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; | |
68c817a1 | 47 | mod_timer(&ls->ls_timer, jiffies + (dlm_config.ci_recover_timer * HZ)); |
e7fd4179 DT |
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; | |
68c817a1 | 58 | ls->ls_timer.expires = jiffies + (dlm_config.ci_recover_timer * HZ); |
e7fd4179 DT |
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 | ||
757a4271 DT |
88 | static void _set_recover_status(struct dlm_ls *ls, uint32_t status) |
89 | { | |
90 | ls->ls_recover_status |= status; | |
91 | } | |
92 | ||
e7fd4179 DT |
93 | void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status) |
94 | { | |
95 | spin_lock(&ls->ls_recover_lock); | |
757a4271 | 96 | _set_recover_status(ls, status); |
e7fd4179 DT |
97 | spin_unlock(&ls->ls_recover_lock); |
98 | } | |
99 | ||
757a4271 DT |
100 | static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status, |
101 | int save_slots) | |
e7fd4179 | 102 | { |
4007685c | 103 | struct dlm_rcom *rc = ls->ls_recover_buf; |
e7fd4179 DT |
104 | struct dlm_member *memb; |
105 | int error = 0, delay; | |
106 | ||
107 | list_for_each_entry(memb, &ls->ls_nodes, list) { | |
108 | delay = 0; | |
109 | for (;;) { | |
110 | if (dlm_recovery_stopped(ls)) { | |
111 | error = -EINTR; | |
112 | goto out; | |
113 | } | |
114 | ||
757a4271 | 115 | error = dlm_rcom_status(ls, memb->nodeid, 0); |
e7fd4179 DT |
116 | if (error) |
117 | goto out; | |
118 | ||
757a4271 DT |
119 | if (save_slots) |
120 | dlm_slot_save(ls, rc, memb); | |
121 | ||
e7fd4179 DT |
122 | if (rc->rc_result & wait_status) |
123 | break; | |
124 | if (delay < 1000) | |
125 | delay += 20; | |
126 | msleep(delay); | |
127 | } | |
128 | } | |
129 | out: | |
130 | return error; | |
131 | } | |
132 | ||
757a4271 DT |
133 | static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status, |
134 | uint32_t status_flags) | |
e7fd4179 | 135 | { |
4007685c | 136 | struct dlm_rcom *rc = ls->ls_recover_buf; |
e7fd4179 DT |
137 | int error = 0, delay = 0, nodeid = ls->ls_low_nodeid; |
138 | ||
139 | for (;;) { | |
140 | if (dlm_recovery_stopped(ls)) { | |
141 | error = -EINTR; | |
142 | goto out; | |
143 | } | |
144 | ||
757a4271 | 145 | error = dlm_rcom_status(ls, nodeid, status_flags); |
e7fd4179 DT |
146 | if (error) |
147 | break; | |
148 | ||
149 | if (rc->rc_result & wait_status) | |
150 | break; | |
151 | if (delay < 1000) | |
152 | delay += 20; | |
153 | msleep(delay); | |
154 | } | |
155 | out: | |
156 | return error; | |
157 | } | |
158 | ||
159 | static int wait_status(struct dlm_ls *ls, uint32_t status) | |
160 | { | |
161 | uint32_t status_all = status << 1; | |
162 | int error; | |
163 | ||
164 | if (ls->ls_low_nodeid == dlm_our_nodeid()) { | |
757a4271 | 165 | error = wait_status_all(ls, status, 0); |
e7fd4179 DT |
166 | if (!error) |
167 | dlm_set_recover_status(ls, status_all); | |
168 | } else | |
757a4271 | 169 | error = wait_status_low(ls, status_all, 0); |
e7fd4179 DT |
170 | |
171 | return error; | |
172 | } | |
173 | ||
174 | int dlm_recover_members_wait(struct dlm_ls *ls) | |
175 | { | |
757a4271 DT |
176 | struct dlm_member *memb; |
177 | struct dlm_slot *slots; | |
178 | int num_slots, slots_size; | |
179 | int error, rv; | |
180 | uint32_t gen; | |
181 | ||
182 | list_for_each_entry(memb, &ls->ls_nodes, list) { | |
183 | memb->slot = -1; | |
184 | memb->generation = 0; | |
185 | } | |
186 | ||
187 | if (ls->ls_low_nodeid == dlm_our_nodeid()) { | |
188 | error = wait_status_all(ls, DLM_RS_NODES, 1); | |
189 | if (error) | |
190 | goto out; | |
191 | ||
192 | /* slots array is sparse, slots_size may be > num_slots */ | |
193 | ||
194 | rv = dlm_slots_assign(ls, &num_slots, &slots_size, &slots, &gen); | |
195 | if (!rv) { | |
196 | spin_lock(&ls->ls_recover_lock); | |
197 | _set_recover_status(ls, DLM_RS_NODES_ALL); | |
198 | ls->ls_num_slots = num_slots; | |
199 | ls->ls_slots_size = slots_size; | |
200 | ls->ls_slots = slots; | |
201 | ls->ls_generation = gen; | |
202 | spin_unlock(&ls->ls_recover_lock); | |
203 | } else { | |
204 | dlm_set_recover_status(ls, DLM_RS_NODES_ALL); | |
205 | } | |
206 | } else { | |
207 | error = wait_status_low(ls, DLM_RS_NODES_ALL, DLM_RSF_NEED_SLOTS); | |
208 | if (error) | |
209 | goto out; | |
210 | ||
211 | dlm_slots_copy_in(ls); | |
212 | } | |
213 | out: | |
214 | return error; | |
e7fd4179 DT |
215 | } |
216 | ||
217 | int dlm_recover_directory_wait(struct dlm_ls *ls) | |
218 | { | |
219 | return wait_status(ls, DLM_RS_DIR); | |
220 | } | |
221 | ||
222 | int dlm_recover_locks_wait(struct dlm_ls *ls) | |
223 | { | |
224 | return wait_status(ls, DLM_RS_LOCKS); | |
225 | } | |
226 | ||
227 | int dlm_recover_done_wait(struct dlm_ls *ls) | |
228 | { | |
229 | return wait_status(ls, DLM_RS_DONE); | |
230 | } | |
231 | ||
232 | /* | |
233 | * The recover_list contains all the rsb's for which we've requested the new | |
234 | * master nodeid. As replies are returned from the resource directories the | |
235 | * rsb's are removed from the list. When the list is empty we're done. | |
236 | * | |
237 | * The recover_list is later similarly used for all rsb's for which we've sent | |
238 | * new lkb's and need to receive new corresponding lkid's. | |
239 | * | |
240 | * We use the address of the rsb struct as a simple local identifier for the | |
241 | * rsb so we can match an rcom reply with the rsb it was sent for. | |
242 | */ | |
243 | ||
244 | static int recover_list_empty(struct dlm_ls *ls) | |
245 | { | |
246 | int empty; | |
247 | ||
248 | spin_lock(&ls->ls_recover_list_lock); | |
249 | empty = list_empty(&ls->ls_recover_list); | |
250 | spin_unlock(&ls->ls_recover_list_lock); | |
251 | ||
252 | return empty; | |
253 | } | |
254 | ||
255 | static void recover_list_add(struct dlm_rsb *r) | |
256 | { | |
257 | struct dlm_ls *ls = r->res_ls; | |
258 | ||
259 | spin_lock(&ls->ls_recover_list_lock); | |
260 | if (list_empty(&r->res_recover_list)) { | |
261 | list_add_tail(&r->res_recover_list, &ls->ls_recover_list); | |
262 | ls->ls_recover_list_count++; | |
263 | dlm_hold_rsb(r); | |
264 | } | |
265 | spin_unlock(&ls->ls_recover_list_lock); | |
266 | } | |
267 | ||
268 | static void recover_list_del(struct dlm_rsb *r) | |
269 | { | |
270 | struct dlm_ls *ls = r->res_ls; | |
271 | ||
272 | spin_lock(&ls->ls_recover_list_lock); | |
273 | list_del_init(&r->res_recover_list); | |
274 | ls->ls_recover_list_count--; | |
275 | spin_unlock(&ls->ls_recover_list_lock); | |
276 | ||
277 | dlm_put_rsb(r); | |
278 | } | |
279 | ||
280 | static struct dlm_rsb *recover_list_find(struct dlm_ls *ls, uint64_t id) | |
281 | { | |
282 | struct dlm_rsb *r = NULL; | |
283 | ||
284 | spin_lock(&ls->ls_recover_list_lock); | |
285 | ||
286 | list_for_each_entry(r, &ls->ls_recover_list, res_recover_list) { | |
287 | if (id == (unsigned long) r) | |
288 | goto out; | |
289 | } | |
290 | r = NULL; | |
291 | out: | |
292 | spin_unlock(&ls->ls_recover_list_lock); | |
293 | return r; | |
294 | } | |
295 | ||
296 | static void recover_list_clear(struct dlm_ls *ls) | |
297 | { | |
298 | struct dlm_rsb *r, *s; | |
299 | ||
300 | spin_lock(&ls->ls_recover_list_lock); | |
301 | list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) { | |
302 | list_del_init(&r->res_recover_list); | |
52069809 | 303 | r->res_recover_locks_count = 0; |
e7fd4179 DT |
304 | dlm_put_rsb(r); |
305 | ls->ls_recover_list_count--; | |
306 | } | |
307 | ||
308 | if (ls->ls_recover_list_count != 0) { | |
309 | log_error(ls, "warning: recover_list_count %d", | |
310 | ls->ls_recover_list_count); | |
311 | ls->ls_recover_list_count = 0; | |
312 | } | |
313 | spin_unlock(&ls->ls_recover_list_lock); | |
314 | } | |
315 | ||
316 | ||
317 | /* Master recovery: find new master node for rsb's that were | |
318 | mastered on nodes that have been removed. | |
319 | ||
320 | dlm_recover_masters | |
321 | recover_master | |
322 | dlm_send_rcom_lookup -> receive_rcom_lookup | |
323 | dlm_dir_lookup | |
324 | receive_rcom_lookup_reply <- | |
325 | dlm_recover_master_reply | |
326 | set_new_master | |
327 | set_master_lkbs | |
328 | set_lock_master | |
329 | */ | |
330 | ||
331 | /* | |
332 | * Set the lock master for all LKBs in a lock queue | |
333 | * If we are the new master of the rsb, we may have received new | |
334 | * MSTCPY locks from other nodes already which we need to ignore | |
335 | * when setting the new nodeid. | |
336 | */ | |
337 | ||
338 | static void set_lock_master(struct list_head *queue, int nodeid) | |
339 | { | |
340 | struct dlm_lkb *lkb; | |
341 | ||
4875647a DT |
342 | list_for_each_entry(lkb, queue, lkb_statequeue) { |
343 | if (!(lkb->lkb_flags & DLM_IFL_MSTCPY)) { | |
e7fd4179 | 344 | lkb->lkb_nodeid = nodeid; |
4875647a DT |
345 | lkb->lkb_remid = 0; |
346 | } | |
347 | } | |
e7fd4179 DT |
348 | } |
349 | ||
350 | static void set_master_lkbs(struct dlm_rsb *r) | |
351 | { | |
352 | set_lock_master(&r->res_grantqueue, r->res_nodeid); | |
353 | set_lock_master(&r->res_convertqueue, r->res_nodeid); | |
354 | set_lock_master(&r->res_waitqueue, r->res_nodeid); | |
355 | } | |
356 | ||
357 | /* | |
25985edc | 358 | * Propagate the new master nodeid to locks |
e7fd4179 | 359 | * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider. |
4875647a | 360 | * The NEW_MASTER2 flag tells recover_lvb() and recover_grant() which |
f7da790d | 361 | * rsb's to consider. |
e7fd4179 DT |
362 | */ |
363 | ||
c04fecb4 | 364 | static void set_new_master(struct dlm_rsb *r) |
e7fd4179 | 365 | { |
e7fd4179 DT |
366 | set_master_lkbs(r); |
367 | rsb_set_flag(r, RSB_NEW_MASTER); | |
368 | rsb_set_flag(r, RSB_NEW_MASTER2); | |
e7fd4179 DT |
369 | } |
370 | ||
371 | /* | |
372 | * We do async lookups on rsb's that need new masters. The rsb's | |
373 | * waiting for a lookup reply are kept on the recover_list. | |
c04fecb4 DT |
374 | * |
375 | * Another node recovering the master may have sent us a rcom lookup, | |
376 | * and our dlm_master_lookup() set it as the new master, along with | |
377 | * NEW_MASTER so that we'll recover it here (this implies dir_nodeid | |
378 | * equals our_nodeid below). | |
e7fd4179 DT |
379 | */ |
380 | ||
c04fecb4 | 381 | static int recover_master(struct dlm_rsb *r, unsigned int *count) |
e7fd4179 DT |
382 | { |
383 | struct dlm_ls *ls = r->res_ls; | |
c04fecb4 DT |
384 | int our_nodeid, dir_nodeid; |
385 | int is_removed = 0; | |
386 | int error; | |
387 | ||
388 | if (is_master(r)) | |
389 | return 0; | |
390 | ||
391 | is_removed = dlm_is_removed(ls, r->res_nodeid); | |
392 | ||
393 | if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER)) | |
394 | return 0; | |
395 | ||
396 | our_nodeid = dlm_our_nodeid(); | |
397 | dir_nodeid = dlm_dir_nodeid(r); | |
e7fd4179 DT |
398 | |
399 | if (dir_nodeid == our_nodeid) { | |
c04fecb4 DT |
400 | if (is_removed) { |
401 | r->res_master_nodeid = our_nodeid; | |
402 | r->res_nodeid = 0; | |
403 | } | |
e7fd4179 | 404 | |
c04fecb4 DT |
405 | /* set master of lkbs to ourself when is_removed, or to |
406 | another new master which we set along with NEW_MASTER | |
407 | in dlm_master_lookup */ | |
408 | set_new_master(r); | |
409 | error = 0; | |
e7fd4179 DT |
410 | } else { |
411 | recover_list_add(r); | |
412 | error = dlm_send_rcom_lookup(r, dir_nodeid); | |
413 | } | |
414 | ||
c04fecb4 | 415 | (*count)++; |
e7fd4179 DT |
416 | return error; |
417 | } | |
418 | ||
419 | /* | |
4875647a DT |
420 | * All MSTCPY locks are purged and rebuilt, even if the master stayed the same. |
421 | * This is necessary because recovery can be started, aborted and restarted, | |
422 | * causing the master nodeid to briefly change during the aborted recovery, and | |
423 | * change back to the original value in the second recovery. The MSTCPY locks | |
424 | * may or may not have been purged during the aborted recovery. Another node | |
425 | * with an outstanding request in waiters list and a request reply saved in the | |
426 | * requestqueue, cannot know whether it should ignore the reply and resend the | |
427 | * request, or accept the reply and complete the request. It must do the | |
428 | * former if the remote node purged MSTCPY locks, and it must do the later if | |
429 | * the remote node did not. This is solved by always purging MSTCPY locks, in | |
430 | * which case, the request reply would always be ignored and the request | |
431 | * resent. | |
e7fd4179 DT |
432 | */ |
433 | ||
c04fecb4 | 434 | static int recover_master_static(struct dlm_rsb *r, unsigned int *count) |
e7fd4179 | 435 | { |
4875647a DT |
436 | int dir_nodeid = dlm_dir_nodeid(r); |
437 | int new_master = dir_nodeid; | |
e7fd4179 | 438 | |
4875647a DT |
439 | if (dir_nodeid == dlm_our_nodeid()) |
440 | new_master = 0; | |
e7fd4179 | 441 | |
4875647a | 442 | dlm_purge_mstcpy_locks(r); |
c04fecb4 DT |
443 | r->res_master_nodeid = dir_nodeid; |
444 | r->res_nodeid = new_master; | |
445 | set_new_master(r); | |
446 | (*count)++; | |
447 | return 0; | |
e7fd4179 DT |
448 | } |
449 | ||
450 | /* | |
451 | * Go through local root resources and for each rsb which has a master which | |
452 | * has departed, get the new master nodeid from the directory. The dir will | |
453 | * assign mastery to the first node to look up the new master. That means | |
454 | * we'll discover in this lookup if we're the new master of any rsb's. | |
455 | * | |
456 | * We fire off all the dir lookup requests individually and asynchronously to | |
457 | * the correct dir node. | |
458 | */ | |
459 | ||
460 | int dlm_recover_masters(struct dlm_ls *ls) | |
461 | { | |
462 | struct dlm_rsb *r; | |
c04fecb4 DT |
463 | unsigned int total = 0; |
464 | unsigned int count = 0; | |
465 | int nodir = dlm_no_directory(ls); | |
466 | int error; | |
e7fd4179 DT |
467 | |
468 | log_debug(ls, "dlm_recover_masters"); | |
469 | ||
470 | down_read(&ls->ls_root_sem); | |
471 | list_for_each_entry(r, &ls->ls_root_list, res_root_list) { | |
472 | if (dlm_recovery_stopped(ls)) { | |
473 | up_read(&ls->ls_root_sem); | |
474 | error = -EINTR; | |
475 | goto out; | |
476 | } | |
477 | ||
c04fecb4 DT |
478 | lock_rsb(r); |
479 | if (nodir) | |
480 | error = recover_master_static(r, &count); | |
481 | else | |
482 | error = recover_master(r, &count); | |
483 | unlock_rsb(r); | |
484 | cond_resched(); | |
485 | total++; | |
e7fd4179 | 486 | |
c04fecb4 DT |
487 | if (error) { |
488 | up_read(&ls->ls_root_sem); | |
489 | goto out; | |
490 | } | |
e7fd4179 DT |
491 | } |
492 | up_read(&ls->ls_root_sem); | |
493 | ||
c04fecb4 | 494 | log_debug(ls, "dlm_recover_masters %u of %u", count, total); |
e7fd4179 DT |
495 | |
496 | error = dlm_wait_function(ls, &recover_list_empty); | |
497 | out: | |
498 | if (error) | |
499 | recover_list_clear(ls); | |
500 | return error; | |
501 | } | |
502 | ||
503 | int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc) | |
504 | { | |
505 | struct dlm_rsb *r; | |
c04fecb4 | 506 | int ret_nodeid, new_master; |
e7fd4179 DT |
507 | |
508 | r = recover_list_find(ls, rc->rc_id); | |
509 | if (!r) { | |
90135925 | 510 | log_error(ls, "dlm_recover_master_reply no id %llx", |
9229f013 | 511 | (unsigned long long)rc->rc_id); |
e7fd4179 DT |
512 | goto out; |
513 | } | |
514 | ||
c04fecb4 DT |
515 | ret_nodeid = rc->rc_result; |
516 | ||
517 | if (ret_nodeid == dlm_our_nodeid()) | |
518 | new_master = 0; | |
519 | else | |
520 | new_master = ret_nodeid; | |
e7fd4179 | 521 | |
4875647a | 522 | lock_rsb(r); |
c04fecb4 DT |
523 | r->res_master_nodeid = ret_nodeid; |
524 | r->res_nodeid = new_master; | |
525 | set_new_master(r); | |
4875647a | 526 | unlock_rsb(r); |
e7fd4179 DT |
527 | recover_list_del(r); |
528 | ||
529 | if (recover_list_empty(ls)) | |
530 | wake_up(&ls->ls_wait_general); | |
531 | out: | |
532 | return 0; | |
533 | } | |
534 | ||
535 | ||
536 | /* Lock recovery: rebuild the process-copy locks we hold on a | |
537 | remastered rsb on the new rsb master. | |
538 | ||
539 | dlm_recover_locks | |
540 | recover_locks | |
541 | recover_locks_queue | |
542 | dlm_send_rcom_lock -> receive_rcom_lock | |
543 | dlm_recover_master_copy | |
544 | receive_rcom_lock_reply <- | |
545 | dlm_recover_process_copy | |
546 | */ | |
547 | ||
548 | ||
549 | /* | |
550 | * keep a count of the number of lkb's we send to the new master; when we get | |
551 | * an equal number of replies then recovery for the rsb is done | |
552 | */ | |
553 | ||
554 | static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head) | |
555 | { | |
556 | struct dlm_lkb *lkb; | |
557 | int error = 0; | |
558 | ||
559 | list_for_each_entry(lkb, head, lkb_statequeue) { | |
560 | error = dlm_send_rcom_lock(r, lkb); | |
561 | if (error) | |
562 | break; | |
563 | r->res_recover_locks_count++; | |
564 | } | |
565 | ||
566 | return error; | |
567 | } | |
568 | ||
e7fd4179 DT |
569 | static int recover_locks(struct dlm_rsb *r) |
570 | { | |
571 | int error = 0; | |
572 | ||
573 | lock_rsb(r); | |
e7fd4179 | 574 | |
a345da3e | 575 | DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r);); |
e7fd4179 DT |
576 | |
577 | error = recover_locks_queue(r, &r->res_grantqueue); | |
578 | if (error) | |
579 | goto out; | |
580 | error = recover_locks_queue(r, &r->res_convertqueue); | |
581 | if (error) | |
582 | goto out; | |
583 | error = recover_locks_queue(r, &r->res_waitqueue); | |
584 | if (error) | |
585 | goto out; | |
586 | ||
587 | if (r->res_recover_locks_count) | |
588 | recover_list_add(r); | |
589 | else | |
590 | rsb_clear_flag(r, RSB_NEW_MASTER); | |
591 | out: | |
592 | unlock_rsb(r); | |
593 | return error; | |
594 | } | |
595 | ||
596 | int dlm_recover_locks(struct dlm_ls *ls) | |
597 | { | |
598 | struct dlm_rsb *r; | |
599 | int error, count = 0; | |
600 | ||
e7fd4179 DT |
601 | down_read(&ls->ls_root_sem); |
602 | list_for_each_entry(r, &ls->ls_root_list, res_root_list) { | |
603 | if (is_master(r)) { | |
604 | rsb_clear_flag(r, RSB_NEW_MASTER); | |
605 | continue; | |
606 | } | |
607 | ||
608 | if (!rsb_flag(r, RSB_NEW_MASTER)) | |
609 | continue; | |
610 | ||
611 | if (dlm_recovery_stopped(ls)) { | |
612 | error = -EINTR; | |
613 | up_read(&ls->ls_root_sem); | |
614 | goto out; | |
615 | } | |
616 | ||
617 | error = recover_locks(r); | |
618 | if (error) { | |
619 | up_read(&ls->ls_root_sem); | |
620 | goto out; | |
621 | } | |
622 | ||
623 | count += r->res_recover_locks_count; | |
624 | } | |
625 | up_read(&ls->ls_root_sem); | |
626 | ||
4875647a | 627 | log_debug(ls, "dlm_recover_locks %d out", count); |
e7fd4179 DT |
628 | |
629 | error = dlm_wait_function(ls, &recover_list_empty); | |
630 | out: | |
631 | if (error) | |
632 | recover_list_clear(ls); | |
e7fd4179 DT |
633 | return error; |
634 | } | |
635 | ||
636 | void dlm_recovered_lock(struct dlm_rsb *r) | |
637 | { | |
a345da3e | 638 | DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r);); |
e7fd4179 DT |
639 | |
640 | r->res_recover_locks_count--; | |
641 | if (!r->res_recover_locks_count) { | |
642 | rsb_clear_flag(r, RSB_NEW_MASTER); | |
643 | recover_list_del(r); | |
644 | } | |
645 | ||
646 | if (recover_list_empty(r->res_ls)) | |
647 | wake_up(&r->res_ls->ls_wait_general); | |
648 | } | |
649 | ||
650 | /* | |
651 | * The lvb needs to be recovered on all master rsb's. This includes setting | |
652 | * the VALNOTVALID flag if necessary, and determining the correct lvb contents | |
653 | * based on the lvb's of the locks held on the rsb. | |
654 | * | |
655 | * RSB_VALNOTVALID is set if there are only NL/CR locks on the rsb. If it | |
656 | * was already set prior to recovery, it's not cleared, regardless of locks. | |
657 | * | |
658 | * The LVB contents are only considered for changing when this is a new master | |
659 | * of the rsb (NEW_MASTER2). Then, the rsb's lvb is taken from any lkb with | |
660 | * mode > CR. If no lkb's exist with mode above CR, the lvb contents are taken | |
661 | * from the lkb with the largest lvb sequence number. | |
662 | */ | |
663 | ||
664 | static void recover_lvb(struct dlm_rsb *r) | |
665 | { | |
666 | struct dlm_lkb *lkb, *high_lkb = NULL; | |
667 | uint32_t high_seq = 0; | |
90135925 DT |
668 | int lock_lvb_exists = 0; |
669 | int big_lock_exists = 0; | |
e7fd4179 DT |
670 | int lvblen = r->res_ls->ls_lvblen; |
671 | ||
672 | list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) { | |
673 | if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) | |
674 | continue; | |
675 | ||
90135925 | 676 | lock_lvb_exists = 1; |
e7fd4179 DT |
677 | |
678 | if (lkb->lkb_grmode > DLM_LOCK_CR) { | |
90135925 | 679 | big_lock_exists = 1; |
e7fd4179 DT |
680 | goto setflag; |
681 | } | |
682 | ||
683 | if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) { | |
684 | high_lkb = lkb; | |
685 | high_seq = lkb->lkb_lvbseq; | |
686 | } | |
687 | } | |
688 | ||
689 | list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) { | |
690 | if (!(lkb->lkb_exflags & DLM_LKF_VALBLK)) | |
691 | continue; | |
692 | ||
90135925 | 693 | lock_lvb_exists = 1; |
e7fd4179 DT |
694 | |
695 | if (lkb->lkb_grmode > DLM_LOCK_CR) { | |
90135925 | 696 | big_lock_exists = 1; |
e7fd4179 DT |
697 | goto setflag; |
698 | } | |
699 | ||
700 | if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) { | |
701 | high_lkb = lkb; | |
702 | high_seq = lkb->lkb_lvbseq; | |
703 | } | |
704 | } | |
705 | ||
706 | setflag: | |
707 | if (!lock_lvb_exists) | |
708 | goto out; | |
709 | ||
710 | if (!big_lock_exists) | |
711 | rsb_set_flag(r, RSB_VALNOTVALID); | |
712 | ||
713 | /* don't mess with the lvb unless we're the new master */ | |
714 | if (!rsb_flag(r, RSB_NEW_MASTER2)) | |
715 | goto out; | |
716 | ||
717 | if (!r->res_lvbptr) { | |
52bda2b5 | 718 | r->res_lvbptr = dlm_allocate_lvb(r->res_ls); |
e7fd4179 DT |
719 | if (!r->res_lvbptr) |
720 | goto out; | |
721 | } | |
722 | ||
723 | if (big_lock_exists) { | |
724 | r->res_lvbseq = lkb->lkb_lvbseq; | |
725 | memcpy(r->res_lvbptr, lkb->lkb_lvbptr, lvblen); | |
726 | } else if (high_lkb) { | |
727 | r->res_lvbseq = high_lkb->lkb_lvbseq; | |
728 | memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen); | |
729 | } else { | |
730 | r->res_lvbseq = 0; | |
731 | memset(r->res_lvbptr, 0, lvblen); | |
732 | } | |
733 | out: | |
734 | return; | |
735 | } | |
736 | ||
737 | /* All master rsb's flagged RECOVER_CONVERT need to be looked at. The locks | |
738 | converting PR->CW or CW->PR need to have their lkb_grmode set. */ | |
739 | ||
740 | static void recover_conversion(struct dlm_rsb *r) | |
741 | { | |
742 | struct dlm_lkb *lkb; | |
743 | int grmode = -1; | |
744 | ||
745 | list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) { | |
746 | if (lkb->lkb_grmode == DLM_LOCK_PR || | |
747 | lkb->lkb_grmode == DLM_LOCK_CW) { | |
748 | grmode = lkb->lkb_grmode; | |
749 | break; | |
750 | } | |
751 | } | |
752 | ||
753 | list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) { | |
754 | if (lkb->lkb_grmode != DLM_LOCK_IV) | |
755 | continue; | |
756 | if (grmode == -1) | |
757 | lkb->lkb_grmode = lkb->lkb_rqmode; | |
758 | else | |
759 | lkb->lkb_grmode = grmode; | |
760 | } | |
761 | } | |
762 | ||
f7da790d | 763 | /* We've become the new master for this rsb and waiting/converting locks may |
4875647a | 764 | need to be granted in dlm_recover_grant() due to locks that may have |
f7da790d DT |
765 | existed from a removed node. */ |
766 | ||
4875647a | 767 | static void recover_grant(struct dlm_rsb *r) |
f7da790d DT |
768 | { |
769 | if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue)) | |
4875647a | 770 | rsb_set_flag(r, RSB_RECOVER_GRANT); |
f7da790d DT |
771 | } |
772 | ||
e7fd4179 DT |
773 | void dlm_recover_rsbs(struct dlm_ls *ls) |
774 | { | |
775 | struct dlm_rsb *r; | |
4875647a | 776 | unsigned int count = 0; |
e7fd4179 DT |
777 | |
778 | down_read(&ls->ls_root_sem); | |
779 | list_for_each_entry(r, &ls->ls_root_list, res_root_list) { | |
780 | lock_rsb(r); | |
781 | if (is_master(r)) { | |
782 | if (rsb_flag(r, RSB_RECOVER_CONVERT)) | |
783 | recover_conversion(r); | |
f7da790d | 784 | if (rsb_flag(r, RSB_NEW_MASTER2)) |
4875647a | 785 | recover_grant(r); |
e7fd4179 DT |
786 | recover_lvb(r); |
787 | count++; | |
788 | } | |
789 | rsb_clear_flag(r, RSB_RECOVER_CONVERT); | |
f7da790d | 790 | rsb_clear_flag(r, RSB_NEW_MASTER2); |
e7fd4179 DT |
791 | unlock_rsb(r); |
792 | } | |
793 | up_read(&ls->ls_root_sem); | |
794 | ||
4875647a DT |
795 | if (count) |
796 | log_debug(ls, "dlm_recover_rsbs %d done", count); | |
e7fd4179 DT |
797 | } |
798 | ||
799 | /* Create a single list of all root rsb's to be used during recovery */ | |
800 | ||
801 | int dlm_create_root_list(struct dlm_ls *ls) | |
802 | { | |
9beb3bf5 | 803 | struct rb_node *n; |
e7fd4179 DT |
804 | struct dlm_rsb *r; |
805 | int i, error = 0; | |
806 | ||
807 | down_write(&ls->ls_root_sem); | |
808 | if (!list_empty(&ls->ls_root_list)) { | |
809 | log_error(ls, "root list not empty"); | |
810 | error = -EINVAL; | |
811 | goto out; | |
812 | } | |
813 | ||
814 | for (i = 0; i < ls->ls_rsbtbl_size; i++) { | |
c7be761a | 815 | spin_lock(&ls->ls_rsbtbl[i].lock); |
9beb3bf5 BP |
816 | for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) { |
817 | r = rb_entry(n, struct dlm_rsb, res_hashnode); | |
e7fd4179 DT |
818 | list_add(&r->res_root_list, &ls->ls_root_list); |
819 | dlm_hold_rsb(r); | |
820 | } | |
85f0379a | 821 | |
c04fecb4 DT |
822 | if (!RB_EMPTY_ROOT(&ls->ls_rsbtbl[i].toss)) |
823 | log_error(ls, "dlm_create_root_list toss not empty"); | |
c7be761a | 824 | spin_unlock(&ls->ls_rsbtbl[i].lock); |
e7fd4179 DT |
825 | } |
826 | out: | |
827 | up_write(&ls->ls_root_sem); | |
828 | return error; | |
829 | } | |
830 | ||
831 | void dlm_release_root_list(struct dlm_ls *ls) | |
832 | { | |
833 | struct dlm_rsb *r, *safe; | |
834 | ||
835 | down_write(&ls->ls_root_sem); | |
836 | list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) { | |
837 | list_del_init(&r->res_root_list); | |
838 | dlm_put_rsb(r); | |
839 | } | |
840 | up_write(&ls->ls_root_sem); | |
841 | } | |
842 | ||
c04fecb4 | 843 | void dlm_clear_toss(struct dlm_ls *ls) |
e7fd4179 | 844 | { |
9beb3bf5 | 845 | struct rb_node *n, *next; |
c04fecb4 DT |
846 | struct dlm_rsb *r; |
847 | unsigned int count = 0; | |
e7fd4179 DT |
848 | int i; |
849 | ||
850 | for (i = 0; i < ls->ls_rsbtbl_size; i++) { | |
c7be761a | 851 | spin_lock(&ls->ls_rsbtbl[i].lock); |
9beb3bf5 | 852 | for (n = rb_first(&ls->ls_rsbtbl[i].toss); n; n = next) { |
c04fecb4 DT |
853 | next = rb_next(n); |
854 | r = rb_entry(n, struct dlm_rsb, res_hashnode); | |
855 | rb_erase(n, &ls->ls_rsbtbl[i].toss); | |
856 | dlm_free_rsb(r); | |
857 | count++; | |
e7fd4179 | 858 | } |
c7be761a | 859 | spin_unlock(&ls->ls_rsbtbl[i].lock); |
e7fd4179 | 860 | } |
c04fecb4 DT |
861 | |
862 | if (count) | |
863 | log_debug(ls, "dlm_clear_toss %u done", count); | |
e7fd4179 DT |
864 | } |
865 |