_locks(r);
683 if (error) {
684 up_read(&ls->ls_root_sem);
685 goto out;
686 }
687
688 count += r->res_recover_locks_count;
689 }
690 up_read(&ls->ls_root_sem);
691
David Teigland48756472012-04-26 15:54:29 -0500692 log_debug(ls, "dlm_recover_locks %d out", count);
David Teiglande7fd4172006-01-18 09:30:29 +0000693
694 error = dlm_wait_function(ls, &recover_list_empty);
695 out:
696 if (error)
697 recover_list_clear(ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000698 return error;
699}
700
701void dlm_recovered_lock(struct dlm_rsb *r)
702{
David Teiglanda345da32006-08-18 11:54:25 -0500703 DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
David Teiglande7fd4172006-01-18 09:30:29 +0000704
705 r->res_recover_locks_count--;
706 if (!r->res_recover_locks_count) {
707 rsb_clear_flag(r, RSB_NEW_MASTER);
708 recover_list_del(r);
709 }
710
711 if (recover_list_empty(r->res_ls))
712 wake_up(&r->res_ls->ls_wait_general);
713}
714
715/*
716 * The lvb needs to be recovered on all master rsb's. This includes setting
717 * the VALNOTVALID flag if necessary, and determining the correct lvb contents
718 * based on the lvb's of the locks held on the rsb.
719 *
720 * RSB_VALNOTVALID is set if there are only NL/CR locks on the rsb. If it
721 * was already set prior to recovery, it's not cleared, regardless of locks.
722 *
723 * The LVB contents are only considered for changing when this is a new master
724 * of the rsb (NEW_MASTER2). Then, the rsb's lvb is taken from any lkb with
725 * mode > CR. If no lkb's exist with mode above CR, the lvb contents are taken
726 * from the lkb with the largest lvb sequence number.
727 */
728
729static void recover_lvb(struct dlm_rsb *r)
730{
731 struct dlm_lkb *lkb, *high_lkb = NULL;
732 uint32_t high_seq = 0;
David Teigland90135922006-01-20 08:47:07 +0000733 int lock_lvb_exists = 0;
734 int big_lock_exists = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000735 int lvblen = r->res_ls->ls_lvblen;
736
737 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
738 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
739 continue;
740
David Teigland90135922006-01-20 08:47:07 +0000741 lock_lvb_exists = 1;
David Teiglande7fd4172006-01-18 09:30:29 +0000742
743 if (lkb->lkb_grmode > DLM_LOCK_CR) {
David Teigland90135922006-01-20 08:47:07 +0000744 big_lock_exists = 1;
David Teiglande7fd4172006-01-18 09:30:29 +0000745 goto setflag;
746 }
747
748 if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
749 high_lkb = lkb;
750 high_seq = lkb->lkb_lvbseq;
751 }
752 }
753
754 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
755 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
756 continue;
757
David Teigland90135922006-01-20 08:47:07 +0000758 lock_lvb_exists = 1;
David Teiglande7fd4172006-01-18 09:30:29 +0000759
760 if (lkb->lkb_grmode > DLM_LOCK_CR) {
David Teigland90135922006-01-20 08:47:07 +0000761 big_lock_exists = 1;
David Teiglande7fd4172006-01-18 09:30:29 +0000762 goto setflag;
763 }
764
765 if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
766 high_lkb = lkb;
767 high_seq = lkb->lkb_lvbseq;
768 }
769 }
770
771 setflag:
772 if (!lock_lvb_exists)
773 goto out;
774
775 if (!big_lock_exists)
776 rsb_set_flag(r, RSB_VALNOTVALID);
777
778 /* don't mess with the lvb unless we're the new master */
779 if (!rsb_flag(r, RSB_NEW_MASTER2))
780 goto out;
781
782 if (!r->res_lvbptr) {
David Teigland52bda2b2007-11-07 09:06:49 -0600783 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000784 if (!r->res_lvbptr)
785 goto out;
786 }
787
788 if (big_lock_exists) {
789 r->res_lvbseq = lkb->lkb_lvbseq;
790 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, lvblen);
791 } else if (high_lkb) {
792 r->res_lvbseq = high_lkb->lkb_lvbseq;
793 memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
794 } else {
795 r->res_lvbseq = 0;
796 memset(r->res_lvbptr, 0, lvblen);
797 }
798 out:
799 return;
800}
801
802/* All master rsb's flagged RECOVER_CONVERT need to be looked at. The locks
803 converting PR->CW or CW->PR need to have their lkb_grmode set. */
804
805static void recover_conversion(struct dlm_rsb *r)
806{
David Teiglandc503a622012-06-05 15:55:19 -0500807 struct dlm_ls *ls = r->res_ls;
David Teiglande7fd4172006-01-18 09:30:29 +0000808 struct dlm_lkb *lkb;
809 int grmode = -1;
810
811 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
812 if (lkb->lkb_grmode == DLM_LOCK_PR ||
813 lkb->lkb_grmode == DLM_LOCK_CW) {
814 grmode = lkb->lkb_grmode;
815 break;
816 }
817 }
818
819 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
820 if (lkb->lkb_grmode != DLM_LOCK_IV)
821 continue;
David Teiglandc503a622012-06-05 15:55:19 -0500822 if (grmode == -1) {
823 log_debug(ls, "recover_conversion %x set gr to rq %d",
824 lkb->lkb_id, lkb->lkb_rqmode);
David Teiglande7fd4172006-01-18 09:30:29 +0000825 lkb->lkb_grmode = lkb->lkb_rqmode;
David Teiglandc503a622012-06-05 15:55:19 -0500826 } else {
827 log_debug(ls, "recover_conversion %x set gr %d",
828 lkb->lkb_id, grmode);
David Teiglande7fd4172006-01-18 09:30:29 +0000829 lkb->lkb_grmode = grmode;
David Teiglandc503a622012-06-05 15:55:19 -0500830 }
David Teiglande7fd4172006-01-18 09:30:29 +0000831 }
832}
833
David Teiglandf7da7902006-07-25 13:53:33 -0500834/* We've become the new master for this rsb and waiting/converting locks may
David Teigland48756472012-04-26 15:54:29 -0500835 need to be granted in dlm_recover_grant() due to locks that may have
David Teiglandf7da7902006-07-25 13:53:33 -0500836 existed from a removed node. */
837
David Teigland48756472012-04-26 15:54:29 -0500838static void recover_grant(struct dlm_rsb *r)
David Teiglandf7da7902006-07-25 13:53:33 -0500839{
840 if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
David Teigland48756472012-04-26 15:54:29 -0500841 rsb_set_flag(r, RSB_RECOVER_GRANT);
David Teiglandf7da7902006-07-25 13:53:33 -0500842}
843
David Teiglande7fd4172006-01-18 09:30:29 +0000844void dlm_recover_rsbs(struct dlm_ls *ls)
845{
846 struct dlm_rsb *r;
David Teigland48756472012-04-26 15:54:29 -0500847 unsigned int count = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000848
849 down_read(&ls->ls_root_sem);
850 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
851 lock_rsb(r);
852 if (is_master(r)) {
853 if (rsb_flag(r, RSB_RECOVER_CONVERT))
854 recover_conversion(r);
David Teiglandf7da7902006-07-25 13:53:33 -0500855 if (rsb_flag(r, RSB_NEW_MASTER2))
David Teigland48756472012-04-26 15:54:29 -0500856 recover_grant(r);
David Teiglande7fd4172006-01-18 09:30:29 +0000857 recover_lvb(r);
858 count++;
859 }
860 rsb_clear_flag(r, RSB_RECOVER_CONVERT);
David Teiglandf7da7902006-07-25 13:53:33 -0500861 rsb_clear_flag(r, RSB_NEW_MASTER2);
David Teiglande7fd4172006-01-18 09:30:29 +0000862 unlock_rsb(r);
863 }
864 up_read(&ls->ls_root_sem);
865
David Teigland48756472012-04-26 15:54:29 -0500866 if (count)
867 log_debug(ls, "dlm_recover_rsbs %d done", count);
David Teiglande7fd4172006-01-18 09:30:29 +0000868}
869
870/* Create a single list of all root rsb's to be used during recovery */
871
872int dlm_create_root_list(struct dlm_ls *ls)
873{
Bob Peterson9beb3bf2011-10-26 15:24:55 -0500874 struct rb_node *n;
David Teiglande7fd4172006-01-18 09:30:29 +0000875 struct dlm_rsb *r;
876 int i, error = 0;
877
878 down_write(&ls->ls_root_sem);
879 if (!list_empty(&ls->ls_root_list)) {
880 log_error(ls, "root list not empty");
881 error = -EINVAL;
882 goto out;
883 }
884
885 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
David Teiglandc7be7612009-01-07 16:50:41 -0600886 spin_lock(&ls->ls_rsbtbl[i].lock);
Bob Peterson9beb3bf2011-10-26 15:24:55 -0500887 for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
888 r = rb_entry(n, struct dlm_rsb, res_hashnode);
David Teiglande7fd4172006-01-18 09:30:29 +0000889 list_add(&r->res_root_list, &ls->ls_root_list);
890 dlm_hold_rsb(r);
891 }
David Teigland85f03792008-01-16 13:02:31 -0600892
David Teiglandc04fecb2012-05-10 10:18:07 -0500893 if (!RB_EMPTY_ROOT(&ls->ls_rsbtbl[i].toss))
894 log_error(ls, "dlm_create_root_list toss not empty");
David Teiglandc7be7612009-01-07 16:50:41 -0600895 spin_unlock(&ls->ls_rsbtbl[i].lock);
David Teiglande7fd4172006-01-18 09:30:29 +0000896 }
897 out:
898 up_write(&ls->ls_root_sem);
899 return error;
900}
901
902void dlm_release_root_list(struct dlm_ls *ls)
903{
904 struct dlm_rsb *r, *safe;
905
906 down_write(&ls->ls_root_sem);
907 list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
908 list_del_init(&r->res_root_list);
909 dlm_put_rsb(r);
910 }
911 up_write(&ls->ls_root_sem);
912}
913
David Teiglandc04fecb2012-05-10 10:18:07 -0500914void dlm_clear_toss(struct dlm_ls *ls)
David Teiglande7fd4172006-01-18 09:30:29 +0000915{
Bob Peterson9beb3bf2011-10-26 15:24:55 -0500916 struct rb_node *n, *next;
David Teiglandc04fecb2012-05-10 10:18:07 -0500917 struct dlm_rsb *r;
918 unsigned int count = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000919 int i;
920
921 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
David Teiglandc7be7612009-01-07 16:50:41 -0600922 spin_lock(&ls->ls_rsbtbl[i].lock);
Bob Peterson9beb3bf2011-10-26 15:24:55 -0500923 for (n = rb_first(&ls->ls_rsbtbl[i].toss); n; n = next) {
David Teiglandc04fecb2012-05-10 10:18:07 -0500924 next = rb_next(n);
925 r = rb_entry(n, struct dlm_rsb, res_hashnode);
926 rb_erase(n, &ls->ls_rsbtbl[i].toss);
927 dlm_free_rsb(r);
928 count++;
David Teiglande7fd4172006-01-18 09:30:29 +0000929 }
David Teiglandc7be7612009-01-07 16:50:41 -0600930 spin_unlock(&ls->ls_rsbtbl[i].lock);
David Teiglande7fd4172006-01-18 09:30:29 +0000931 }
David Teiglandc04fecb2012-05-10 10:18:07 -0500932
933 if (count)
934 log_debug(ls, "dlm_clear_toss %u done", count);
David Teiglande7fd4172006-01-18 09:30:29 +0000935}
936