Commit | Line | Data |
---|---|---|
e7fd4179 DT |
1 | /****************************************************************************** |
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 | |
39 | * timer to detect the result. A timer wakes us up periodically while waiting | |
40 | * to see if we should abort due to a node failure. This should only be called | |
41 | * by the dlm_recoverd thread. | |
42 | */ | |
43 | ||
44 | static void dlm_wait_timer_fn(unsigned long data) | |
45 | { | |
46 | struct dlm_ls *ls = (struct dlm_ls *) data; | |
47 | mod_timer(&ls->ls_timer, jiffies + (dlm_config.recover_timer * HZ)); | |
48 | wake_up(&ls->ls_wait_general); | |
49 | } | |
50 | ||
51 | int dlm_wait_function(struct dlm_ls *ls, int (*testfn) (struct dlm_ls *ls)) | |
52 | { | |
53 | int error = 0; | |
54 | ||
55 | init_timer(&ls->ls_timer); | |
56 | ls->ls_timer.function = dlm_wait_timer_fn; | |
57 | ls->ls_timer.data = (long) ls; | |
58 | ls->ls_timer.expires = jiffies + (dlm_config.recover_timer * HZ); | |
59 | add_timer(&ls->ls_timer); | |
60 | ||
61 | wait_event(ls->ls_wait_general, testfn(ls) || dlm_recovery_stopped(ls)); | |
62 | del_timer_sync(&ls->ls_timer); | |
63 | ||
64 | if (dlm_recovery_stopped(ls)) { | |
65 | log_debug(ls, "dlm_wait_function aborted"); | |
66 | error = -EINTR; | |
67 | } | |
68 | return error; | |
69 | } | |
70 | ||
71 | /* | |
72 | * An efficient way for all nodes to wait for all others to have a certain | |
73 | * status. The node with the lowest nodeid polls all the others for their | |
74 | * status (wait_status_all) and all the others poll the node with the low id | |
75 | * for its accumulated result (wait_status_low). When all nodes have set | |
76 | * status flag X, then status flag X_ALL will be set on the low nodeid. | |
77 | */ | |
78 | ||
79 | uint32_t dlm_recover_status(struct dlm_ls *ls) | |
80 | { | |
81 | uint32_t status; | |
82 | spin_lock(&ls->ls_recover_lock); | |
83 | status = ls->ls_recover_status; | |
84 | spin_unlock(&ls->ls_recover_lock); | |
85 | return status; | |
86 | } | |
87 | ||
88 | void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status) | |
89 | { | |
90 | spin_lock(&ls->ls_recover_lock); | |
91 | ls->ls_recover_status |= status; | |
92 | spin_unlock(&ls->ls_recover_lock); | |
93 | } | |
94 | ||
95 | static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status) | |
96 | { | |
97 | struct dlm_rcom *rc = (struct dlm_rcom *) ls->ls_recover_buf; | |
98 | struct dlm_member *memb; | |
99 | int error = 0, delay; | |
100 | ||
101 | list_for_each_entry(memb, &ls->ls_nodes, list) { | |
102 | delay = 0; | |
103 | for (;;) { | |
104 | if (dlm_recovery_stopped(ls)) { | |
105 | error = -EINTR; | |
106 | goto out; | |
107 | } | |
108 | ||
109 | error = dlm_rcom_status(ls, memb->nodeid); | |
110 | if (error) | |
111 | goto out; | |
112 | ||
113 | if (rc->rc_result & wait_status) | |
114 | break; | |
115 | if (delay < 1000) | |
116 | delay += 20; | |
117 | msleep(delay); | |
118 | } | |
119 | } | |
120 | out: | |
121 | return error; | |
122 | } | |
123 | ||
124 | static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status) | |
125 | { | |
126 | struct dlm_rcom *rc = (struct dlm_rcom *) ls->ls_recover_buf; | |
127 | int error = 0, delay = 0, nodeid = ls->ls_low_nodeid; | |
128 | ||
129 | for (;;) { | |
130 | if (dlm_recovery_stopped(ls)) { | |
131 | error = -EINTR; | |
132 | goto out; | |
133 | } | |
134 | ||
135 | error = dlm_rcom_status(ls, nodeid); | |
136 | if (error) | |
137 | break; | |
138 | ||
139 | if (rc->rc_result & wait_status) | |
140 | break; | |
141 | if (delay < 1000) | |
142 | delay += 20; | |
143 | msleep(delay); | |
144 | } | |
145 | out: | |
146 | return error; | |
147 | } | |
148 | ||
149 | static int wait_status(struct dlm_ls *ls, uint32_t status) | |
150 | { | |
151 | uint32_t status_all = status << 1; | |
152 | int error; | |
153 | ||
154 | if (ls->ls_low_nodeid == dlm_our_nodeid()) { | |
155 | error = wait_status_all(ls, status); | |
156 | if (!error) | |
157 | dlm_set_recover_status(ls, status_all); | |
158 | } else | |
159 | error = wait_status_low(ls, status_all); | |
160 | ||
161 | return error; | |
162 | } | |
163 | ||
164 | int dlm_recover_members_wait(struct dlm_ls *ls) | |
165 | { | |
166 | return wait_status(ls, DLM_RS_NODES); | |
167 | } | |
168 | ||
169 | int dlm_recover_directory_wait(struct dlm_ls *ls) | |
170 | { | |
171 | return wait_status(ls, DLM_RS_DIR); | |
172 | } | |
173 | ||
174 | int dlm_recover_locks_wait(struct dlm_ls *ls) | |
175 | { | |
176 | return wait_status(ls, DLM_RS_LOCKS); | |
177 | } | |
178 | ||
179 | int dlm_recover_done_wait(struct dlm_ls *ls) | |
180 | { | |
181 | return wait_status(ls, DLM_RS_DONE); | |
182 | } | |
183 | ||
184 | /* | |
185 | * The recover_list contains all the rsb's for which we've requested the new | |
186 | * master nodeid. As replies are returned from the resource directories the | |
187 | * rsb's are removed from the list. When the list is empty we're done. | |
188 | * | |
189 | * The recover_list is later similarly used for all rsb's for which we've sent | |
190 | * new lkb's and need to receive new corresponding lkid's. | |
191 | * | |
192 | * We use the address of the rsb struct as a simple local identifier for the | |
193 | * rsb so we can match an rcom reply with the rsb it was sent for. | |
194 | */ | |
195 | ||
196 | static int recover_list_empty(struct dlm_ls *ls) | |
197 | { | |
198 | int empty; | |
199 | ||
200 | spin_lock(&ls->ls_recover_list_lock); | |
201 | empty = list_empty(&ls->ls_recover_list); | |
202 | spin_unlock(&ls->ls_recover_list_lock); | |
203 | ||
204 | return empty; | |
205 | } | |
206 | ||
207 | static void recover_list_add(struct dlm_rsb *r) | |
208 | { | |
209 | struct dlm_ls *ls = r->res_ls; | |
210 | ||
211 | spin_lock(&ls->ls_recover_list_lock); | |
212 | if (list_empty(&r->res_recover_list)) { | |
213 | list_add_tail(&r->res_recover_list, &ls->ls_recover_list); | |
214 | ls->ls_recover_list_count++; | |
215 | dlm_hold_rsb(r); | |
216 | } | |
217 | spin_unlock(&ls->ls_recover_list_lock); | |
218 | } | |
219 | ||
220 | static void recover_list_del(struct dlm_rsb *r) | |
221 | { | |
222 | struct dlm_ls *ls = r->res_ls; | |
223 | ||
224 | spin_lock(&ls->ls_recover_list_lock); | |
225 | list_del_init(&r->res_recover_list); | |
226 | ls->ls_recover_list_count--; | |
227 | spin_unlock(&ls->ls_recover_list_lock); | |
228 | ||
229 | dlm_put_rsb(r); | |
230 | } | |
231 | ||
232 | static struct dlm_rsb *recover_list_find(struct dlm_ls *ls, uint64_t id) | |
233 | { | |
234 | struct dlm_rsb *r = NULL; | |
235 | ||
236 | spin_lock(&ls->ls_recover_list_lock); | |
237 | ||
238 | list_for_each_entry(r, &ls->ls_recover_list, res_recover_list) { | |
239 | if (id == (unsigned long) r) | |
240 | goto out; | |
241 | } | |
242 | r = NULL; | |
243 | out: | |
244 | spin_unlock(&ls->ls_recover_list_lock); | |
245 | return r; | |
246 | } | |
247 | ||
248 | static void recover_list_clear(struct dlm_ls *ls) | |
249 | { | |
250 | struct dlm_rsb *r, *s; | |
251 | ||
252 | spin_lock(&ls->ls_recover_list_lock); | |
253 | list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) { | |
254 | list_del_init(&r->res_recover_list); | |
255 | dlm_put_rsb(r); | |
256 | ls->ls_recover_list_count--; | |
257 | } | |
258 | ||
259 | if (ls->ls_recover_list_count != 0) { | |
260 | log_error(ls, "warning: recover_list_count %d", | |
261 | ls->ls_recover_list_count); | |
262 | ls->ls_recover_list_count = 0; | |
263 | } | |
264 | spin_unlock(&ls->ls_recover_list_lock); | |
265 | } | |
266 | ||
267 | ||
268 | /* Master recovery: find new master node for rsb's that were | |
269 | mastered on nodes that have been removed. | |
270 | ||
271 | dlm_recover_masters | |
272 | recover_master | |
273 | dlm_send_rcom_lookup -> receive_rcom_lookup | |
274 | dlm_dir_lookup | |
275 | receive_rcom_lookup_reply <- | |
276 | dlm_recover_master_reply | |
277 | set_new_master | |
278 | set_master_lkbs | |
279 | set_lock_master | |
280 | */ | |
281 | ||
282 | /* | |
283 | * Set the lock master for all LKBs in a lock queue | |
284 | * If we are the new master of the rsb, we may have received new | |
285 | * MSTCPY locks from other nodes already which we need to ignore | |
286 | * when setting the new nodeid. | |
287 | */ | |
288 | ||
289 | static void set_lock_master(struct list_head *queue, int nodeid) | |
290 | { | |
291 | struct dlm_lkb *lkb; | |
292 | ||
293 | list_for_each_entry(lkb, queue, lkb_statequeue) | |
294 | if (!(lkb->lkb_flags & DLM_IFL_MSTCPY)) | |
295 | lkb->lkb_nodeid = nodeid; | |
296 | } | |
297 | ||
298 | static void set_master_lkbs(struct dlm_rsb *r) | |
299 | { | |
300 | set_lock_master(&r->res_grantqueue, r->res_nodeid); | |
301 | set_lock_master(&r->res_convertqueue, r->res_nodeid); | |
302 | set_lock_master(&r->res_waitqueue, r->res_nodeid); | |
303 | } | |
304 | ||
305 | /* | |
306 | * Propogate the new master nodeid to locks | |
307 | * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider. | |
f7da790d DT |
308 | * The NEW_MASTER2 flag tells recover_lvb() and set_locks_purged() which |
309 | * rsb's to consider. | |
e7fd4179 DT |
310 | */ |
311 | ||
312 | static void set_new_master(struct dlm_rsb *r, int nodeid) | |
313 | { | |
314 | lock_rsb(r); | |
315 | r->res_nodeid = nodeid; | |
316 | set_master_lkbs(r); | |
317 | rsb_set_flag(r, RSB_NEW_MASTER); | |
318 | rsb_set_flag(r, RSB_NEW_MASTER2); | |
319 | unlock_rsb(r); | |
320 | } | |
321 | ||
322 | /* | |
323 | * We do async lookups on rsb's that need new masters. The rsb's | |
324 | * waiting for a lookup reply are kept on the recover_list. | |
325 | */ | |
326 | ||
327 | static int recover_master(struct dlm_rsb *r) | |
328 | { | |
329 | struct dlm_ls *ls = r->res_ls; | |
330 | int error, dir_nodeid, ret_nodeid, our_nodeid = dlm_our_nodeid(); | |
331 | ||
332 | dir_nodeid = dlm_dir_nodeid(r); | |
333 | ||
334 | if (dir_nodeid == our_nodeid) { | |
335 | error = dlm_dir_lookup(ls, our_nodeid, r->res_name, | |
336 | r->res_length, &ret_nodeid); | |
337 | if (error) | |
338 | log_error(ls, "recover dir lookup error %d", error); | |
339 | ||
340 | if (ret_nodeid == our_nodeid) | |
341 | ret_nodeid = 0; | |
342 | set_new_master(r, ret_nodeid); | |
343 | } else { | |
344 | recover_list_add(r); | |
345 | error = dlm_send_rcom_lookup(r, dir_nodeid); | |
346 | } | |
347 | ||
348 | return error; | |
349 | } | |
350 | ||
351 | /* | |
352 | * When not using a directory, most resource names will hash to a new static | |
353 | * master nodeid and the resource will need to be remastered. | |
354 | */ | |
355 | ||
356 | static int recover_master_static(struct dlm_rsb *r) | |
357 | { | |
358 | int master = dlm_dir_nodeid(r); | |
359 | ||
360 | if (master == dlm_our_nodeid()) | |
361 | master = 0; | |
362 | ||
363 | if (r->res_nodeid != master) { | |
364 | if (is_master(r)) | |
365 | dlm_purge_mstcpy_locks(r); | |
366 | set_new_master(r, master); | |
367 | return 1; | |
368 | } | |
369 | return 0; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Go through local root resources and for each rsb which has a master which | |
374 | * has departed, get the new master nodeid from the directory. The dir will | |
375 | * assign mastery to the first node to look up the new master. That means | |
376 | * we'll discover in this lookup if we're the new master of any rsb's. | |
377 | * | |
378 | * We fire off all the dir lookup requests individually and asynchronously to | |
379 | * the correct dir node. | |
380 | */ | |
381 | ||
382 | int dlm_recover_masters(struct dlm_ls *ls) | |
383 | { | |
384 | struct dlm_rsb *r; | |
385 | int error = 0, count = 0; | |
386 | ||
387 | log_debug(ls, "dlm_recover_masters"); | |
388 | ||
389 | down_read(&ls->ls_root_sem); | |
390 | list_for_each_entry(r, &ls->ls_root_list, res_root_list) { | |
391 | if (dlm_recovery_stopped(ls)) { | |
392 | up_read(&ls->ls_root_sem); | |
393 | error = -EINTR; | |
394 | goto out; | |
395 | } | |
396 | ||
397 | if (dlm_no_directory(ls)) | |
398 | count += recover_master_static(r); | |
399 | else if (!is_master(r) && dlm_is_removed(ls, r->res_nodeid)) { | |
400 | recover_master(r); | |
401 | count++; | |
402 | } | |
403 | ||
404 | schedule(); | |
405 | } | |
406 | up_read(&ls->ls_root_sem); | |
407 | ||
408 | log_debug(ls, "dlm_recover_masters %d resources", count); | |
409 | ||
410 | error = dlm_wait_function(ls, &recover_list_empty); | |
411 | out: | |
412 | if (error) | |
413 | recover_list_clear(ls); | |
414 | return error; | |
415 | } | |
416 | ||
417 | int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc) | |
418 | { | |
419 | struct dlm_rsb *r; | |
420 | int nodeid; | |
421 | ||
422 | r = recover_list_find(ls, rc->rc_id); | |
423 | if (!r) { | |
90135925 | 424 | log_error(ls, "dlm_recover_master_reply no id %llx", |
9229f013 | 425 | (unsigned long long)rc->rc_id); |
e7fd4179 DT |
426 | goto out; |
427 | } | |
428 | ||
429 | nodeid = rc->rc_result; | |
430 | if (nodeid == dlm_our_nodeid()) | |
431 | nodeid = 0; | |
432 | ||
433 | set_new_master(r, nodeid); | |
434 | recover_list_del(r); | |
435 | ||
436 | if (recover_list_empty(ls)) | |
437 | wake_up(&ls->ls_wait_general); | |
438 | out: | |
439 | return 0; | |
440 | } | |
441 | ||
442 | ||
443 | /* Lock recovery: rebuild the process-copy locks we hold on a | |
444 | remastered rsb on the new rsb master. | |
445 | ||
446 | dlm_recover_locks | |
447 | recover_locks | |
448 | recover_locks_queue | |
449 | dlm_send_rcom_lock -> receive_rcom_lock | |
450 | dlm_recover_master_copy | |
451 | receive_rcom_lock_reply <- | |
452 | dlm_recover_process_copy | |
453 | */ | |
454 | ||
455 | ||
456 | /* | |
457 | * keep a count of the number of lkb's we send to the new master; when we get | |
458 | * an equal number of replies then recovery for the rsb is done | |
459 | */ | |
460 | ||
461 | static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head) | |
462 | { | |
463 | struct dlm_lkb *lkb; | |
464 | int error = 0; | |
465 | ||
466 | list_for_each_entry(lkb, head, lkb_statequeue) { | |
467 | error = dlm_send_rcom_lock(r, lkb); | |
468 | if (error) | |
469 | break; | |
470 | r->res_recover_locks_count++; | |
471 | } | |
472 | ||
473 | return error; | |
474 | } | |
475 | ||
476 | static int all_queues_empty(struct dlm_rsb *r) | |
477 | { | |
478 | if (!list_empty(&r->res_grantqueue) || | |
479 | !list_empty(&r->res_convertqueue) || | |
480 | !list_empty(&r->res_waitqueue)) | |
90135925 DT |
481 | return 0; |
482 | return 1; | |
e7fd4179 DT |
483 | } |
484 | ||
485 | static int recover_locks(struct dlm_rsb *r) | |
486 | { | |
487 | int error = 0; | |
488 | ||
489 | lock_rsb(r); | |
490 | if (all_queues_empty(r)) | |
491 | goto out; | |
492 | ||
493 | DLM_ASSERT(!r->res_recover_locks_count, dlm_print_rsb(r);); | |
494 | ||
495 | error = recover_locks_queue(r, &r->res_grantqueue); | |
496 | if (error) | |
497 | goto out; | |
498 | error = recover_locks_queue(r, &r->res_convertqueue); | |
499 | if (error) | |
500 | goto out; | |
501 | error = recover_locks_queue(r, &r->res_waitqueue); | |
502 | if (error) | |
503 | goto out; | |
504 | ||
505 | if (r->res_recover_locks_count) | |
506 | recover_list_add(r); | |
507 | else | |
508 | rsb_clear_flag(r, RSB_NEW_MASTER); | |
509 | out: | |
510 | unlock_rsb(r); | |
511 | return error; | |
512 | } | |
513 | ||
514 | int dlm_recover_locks(struct dlm_ls *ls) | |
515 | { | |
516 | struct dlm_rsb *r; | |
517 | int error, count = 0; | |
518 | ||
519 | log_debug(ls, "dlm_recover_locks"); | |
520 | ||
521 | down_read(&ls->ls_root_sem); | |
522 | list_for_each_entry(r, &ls->ls_root_list, res_root_list) { | |
523 | if (is_master(r)) { | |
524 | rsb_clear_flag(r, RSB_NEW_MASTER); | |
525 | continue; | |
526 | } | |
527 | ||
528 | if (!rsb_flag(r, RSB_NEW_MASTER)) | |
529 | continue; | |
530 | ||
531 | if (dlm_recovery_stopped(ls)) { | |
532 | error = -EINTR; | |
533 | up_read(&ls->ls_root_sem); | |
534 | goto out; | |
535 | } | |
536 | ||
537 | error = recover_locks(r); | |
538 | if (error) { | |
539 | up_read(&ls->ls_root_sem); | |
540 | goto out; | |
541 | } | |
542 | ||
543 | count += r->res_recover_locks_count; | |
544 | } | |
545 | up_read(&ls->ls_root_sem); | |
546 | ||
547 | log_debug(ls, "dlm_recover_locks %d locks", count); | |
548 | ||
549 | error = dlm_wait_function(ls, &recover_list_empty); | |
550 | out: | |
551 | if (error) | |
552 | recover_list_clear(ls); | |
553 | else | |
554 | dlm_set_recover_status(ls, DLM_RS_LOCKS); | |
555 | return error; | |
556 | } | |
557 | ||
558 | void dlm_recovered_lock(struct dlm_rsb *r) | |
559 | { | |
560 | DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_print_rsb(r);); | |
561 | ||
562 | r->res_recover_locks_count--; | |
563 | if (!r->res_recover_locks_count) { | |
564 | rsb_clear_flag(r, RSB_NEW_MASTER); | |
565 | recover_list_del(r); | |
566 | } | |
567 | ||
568 | if (recover_list_empty(r->res_ls)) | |
569 | wake_up(&r->res_ls->ls_wait_general); | |
570 | } | |
571 | ||
572 | /* | |
573 | * The lvb needs to be recovered on all master rsb's. This includes setting | |
574 | * the VALNOTVALID flag if necessary, and determining the correct lvb contents | |
575 | * based on the lvb's of the locks held on the rsb. | |
576 | * | |
577 | * RSB_VALNOTVALID is set if there are only NL/CR locks on the rsb. If it | |
578 | * was already set prior to recovery, it's not cleared, regardless of locks. | |
579 | * | |
580 | * The LVB contents are only considered for changing when this is a new master | |
581 | * of the rsb (NEW_MASTER2). Then, the rsb's lvb is taken from any lkb with | |
582 | * mode > CR. If no lkb's exist with mode above CR, the lvb contents are taken | |
583 | * from the lkb with the largest lvb sequence number. | |
584 | */ | |
585 | ||
586 | static void recover_lvb(struct dlm_rsb *r) | |
587 | { | |
588 | struct dlm_lkb *lkb, *high_lkb = NULL; | |
589 | uint32_t high_seq = 0; | |
90135925 DT |
590 | int lock_lvb_exists = 0; |
591 | int big_lock_exists = 0; | |
e7fd4179 DT |
592 | int lvblen = r->res_ls->ls_lvblen; |
593 | ||
594 | list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) { | |
595 | if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) | |
596 | continue; | |
597 | ||
90135925 | 598 | lock_lvb_exists = 1; |
e7fd4179 DT |
599 | |
600 | if (lkb->lkb_grmode > DLM_LOCK_CR) { | |
90135925 | 601 | big_lock_exists = 1; |
e7fd4179 DT |
602 | goto setflag; |
603 | } | |
604 | ||
605 | if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) { | |
606 | high_lkb = lkb; | |
607 | high_seq = lkb->lkb_lvbseq; | |
608 | } | |
609 | } | |
610 | ||
611 | list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) { | |
612 | if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) | |
613 | continue; | |
614 | ||
90135925 | 615 | lock_lvb_exists = 1; |
e7fd4179 DT |
616 | |
617 | if (lkb->lkb_grmode > DLM_LOCK_CR) { | |
90135925 | 618 | big_lock_exists = 1; |
e7fd4179 DT |
619 | goto setflag; |
620 | } | |
621 | ||
622 | if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) { | |
623 | high_lkb = lkb; | |
624 | high_seq = lkb->lkb_lvbseq; | |
625 | } | |
626 | } | |
627 | ||
628 | setflag: | |
629 | if (!lock_lvb_exists) | |
630 | goto out; | |
631 | ||
632 | if (!big_lock_exists) | |
633 | rsb_set_flag(r, RSB_VALNOTVALID); | |
634 | ||
635 | /* don't mess with the lvb unless we're the new master */ | |
636 | if (!rsb_flag(r, RSB_NEW_MASTER2)) | |
637 | goto out; | |
638 | ||
639 | if (!r->res_lvbptr) { | |
640 | r->res_lvbptr = allocate_lvb(r->res_ls); | |
641 | if (!r->res_lvbptr) | |
642 | goto out; | |
643 | } | |
644 | ||
645 | if (big_lock_exists) { | |
646 | r->res_lvbseq = lkb->lkb_lvbseq; | |
647 | memcpy(r->res_lvbptr, lkb->lkb_lvbptr, lvblen); | |
648 | } else if (high_lkb) { | |
649 | r->res_lvbseq = high_lkb->lkb_lvbseq; | |
650 | memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen); | |
651 | } else { | |
652 | r->res_lvbseq = 0; | |
653 | memset(r->res_lvbptr, 0, lvblen); | |
654 | } | |
655 | out: | |
656 | return; | |
657 | } | |
658 | ||
659 | /* All master rsb's flagged RECOVER_CONVERT need to be looked at. The locks | |
660 | converting PR->CW or CW->PR need to have their lkb_grmode set. */ | |
661 | ||
662 | static void recover_conversion(struct dlm_rsb *r) | |
663 | { | |
664 | struct dlm_lkb *lkb; | |
665 | int grmode = -1; | |
666 | ||
667 | list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) { | |
668 | if (lkb->lkb_grmode == DLM_LOCK_PR || | |
669 | lkb->lkb_grmode == DLM_LOCK_CW) { | |
670 | grmode = lkb->lkb_grmode; | |
671 | break; | |
672 | } | |
673 | } | |
674 | ||
675 | list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) { | |
676 | if (lkb->lkb_grmode != DLM_LOCK_IV) | |
677 | continue; | |
678 | if (grmode == -1) | |
679 | lkb->lkb_grmode = lkb->lkb_rqmode; | |
680 | else | |
681 | lkb->lkb_grmode = grmode; | |
682 | } | |
683 | } | |
684 | ||
f7da790d DT |
685 | /* We've become the new master for this rsb and waiting/converting locks may |
686 | need to be granted in dlm_grant_after_purge() due to locks that may have | |
687 | existed from a removed node. */ | |
688 | ||
689 | static void set_locks_purged(struct dlm_rsb *r) | |
690 | { | |
691 | if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue)) | |
692 | rsb_set_flag(r, RSB_LOCKS_PURGED); | |
693 | } | |
694 | ||
e7fd4179 DT |
695 | void dlm_recover_rsbs(struct dlm_ls *ls) |
696 | { | |
697 | struct dlm_rsb *r; | |
698 | int count = 0; | |
699 | ||
700 | log_debug(ls, "dlm_recover_rsbs"); | |
701 | ||
702 | down_read(&ls->ls_root_sem); | |
703 | list_for_each_entry(r, &ls->ls_root_list, res_root_list) { | |
704 | lock_rsb(r); | |
705 | if (is_master(r)) { | |
706 | if (rsb_flag(r, RSB_RECOVER_CONVERT)) | |
707 | recover_conversion(r); | |
f7da790d DT |
708 | if (rsb_flag(r, RSB_NEW_MASTER2)) |
709 | set_locks_purged(r); | |
e7fd4179 DT |
710 | recover_lvb(r); |
711 | count++; | |
712 | } | |
713 | rsb_clear_flag(r, RSB_RECOVER_CONVERT); | |
f7da790d | 714 | rsb_clear_flag(r, RSB_NEW_MASTER2); |
e7fd4179 DT |
715 | unlock_rsb(r); |
716 | } | |
717 | up_read(&ls->ls_root_sem); | |
718 | ||
719 | log_debug(ls, "dlm_recover_rsbs %d rsbs", count); | |
720 | } | |
721 | ||
722 | /* Create a single list of all root rsb's to be used during recovery */ | |
723 | ||
724 | int dlm_create_root_list(struct dlm_ls *ls) | |
725 | { | |
726 | struct dlm_rsb *r; | |
727 | int i, error = 0; | |
728 | ||
729 | down_write(&ls->ls_root_sem); | |
730 | if (!list_empty(&ls->ls_root_list)) { | |
731 | log_error(ls, "root list not empty"); | |
732 | error = -EINVAL; | |
733 | goto out; | |
734 | } | |
735 | ||
736 | for (i = 0; i < ls->ls_rsbtbl_size; i++) { | |
737 | read_lock(&ls->ls_rsbtbl[i].lock); | |
738 | list_for_each_entry(r, &ls->ls_rsbtbl[i].list, res_hashchain) { | |
739 | list_add(&r->res_root_list, &ls->ls_root_list); | |
740 | dlm_hold_rsb(r); | |
741 | } | |
742 | read_unlock(&ls->ls_rsbtbl[i].lock); | |
743 | } | |
744 | out: | |
745 | up_write(&ls->ls_root_sem); | |
746 | return error; | |
747 | } | |
748 | ||
749 | void dlm_release_root_list(struct dlm_ls *ls) | |
750 | { | |
751 | struct dlm_rsb *r, *safe; | |
752 | ||
753 | down_write(&ls->ls_root_sem); | |
754 | list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) { | |
755 | list_del_init(&r->res_root_list); | |
756 | dlm_put_rsb(r); | |
757 | } | |
758 | up_write(&ls->ls_root_sem); | |
759 | } | |
760 | ||
761 | void dlm_clear_toss_list(struct dlm_ls *ls) | |
762 | { | |
763 | struct dlm_rsb *r, *safe; | |
764 | int i; | |
765 | ||
766 | for (i = 0; i < ls->ls_rsbtbl_size; i++) { | |
767 | write_lock(&ls->ls_rsbtbl[i].lock); | |
768 | list_for_each_entry_safe(r, safe, &ls->ls_rsbtbl[i].toss, | |
769 | res_hashchain) { | |
770 | list_del(&r->res_hashchain); | |
771 | free_rsb(r); | |
772 | } | |
773 | write_unlock(&ls->ls_rsbtbl[i].lock); | |
774 | } | |
775 | } | |
776 |