projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: gracefully handle stale create_lock messages.
[linux-2.6-block.git] / fs / ocfs2 / dlm / dlmmaster.c
CommitLineData
6714d8e8
KH		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmmod.c
5 *
6 * standalone DLM module
7 *
8 * Copyright (C) 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 *
25 */
26
27
28#include <linux/module.h>
29#include <linux/fs.h>
30#include <linux/types.h>
31#include <linux/slab.h>
32#include <linux/highmem.h>
33#include <linux/utsname.h>
34#include <linux/init.h>
35#include <linux/sysctl.h>
36#include <linux/random.h>
37#include <linux/blkdev.h>
38#include <linux/socket.h>
39#include <linux/inet.h>
40#include <linux/spinlock.h>
41#include <linux/delay.h>
42
43
44#include "cluster/heartbeat.h"
45#include "cluster/nodemanager.h"
46#include "cluster/tcp.h"
47
48#include "dlmapi.h"
49#include "dlmcommon.h"
50#include "dlmdebug.h"
82353b59 51#include "dlmdomain.h"
6714d8e8
KH		 52
53#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
54#include "cluster/masklog.h"
55
56enum dlm_mle_type {
57 DLM_MLE_BLOCK,
58 DLM_MLE_MASTER,
59 DLM_MLE_MIGRATION
60};
61
62struct dlm_lock_name
63{
64 u8 len;
65 u8 name[DLM_LOCKID_NAME_MAX];
66};
67
68struct dlm_master_list_entry
69{
70 struct list_head list;
71 struct list_head hb_events;
72 struct dlm_ctxt *dlm;
73 spinlock_t spinlock;
74 wait_queue_head_t wq;
75 atomic_t woken;
76 struct kref mle_refs;
a2bf0477 77 int inuse;
6714d8e8
KH		 78 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
79 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
80 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
81 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
82 u8 master;
83 u8 new_master;
84 enum dlm_mle_type type;
85 struct o2hb_callback_func mle_hb_up;
86 struct o2hb_callback_func mle_hb_down;
87 union {
88 struct dlm_lock_resource *res;
89 struct dlm_lock_name name;
90 } u;
91};
92
93static void dlm_mle_node_down(struct dlm_ctxt *dlm,
94 struct dlm_master_list_entry *mle,
95 struct o2nm_node *node,
96 int idx);
97static void dlm_mle_node_up(struct dlm_ctxt *dlm,
98 struct dlm_master_list_entry *mle,
99 struct o2nm_node *node,
100 int idx);
101
102static void dlm_assert_master_worker(struct dlm_work_item *item, void *data);
103static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
104 unsigned int namelen, void *nodemap,
105 u32 flags);
106
107static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
108 struct dlm_master_list_entry *mle,
109 const char *name,
110 unsigned int namelen)
111{
112 struct dlm_lock_resource *res;
113
114 if (dlm != mle->dlm)
115 return 0;
116
117 if (mle->type == DLM_MLE_BLOCK ||
118 mle->type == DLM_MLE_MIGRATION) {
119 if (namelen != mle->u.name.len ||
120 memcmp(name, mle->u.name.name, namelen)!=0)
121 return 0;
122 } else {
123 res = mle->u.res;
124 if (namelen != res->lockname.len ||
125 memcmp(res->lockname.name, name, namelen) != 0)
126 return 0;
127 }
128 return 1;
129}
130
131#if 0
132/* Code here is included but defined out as it aids debugging */
133
95883719
KH		 134#define dlm_print_nodemap(m) _dlm_print_nodemap(m,#m)
135void _dlm_print_nodemap(unsigned long *map, const char *mapname)
136{
137 int i;
138 printk("%s=[ ", mapname);
139 for (i=0; i<O2NM_MAX_NODES; i++)
140 if (test_bit(i, map))
141 printk("%d ", i);
142 printk("]");
143}
144
6714d8e8
KH		 145void dlm_print_one_mle(struct dlm_master_list_entry *mle)
146{
95883719 147 int refs;
6714d8e8
KH		 148 char *type;
149 char attached;
150 u8 master;
151 unsigned int namelen;
152 const char *name;
153 struct kref *k;
95883719
KH		 154 unsigned long *maybe = mle->maybe_map,
155 *vote = mle->vote_map,
156 *resp = mle->response_map,
157 *node = mle->node_map;
6714d8e8
KH		 158
159 k = &mle->mle_refs;
160 if (mle->type == DLM_MLE_BLOCK)
161 type = "BLK";
162 else if (mle->type == DLM_MLE_MASTER)
163 type = "MAS";
164 else
165 type = "MIG";
166 refs = atomic_read(&k->refcount);
167 master = mle->master;
168 attached = (list_empty(&mle->hb_events) ? 'N' : 'Y');
169
170 if (mle->type != DLM_MLE_MASTER) {
171 namelen = mle->u.name.len;
172 name = mle->u.name.name;
173 } else {
174 namelen = mle->u.res->lockname.len;
175 name = mle->u.res->lockname.name;
176 }
177
95883719
KH		 178 mlog(ML_NOTICE, "%.*s: %3s refs=%3d mas=%3u new=%3u evt=%c inuse=%d ",
179 namelen, name, type, refs, master, mle->new_master, attached,
180 mle->inuse);
181 dlm_print_nodemap(maybe);
182 printk(", ");
183 dlm_print_nodemap(vote);
184 printk(", ");
185 dlm_print_nodemap(resp);
186 printk(", ");
187 dlm_print_nodemap(node);
188 printk(", ");
189 printk("\n");
6714d8e8
KH		 190}
191
192static void dlm_dump_mles(struct dlm_ctxt *dlm)
193{
194 struct dlm_master_list_entry *mle;
195 struct list_head *iter;
196
197 mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name);
6714d8e8
KH		 198 spin_lock(&dlm->master_lock);
199 list_for_each(iter, &dlm->master_list) {
200 mle = list_entry(iter, struct dlm_master_list_entry, list);
201 dlm_print_one_mle(mle);
202 }
203 spin_unlock(&dlm->master_lock);
204}
205
6714d8e8
KH		 206int dlm_dump_all_mles(const char __user *data, unsigned int len)
207{
208 struct list_head *iter;
209 struct dlm_ctxt *dlm;
210
211 spin_lock(&dlm_domain_lock);
212 list_for_each(iter, &dlm_domains) {
213 dlm = list_entry (iter, struct dlm_ctxt, list);
214 mlog(ML_NOTICE, "found dlm: %p, name=%s\n", dlm, dlm->name);
215 dlm_dump_mles(dlm);
216 }
217 spin_unlock(&dlm_domain_lock);
218 return len;
219}
220EXPORT_SYMBOL_GPL(dlm_dump_all_mles);
221
222#endif /* 0 */
223
224
225static kmem_cache_t *dlm_mle_cache = NULL;
226
227
228static void dlm_mle_release(struct kref *kref);
229static void dlm_init_mle(struct dlm_master_list_entry *mle,
230 enum dlm_mle_type type,
231 struct dlm_ctxt *dlm,
232 struct dlm_lock_resource *res,
233 const char *name,
234 unsigned int namelen);
235static void dlm_put_mle(struct dlm_master_list_entry *mle);
236static void __dlm_put_mle(struct dlm_master_list_entry *mle);
237static int dlm_find_mle(struct dlm_ctxt *dlm,
238 struct dlm_master_list_entry **mle,
239 char *name, unsigned int namelen);
240
241static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to);
242
243
244static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
245 struct dlm_lock_resource *res,
246 struct dlm_master_list_entry *mle,
247 int *blocked);
248static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
249 struct dlm_lock_resource *res,
250 struct dlm_master_list_entry *mle,
251 int blocked);
252static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
253 struct dlm_lock_resource *res,
254 struct dlm_master_list_entry *mle,
255 struct dlm_master_list_entry **oldmle,
256 const char *name, unsigned int namelen,
257 u8 new_master, u8 master);
258
259static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
260 struct dlm_lock_resource *res);
261static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
262 struct dlm_lock_resource *res);
263static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
264 struct dlm_lock_resource *res,
265 u8 target);
c03872f5
KH		 266static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
267 struct dlm_lock_resource *res);
6714d8e8
KH		 268
269
270int dlm_is_host_down(int errno)
271{
272 switch (errno) {
273 case -EBADF:
274 case -ECONNREFUSED:
275 case -ENOTCONN:
276 case -ECONNRESET:
277 case -EPIPE:
278 case -EHOSTDOWN:
279 case -EHOSTUNREACH:
280 case -ETIMEDOUT:
281 case -ECONNABORTED:
282 case -ENETDOWN:
283 case -ENETUNREACH:
284 case -ENETRESET:
285 case -ESHUTDOWN:
286 case -ENOPROTOOPT:
287 case -EINVAL: /* if returned from our tcp code,
288 this means there is no socket */
289 return 1;
290 }
291 return 0;
292}
293
294
295/*
296 * MASTER LIST FUNCTIONS
297 */
298
299
300/*
301 * regarding master list entries and heartbeat callbacks:
302 *
303 * in order to avoid sleeping and allocation that occurs in
304 * heartbeat, master list entries are simply attached to the
305 * dlm's established heartbeat callbacks. the mle is attached
306 * when it is created, and since the dlm->spinlock is held at
307 * that time, any heartbeat event will be properly discovered
308 * by the mle. the mle needs to be detached from the
309 * dlm->mle_hb_events list as soon as heartbeat events are no
310 * longer useful to the mle, and before the mle is freed.
311 *
312 * as a general rule, heartbeat events are no longer needed by
313 * the mle once an "answer" regarding the lock master has been
314 * received.
315 */
316static inline void __dlm_mle_attach_hb_events(struct dlm_ctxt *dlm,
317 struct dlm_master_list_entry *mle)
318{
319 assert_spin_locked(&dlm->spinlock);
320
321 list_add_tail(&mle->hb_events, &dlm->mle_hb_events);
322}
323
324
325static inline void __dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
326 struct dlm_master_list_entry *mle)
327{
328 if (!list_empty(&mle->hb_events))
329 list_del_init(&mle->hb_events);
330}
331
332
333static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
334 struct dlm_master_list_entry *mle)
335{
336 spin_lock(&dlm->spinlock);
337 __dlm_mle_detach_hb_events(dlm, mle);
338 spin_unlock(&dlm->spinlock);
339}
340
a2bf0477
KH		 341static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle)
342{
343 struct dlm_ctxt *dlm;
344 dlm = mle->dlm;
345
346 assert_spin_locked(&dlm->spinlock);
347 assert_spin_locked(&dlm->master_lock);
348 mle->inuse++;
349 kref_get(&mle->mle_refs);
350}
351
352static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle)
353{
354 struct dlm_ctxt *dlm;
355 dlm = mle->dlm;
356
357 spin_lock(&dlm->spinlock);
358 spin_lock(&dlm->master_lock);
359 mle->inuse--;
360 __dlm_put_mle(mle);
361 spin_unlock(&dlm->master_lock);
362 spin_unlock(&dlm->spinlock);
363
364}
365
6714d8e8
KH		 366/* remove from list and free */
367static void __dlm_put_mle(struct dlm_master_list_entry *mle)
368{
369 struct dlm_ctxt *dlm;
370 dlm = mle->dlm;
371
372 assert_spin_locked(&dlm->spinlock);
373 assert_spin_locked(&dlm->master_lock);
aa852354
KH		 374 if (!atomic_read(&mle->mle_refs.refcount)) {
375 /* this may or may not crash, but who cares.
376 * it's a BUG. */
377 mlog(ML_ERROR, "bad mle: %p\n", mle);
378 dlm_print_one_mle(mle);
379 BUG();
380 } else
381 kref_put(&mle->mle_refs, dlm_mle_release);
6714d8e8
KH		 382}
383
384
385/* must not have any spinlocks coming in */
386static void dlm_put_mle(struct dlm_master_list_entry *mle)
387{
388 struct dlm_ctxt *dlm;
389 dlm = mle->dlm;
390
391 spin_lock(&dlm->spinlock);
392 spin_lock(&dlm->master_lock);
393 __dlm_put_mle(mle);
394 spin_unlock(&dlm->master_lock);
395 spin_unlock(&dlm->spinlock);
396}
397
398static inline void dlm_get_mle(struct dlm_master_list_entry *mle)
399{
400 kref_get(&mle->mle_refs);
401}
402
403static void dlm_init_mle(struct dlm_master_list_entry *mle,
404 enum dlm_mle_type type,
405 struct dlm_ctxt *dlm,
406 struct dlm_lock_resource *res,
407 const char *name,
408 unsigned int namelen)
409{
410 assert_spin_locked(&dlm->spinlock);
411
412 mle->dlm = dlm;
413 mle->type = type;
414 INIT_LIST_HEAD(&mle->list);
415 INIT_LIST_HEAD(&mle->hb_events);
416 memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
417 spin_lock_init(&mle->spinlock);
418 init_waitqueue_head(&mle->wq);
419 atomic_set(&mle->woken, 0);
420 kref_init(&mle->mle_refs);
421 memset(mle->response_map, 0, sizeof(mle->response_map));
422 mle->master = O2NM_MAX_NODES;
423 mle->new_master = O2NM_MAX_NODES;
a2bf0477 424 mle->inuse = 0;
6714d8e8
KH		 425
426 if (mle->type == DLM_MLE_MASTER) {
427 BUG_ON(!res);
428 mle->u.res = res;
429 } else if (mle->type == DLM_MLE_BLOCK) {
430 BUG_ON(!name);
431 memcpy(mle->u.name.name, name, namelen);
432 mle->u.name.len = namelen;
433 } else /* DLM_MLE_MIGRATION */ {
434 BUG_ON(!name);
435 memcpy(mle->u.name.name, name, namelen);
436 mle->u.name.len = namelen;
437 }
438
439 /* copy off the node_map and register hb callbacks on our copy */
440 memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map));
441 memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map));
442 clear_bit(dlm->node_num, mle->vote_map);
443 clear_bit(dlm->node_num, mle->node_map);
444
445 /* attach the mle to the domain node up/down events */
446 __dlm_mle_attach_hb_events(dlm, mle);
447}
448
449
450/* returns 1 if found, 0 if not */
451static int dlm_find_mle(struct dlm_ctxt *dlm,
452 struct dlm_master_list_entry **mle,
453 char *name, unsigned int namelen)
454{
455 struct dlm_master_list_entry *tmpmle;
456 struct list_head *iter;
457
458 assert_spin_locked(&dlm->master_lock);
459
460 list_for_each(iter, &dlm->master_list) {
461 tmpmle = list_entry(iter, struct dlm_master_list_entry, list);
462 if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
463 continue;
464 dlm_get_mle(tmpmle);
465 *mle = tmpmle;
466 return 1;
467 }
468 return 0;
469}
470
471void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up)
472{
473 struct dlm_master_list_entry *mle;
474 struct list_head *iter;
475
476 assert_spin_locked(&dlm->spinlock);
477
478 list_for_each(iter, &dlm->mle_hb_events) {
479 mle = list_entry(iter, struct dlm_master_list_entry,
480 hb_events);
481 if (node_up)
482 dlm_mle_node_up(dlm, mle, NULL, idx);
483 else
484 dlm_mle_node_down(dlm, mle, NULL, idx);
485 }
486}
487
488static void dlm_mle_node_down(struct dlm_ctxt *dlm,
489 struct dlm_master_list_entry *mle,
490 struct o2nm_node *node, int idx)
491{
492 spin_lock(&mle->spinlock);
493
494 if (!test_bit(idx, mle->node_map))
495 mlog(0, "node %u already removed from nodemap!\n", idx);
496 else
497 clear_bit(idx, mle->node_map);
498
499 spin_unlock(&mle->spinlock);
500}
501
502static void dlm_mle_node_up(struct dlm_ctxt *dlm,
503 struct dlm_master_list_entry *mle,
504 struct o2nm_node *node, int idx)
505{
506 spin_lock(&mle->spinlock);
507
508 if (test_bit(idx, mle->node_map))
509 mlog(0, "node %u already in node map!\n", idx);
510 else
511 set_bit(idx, mle->node_map);
512
513 spin_unlock(&mle->spinlock);
514}
515
516
517int dlm_init_mle_cache(void)
518{
519 dlm_mle_cache = kmem_cache_create("dlm_mle_cache",
520 sizeof(struct dlm_master_list_entry),
521 0, SLAB_HWCACHE_ALIGN,
522 NULL, NULL);
523 if (dlm_mle_cache == NULL)
524 return -ENOMEM;
525 return 0;
526}
527
528void dlm_destroy_mle_cache(void)
529{
530 if (dlm_mle_cache)
531 kmem_cache_destroy(dlm_mle_cache);
532}
533
534static void dlm_mle_release(struct kref *kref)
535{
536 struct dlm_master_list_entry *mle;
537 struct dlm_ctxt *dlm;
538
539 mlog_entry_void();
540
541 mle = container_of(kref, struct dlm_master_list_entry, mle_refs);
542 dlm = mle->dlm;
543
544 if (mle->type != DLM_MLE_MASTER) {
545 mlog(0, "calling mle_release for %.*s, type %d\n",
546 mle->u.name.len, mle->u.name.name, mle->type);
547 } else {
548 mlog(0, "calling mle_release for %.*s, type %d\n",
549 mle->u.res->lockname.len,
550 mle->u.res->lockname.name, mle->type);
551 }
552 assert_spin_locked(&dlm->spinlock);
553 assert_spin_locked(&dlm->master_lock);
554
555 /* remove from list if not already */
556 if (!list_empty(&mle->list))
557 list_del_init(&mle->list);
558
559 /* detach the mle from the domain node up/down events */
560 __dlm_mle_detach_hb_events(dlm, mle);
561
562 /* NOTE: kfree under spinlock here.
563 * if this is bad, we can move this to a freelist. */
564 kmem_cache_free(dlm_mle_cache, mle);
565}
566
567
568/*
569 * LOCK RESOURCE FUNCTIONS
570 */
571
572static void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
573 struct dlm_lock_resource *res,
574 u8 owner)
575{
576 assert_spin_locked(&res->spinlock);
577
578 mlog_entry("%.*s, %u\n", res->lockname.len, res->lockname.name, owner);
579
580 if (owner == dlm->node_num)
581 atomic_inc(&dlm->local_resources);
582 else if (owner == DLM_LOCK_RES_OWNER_UNKNOWN)
583 atomic_inc(&dlm->unknown_resources);
584 else
585 atomic_inc(&dlm->remote_resources);
586
587 res->owner = owner;
588}
589
590void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
591 struct dlm_lock_resource *res, u8 owner)
592{
593 assert_spin_locked(&res->spinlock);
594
595 if (owner == res->owner)
596 return;
597
598 if (res->owner == dlm->node_num)
599 atomic_dec(&dlm->local_resources);
600 else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN)
601 atomic_dec(&dlm->unknown_resources);
602 else
603 atomic_dec(&dlm->remote_resources);
604
605 dlm_set_lockres_owner(dlm, res, owner);
606}
607
608
609static void dlm_lockres_release(struct kref *kref)
610{
611 struct dlm_lock_resource *res;
612
613 res = container_of(kref, struct dlm_lock_resource, refs);
614
615 /* This should not happen -- all lockres' have a name
616 * associated with them at init time. */
617 BUG_ON(!res->lockname.name);
618
619 mlog(0, "destroying lockres %.*s\n", res->lockname.len,
620 res->lockname.name);
621
a7f90d83
KH		 622 if (!hlist_unhashed(&res->hash_node) ||
623 !list_empty(&res->granted) ||
624 !list_empty(&res->converting) ||
625 !list_empty(&res->blocked) ||
626 !list_empty(&res->dirty) ||
627 !list_empty(&res->recovering) ||
628 !list_empty(&res->purge)) {
629 mlog(ML_ERROR,
630 "Going to BUG for resource %.*s."
631 " We're on a list! [%c%c%c%c%c%c%c]\n",
632 res->lockname.len, res->lockname.name,
633 !hlist_unhashed(&res->hash_node) ? 'H' : ' ',
634 !list_empty(&res->granted) ? 'G' : ' ',
635 !list_empty(&res->converting) ? 'C' : ' ',
636 !list_empty(&res->blocked) ? 'B' : ' ',
637 !list_empty(&res->dirty) ? 'D' : ' ',
638 !list_empty(&res->recovering) ? 'R' : ' ',
639 !list_empty(&res->purge) ? 'P' : ' ');
640
641 dlm_print_one_lock_resource(res);
642 }
643
6714d8e8
KH		 644 /* By the time we're ready to blow this guy away, we shouldn't
645 * be on any lists. */
81f2094a 646 BUG_ON(!hlist_unhashed(&res->hash_node));
6714d8e8
KH		 647 BUG_ON(!list_empty(&res->granted));
648 BUG_ON(!list_empty(&res->converting));
649 BUG_ON(!list_empty(&res->blocked));
650 BUG_ON(!list_empty(&res->dirty));
651 BUG_ON(!list_empty(&res->recovering));
652 BUG_ON(!list_empty(&res->purge));
653
654 kfree(res->lockname.name);
655
656 kfree(res);
657}
658
6714d8e8
KH		 659void dlm_lockres_put(struct dlm_lock_resource *res)
660{
661 kref_put(&res->refs, dlm_lockres_release);
662}
663
664static void dlm_init_lockres(struct dlm_ctxt *dlm,
665 struct dlm_lock_resource *res,
666 const char *name, unsigned int namelen)
667{
668 char *qname;
669
670 /* If we memset here, we lose our reference to the kmalloc'd
671 * res->lockname.name, so be sure to init every field
672 * correctly! */
673
674 qname = (char *) res->lockname.name;
675 memcpy(qname, name, namelen);
676
677 res->lockname.len = namelen;
a3d33291 678 res->lockname.hash = dlm_lockid_hash(name, namelen);
6714d8e8
KH		 679
680 init_waitqueue_head(&res->wq);
681 spin_lock_init(&res->spinlock);
81f2094a 682 INIT_HLIST_NODE(&res->hash_node);
6714d8e8
KH		 683 INIT_LIST_HEAD(&res->granted);
684 INIT_LIST_HEAD(&res->converting);
685 INIT_LIST_HEAD(&res->blocked);
686 INIT_LIST_HEAD(&res->dirty);
687 INIT_LIST_HEAD(&res->recovering);
688 INIT_LIST_HEAD(&res->purge);
689 atomic_set(&res->asts_reserved, 0);
690 res->migration_pending = 0;
691
692 kref_init(&res->refs);
693
694 /* just for consistency */
695 spin_lock(&res->spinlock);
696 dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
697 spin_unlock(&res->spinlock);
698
699 res->state = DLM_LOCK_RES_IN_PROGRESS;
700
701 res->last_used = 0;
702
703 memset(res->lvb, 0, DLM_LVB_LEN);
704}
705
706struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
707 const char *name,
708 unsigned int namelen)
709{
710 struct dlm_lock_resource *res;
711
712 res = kmalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
713 if (!res)
714 return NULL;
715
716 res->lockname.name = kmalloc(namelen, GFP_KERNEL);
717 if (!res->lockname.name) {
718 kfree(res);
719 return NULL;
720 }
721
722 dlm_init_lockres(dlm, res, name, namelen);
723 return res;
724}
725
726/*
727 * lookup a lock resource by name.
728 * may already exist in the hashtable.
729 * lockid is null terminated
730 *
731 * if not, allocate enough for the lockres and for
732 * the temporary structure used in doing the mastering.
733 *
734 * also, do a lookup in the dlm->master_list to see
735 * if another node has begun mastering the same lock.
736 * if so, there should be a block entry in there
737 * for this name, and we should *not* attempt to master
738 * the lock here. need to wait around for that node
739 * to assert_master (or die).
740 *
741 */
742struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
743 const char *lockid,
744 int flags)
745{
746 struct dlm_lock_resource *tmpres=NULL, *res=NULL;
747 struct dlm_master_list_entry *mle = NULL;
748 struct dlm_master_list_entry *alloc_mle = NULL;
749 int blocked = 0;
750 int ret, nodenum;
751 struct dlm_node_iter iter;
a3d33291 752 unsigned int namelen, hash;
6714d8e8 753 int tries = 0;
c03872f5 754 int bit, wait_on_recovery = 0;
6714d8e8
KH		 755
756 BUG_ON(!lockid);
757
758 namelen = strlen(lockid);
a3d33291 759 hash = dlm_lockid_hash(lockid, namelen);
6714d8e8
KH		 760
761 mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
762
763lookup:
764 spin_lock(&dlm->spinlock);
a3d33291 765 tmpres = __dlm_lookup_lockres(dlm, lockid, namelen, hash);
6714d8e8
KH		 766 if (tmpres) {
767 spin_unlock(&dlm->spinlock);
768 mlog(0, "found in hash!\n");
769 if (res)
770 dlm_lockres_put(res);
771 res = tmpres;
772 goto leave;
773 }
774
775 if (!res) {
776 spin_unlock(&dlm->spinlock);
777 mlog(0, "allocating a new resource\n");
778 /* nothing found and we need to allocate one. */
779 alloc_mle = (struct dlm_master_list_entry *)
780 kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
781 if (!alloc_mle)
782 goto leave;
783 res = dlm_new_lockres(dlm, lockid, namelen);
784 if (!res)
785 goto leave;
786 goto lookup;
787 }
788
789 mlog(0, "no lockres found, allocated our own: %p\n", res);
790
791 if (flags & LKM_LOCAL) {
792 /* caller knows it's safe to assume it's not mastered elsewhere
793 * DONE! return right away */
794 spin_lock(&res->spinlock);
795 dlm_change_lockres_owner(dlm, res, dlm->node_num);
796 __dlm_insert_lockres(dlm, res);
797 spin_unlock(&res->spinlock);
798 spin_unlock(&dlm->spinlock);
799 /* lockres still marked IN_PROGRESS */
800 goto wake_waiters;
801 }
802
803 /* check master list to see if another node has started mastering it */
804 spin_lock(&dlm->master_lock);
805
806 /* if we found a block, wait for lock to be mastered by another node */
807 blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen);
808 if (blocked) {
809 if (mle->type == DLM_MLE_MASTER) {
810 mlog(ML_ERROR, "master entry for nonexistent lock!\n");
811 BUG();
812 } else if (mle->type == DLM_MLE_MIGRATION) {
813 /* migration is in progress! */
814 /* the good news is that we now know the
815 * "current" master (mle->master). */
816
817 spin_unlock(&dlm->master_lock);
818 assert_spin_locked(&dlm->spinlock);
819
820 /* set the lockres owner and hash it */
821 spin_lock(&res->spinlock);
822 dlm_set_lockres_owner(dlm, res, mle->master);
823 __dlm_insert_lockres(dlm, res);
824 spin_unlock(&res->spinlock);
825 spin_unlock(&dlm->spinlock);
826
827 /* master is known, detach */
828 dlm_mle_detach_hb_events(dlm, mle);
829 dlm_put_mle(mle);
830 mle = NULL;
831 goto wake_waiters;
832 }
833 } else {
834 /* go ahead and try to master lock on this node */
835 mle = alloc_mle;
836 /* make sure this does not get freed below */
837 alloc_mle = NULL;
838 dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
839 set_bit(dlm->node_num, mle->maybe_map);
840 list_add(&mle->list, &dlm->master_list);
c03872f5
KH		 841
842 /* still holding the dlm spinlock, check the recovery map
843 * to see if there are any nodes that still need to be
844 * considered. these will not appear in the mle nodemap
845 * but they might own this lockres. wait on them. */
846 bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
847 if (bit < O2NM_MAX_NODES) {
848 mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
849 "recover before lock mastery can begin\n",
850 dlm->name, namelen, (char *)lockid, bit);
851 wait_on_recovery = 1;
852 }
6714d8e8
KH		 853 }
854
855 /* at this point there is either a DLM_MLE_BLOCK or a
856 * DLM_MLE_MASTER on the master list, so it's safe to add the
857 * lockres to the hashtable. anyone who finds the lock will
858 * still have to wait on the IN_PROGRESS. */
859
860 /* finally add the lockres to its hash bucket */
861 __dlm_insert_lockres(dlm, res);
862 /* get an extra ref on the mle in case this is a BLOCK
863 * if so, the creator of the BLOCK may try to put the last
864 * ref at this time in the assert master handler, so we
865 * need an extra one to keep from a bad ptr deref. */
a2bf0477 866 dlm_get_mle_inuse(mle);
6714d8e8
KH		 867 spin_unlock(&dlm->master_lock);
868 spin_unlock(&dlm->spinlock);
869
c03872f5
KH		 870 while (wait_on_recovery) {
871 /* any cluster changes that occurred after dropping the
872 * dlm spinlock would be detectable be a change on the mle,
873 * so we only need to clear out the recovery map once. */
874 if (dlm_is_recovery_lock(lockid, namelen)) {
875 mlog(ML_NOTICE, "%s: recovery map is not empty, but "
876 "must master $RECOVERY lock now\n", dlm->name);
877 if (!dlm_pre_master_reco_lockres(dlm, res))
878 wait_on_recovery = 0;
879 else {
880 mlog(0, "%s: waiting 500ms for heartbeat state "
881 "change\n", dlm->name);
882 msleep(500);
883 }
884 continue;
885 }
886
887 dlm_kick_recovery_thread(dlm);
888 msleep(100);
889 dlm_wait_for_recovery(dlm);
890
891 spin_lock(&dlm->spinlock);
892 bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
893 if (bit < O2NM_MAX_NODES) {
894 mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
895 "recover before lock mastery can begin\n",
896 dlm->name, namelen, (char *)lockid, bit);
897 wait_on_recovery = 1;
898 } else
899 wait_on_recovery = 0;
900 spin_unlock(&dlm->spinlock);
901 }
902
6714d8e8
KH		 903 /* must wait for lock to be mastered elsewhere */
904 if (blocked)
905 goto wait;
906
907redo_request:
908 ret = -EINVAL;
909 dlm_node_iter_init(mle->vote_map, &iter);
910 while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
911 ret = dlm_do_master_request(mle, nodenum);
912 if (ret < 0)
913 mlog_errno(ret);
914 if (mle->master != O2NM_MAX_NODES) {
915 /* found a master ! */
9c6510a5
KH		 916 if (mle->master <= nodenum)
917 break;
918 /* if our master request has not reached the master
919 * yet, keep going until it does. this is how the
920 * master will know that asserts are needed back to
921 * the lower nodes. */
922 mlog(0, "%s:%.*s: requests only up to %u but master "
923 "is %u, keep going\n", dlm->name, namelen,
924 lockid, nodenum, mle->master);
6714d8e8
KH		 925 }
926 }
927
928wait:
929 /* keep going until the response map includes all nodes */
930 ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
931 if (ret < 0) {
932 mlog(0, "%s:%.*s: node map changed, redo the "
933 "master request now, blocked=%d\n",
934 dlm->name, res->lockname.len,
935 res->lockname.name, blocked);
936 if (++tries > 20) {
937 mlog(ML_ERROR, "%s:%.*s: spinning on "
938 "dlm_wait_for_lock_mastery, blocked=%d\n",
939 dlm->name, res->lockname.len,
940 res->lockname.name, blocked);
941 dlm_print_one_lock_resource(res);
942 /* dlm_print_one_mle(mle); */
943 tries = 0;
944 }
945 goto redo_request;
946 }
947
948 mlog(0, "lockres mastered by %u\n", res->owner);
949 /* make sure we never continue without this */
950 BUG_ON(res->owner == O2NM_MAX_NODES);
951
952 /* master is known, detach if not already detached */
953 dlm_mle_detach_hb_events(dlm, mle);
954 dlm_put_mle(mle);
955 /* put the extra ref */
a2bf0477 956 dlm_put_mle_inuse(mle);
6714d8e8
KH		 957
958wake_waiters:
959 spin_lock(&res->spinlock);
960 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
961 spin_unlock(&res->spinlock);
962 wake_up(&res->wq);
963
964leave:
965 /* need to free the unused mle */
966 if (alloc_mle)
967 kmem_cache_free(dlm_mle_cache, alloc_mle);
968
969 return res;
970}
971
972
973#define DLM_MASTERY_TIMEOUT_MS 5000
974
975static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
976 struct dlm_lock_resource *res,
977 struct dlm_master_list_entry *mle,
978 int *blocked)
979{
980 u8 m;
981 int ret, bit;
982 int map_changed, voting_done;
983 int assert, sleep;
984
985recheck:
986 ret = 0;
987 assert = 0;
988
989 /* check if another node has already become the owner */
990 spin_lock(&res->spinlock);
991 if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
9c6510a5
KH		 992 mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name,
993 res->lockname.len, res->lockname.name, res->owner);
6714d8e8 994 spin_unlock(&res->spinlock);
9c6510a5
KH		 995 /* this will cause the master to re-assert across
996 * the whole cluster, freeing up mles */
588e0090
KH		 997 if (res->owner != dlm->node_num) {
998 ret = dlm_do_master_request(mle, res->owner);
999 if (ret < 0) {
1000 /* give recovery a chance to run */
1001 mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
1002 msleep(500);
1003 goto recheck;
1004 }
9c6510a5
KH		 1005 }
1006 ret = 0;
6714d8e8
KH		 1007 goto leave;
1008 }
1009 spin_unlock(&res->spinlock);
1010
1011 spin_lock(&mle->spinlock);
1012 m = mle->master;
1013 map_changed = (memcmp(mle->vote_map, mle->node_map,
1014 sizeof(mle->vote_map)) != 0);
1015 voting_done = (memcmp(mle->vote_map, mle->response_map,
1016 sizeof(mle->vote_map)) == 0);
1017
1018 /* restart if we hit any errors */
1019 if (map_changed) {
1020 int b;
1021 mlog(0, "%s: %.*s: node map changed, restarting\n",
1022 dlm->name, res->lockname.len, res->lockname.name);
1023 ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked);
1024 b = (mle->type == DLM_MLE_BLOCK);
1025 if ((*blocked && !b) || (!*blocked && b)) {
1026 mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
1027 dlm->name, res->lockname.len, res->lockname.name,
1028 *blocked, b);
1029 *blocked = b;
1030 }
1031 spin_unlock(&mle->spinlock);
1032 if (ret < 0) {
1033 mlog_errno(ret);
1034 goto leave;
1035 }
1036 mlog(0, "%s:%.*s: restart lock mastery succeeded, "
1037 "rechecking now\n", dlm->name, res->lockname.len,
1038 res->lockname.name);
1039 goto recheck;
aa852354
KH		 1040 } else {
1041 if (!voting_done) {
1042 mlog(0, "map not changed and voting not done "
1043 "for %s:%.*s\n", dlm->name, res->lockname.len,
1044 res->lockname.name);
1045 }
6714d8e8
KH		 1046 }
1047
1048 if (m != O2NM_MAX_NODES) {
1049 /* another node has done an assert!
1050 * all done! */
1051 sleep = 0;
1052 } else {
1053 sleep = 1;
1054 /* have all nodes responded? */
1055 if (voting_done && !*blocked) {
1056 bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
1057 if (dlm->node_num <= bit) {
1058 /* my node number is lowest.
1059 * now tell other nodes that I am
1060 * mastering this. */
1061 mle->master = dlm->node_num;
1062 assert = 1;
1063 sleep = 0;
1064 }
1065 /* if voting is done, but we have not received
1066 * an assert master yet, we must sleep */
1067 }
1068 }
1069
1070 spin_unlock(&mle->spinlock);
1071
1072 /* sleep if we haven't finished voting yet */
1073 if (sleep) {
1074 unsigned long timeo = msecs_to_jiffies(DLM_MASTERY_TIMEOUT_MS);
1075
1076 /*
1077 if (atomic_read(&mle->mle_refs.refcount) < 2)
1078 mlog(ML_ERROR, "mle (%p) refs=%d, name=%.*s\n", mle,
1079 atomic_read(&mle->mle_refs.refcount),
1080 res->lockname.len, res->lockname.name);
1081 */
1082 atomic_set(&mle->woken, 0);
1083 (void)wait_event_timeout(mle->wq,
1084 (atomic_read(&mle->woken) == 1),
1085 timeo);
1086 if (res->owner == O2NM_MAX_NODES) {
1087 mlog(0, "waiting again\n");
1088 goto recheck;
1089 }
1090 mlog(0, "done waiting, master is %u\n", res->owner);
1091 ret = 0;
1092 goto leave;
1093 }
1094
1095 ret = 0; /* done */
1096 if (assert) {
1097 m = dlm->node_num;
1098 mlog(0, "about to master %.*s here, this=%u\n",
1099 res->lockname.len, res->lockname.name, m);
1100 ret = dlm_do_assert_master(dlm, res->lockname.name,
1101 res->lockname.len, mle->vote_map, 0);
1102 if (ret) {
1103 /* This is a failure in the network path,
1104 * not in the response to the assert_master
1105 * (any nonzero response is a BUG on this node).
1106 * Most likely a socket just got disconnected
1107 * due to node death. */
1108 mlog_errno(ret);
1109 }
1110 /* no longer need to restart lock mastery.
1111 * all living nodes have been contacted. */
1112 ret = 0;
1113 }
1114
1115 /* set the lockres owner */
1116 spin_lock(&res->spinlock);
1117 dlm_change_lockres_owner(dlm, res, m);
1118 spin_unlock(&res->spinlock);
1119
1120leave:
1121 return ret;
1122}
1123
1124struct dlm_bitmap_diff_iter
1125{
1126 int curnode;
1127 unsigned long *orig_bm;
1128 unsigned long *cur_bm;
1129 unsigned long diff_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
1130};
1131
1132enum dlm_node_state_change
1133{
1134 NODE_DOWN = -1,
1135 NODE_NO_CHANGE = 0,
1136 NODE_UP
1137};
1138
1139static void dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter *iter,
1140 unsigned long *orig_bm,
1141 unsigned long *cur_bm)
1142{
1143 unsigned long p1, p2;
1144 int i;
1145
1146 iter->curnode = -1;
1147 iter->orig_bm = orig_bm;
1148 iter->cur_bm = cur_bm;
1149
1150 for (i = 0; i < BITS_TO_LONGS(O2NM_MAX_NODES); i++) {
1151 p1 = *(iter->orig_bm + i);
1152 p2 = *(iter->cur_bm + i);
1153 iter->diff_bm[i] = (p1 & ~p2) | (p2 & ~p1);
1154 }
1155}
1156
1157static int dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter *iter,
1158 enum dlm_node_state_change *state)
1159{
1160 int bit;
1161
1162 if (iter->curnode >= O2NM_MAX_NODES)
1163 return -ENOENT;
1164
1165 bit = find_next_bit(iter->diff_bm, O2NM_MAX_NODES,
1166 iter->curnode+1);
1167 if (bit >= O2NM_MAX_NODES) {
1168 iter->curnode = O2NM_MAX_NODES;
1169 return -ENOENT;
1170 }
1171
1172 /* if it was there in the original then this node died */
1173 if (test_bit(bit, iter->orig_bm))
1174 *state = NODE_DOWN;
1175 else
1176 *state = NODE_UP;
1177
1178 iter->curnode = bit;
1179 return bit;
1180}
1181
1182
1183static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
1184 struct dlm_lock_resource *res,
1185 struct dlm_master_list_entry *mle,
1186 int blocked)
1187{
1188 struct dlm_bitmap_diff_iter bdi;
1189 enum dlm_node_state_change sc;
1190 int node;
1191 int ret = 0;
1192
1193 mlog(0, "something happened such that the "
1194 "master process may need to be restarted!\n");
1195
1196 assert_spin_locked(&mle->spinlock);
1197
1198 dlm_bitmap_diff_iter_init(&bdi, mle->vote_map, mle->node_map);
1199 node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1200 while (node >= 0) {
1201 if (sc == NODE_UP) {
e2faea4c
KH		 1202 /* a node came up. clear any old vote from
1203 * the response map and set it in the vote map
1204 * then restart the mastery. */
1205 mlog(ML_NOTICE, "node %d up while restarting\n", node);
6714d8e8
KH		 1206
1207 /* redo the master request, but only for the new node */
1208 mlog(0, "sending request to new node\n");
1209 clear_bit(node, mle->response_map);
1210 set_bit(node, mle->vote_map);
1211 } else {
1212 mlog(ML_ERROR, "node down! %d\n", node);
1213
1214 /* if the node wasn't involved in mastery skip it,
1215 * but clear it out from the maps so that it will
1216 * not affect mastery of this lockres */
1217 clear_bit(node, mle->response_map);
1218 clear_bit(node, mle->vote_map);
1219 if (!test_bit(node, mle->maybe_map))
1220 goto next;
1221
1222 /* if we're already blocked on lock mastery, and the
1223 * dead node wasn't the expected master, or there is
1224 * another node in the maybe_map, keep waiting */
1225 if (blocked) {
1226 int lowest = find_next_bit(mle->maybe_map,
1227 O2NM_MAX_NODES, 0);
1228
1229 /* act like it was never there */
1230 clear_bit(node, mle->maybe_map);
1231
1232 if (node != lowest)
1233 goto next;
1234
1235 mlog(ML_ERROR, "expected master %u died while "
1236 "this node was blocked waiting on it!\n",
1237 node);
1238 lowest = find_next_bit(mle->maybe_map,
1239 O2NM_MAX_NODES,
1240 lowest+1);
1241 if (lowest < O2NM_MAX_NODES) {
1242 mlog(0, "still blocked. waiting "
1243 "on %u now\n", lowest);
1244 goto next;
1245 }
1246
1247 /* mle is an MLE_BLOCK, but there is now
1248 * nothing left to block on. we need to return
1249 * all the way back out and try again with
1250 * an MLE_MASTER. dlm_do_local_recovery_cleanup
1251 * has already run, so the mle refcount is ok */
1252 mlog(0, "no longer blocking. we can "
1253 "try to master this here\n");
1254 mle->type = DLM_MLE_MASTER;
1255 memset(mle->maybe_map, 0,
1256 sizeof(mle->maybe_map));
1257 memset(mle->response_map, 0,
1258 sizeof(mle->maybe_map));
1259 memcpy(mle->vote_map, mle->node_map,
1260 sizeof(mle->node_map));
1261 mle->u.res = res;
1262 set_bit(dlm->node_num, mle->maybe_map);
1263
1264 ret = -EAGAIN;
1265 goto next;
1266 }
1267
1268 clear_bit(node, mle->maybe_map);
1269 if (node > dlm->node_num)
1270 goto next;
1271
1272 mlog(0, "dead node in map!\n");
1273 /* yuck. go back and re-contact all nodes
1274 * in the vote_map, removing this node. */
1275 memset(mle->response_map, 0,
1276 sizeof(mle->response_map));
1277 }
1278 ret = -EAGAIN;
1279next:
1280 node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1281 }
1282 return ret;
1283}
1284
1285
1286/*
1287 * DLM_MASTER_REQUEST_MSG
1288 *
1289 * returns: 0 on success,
1290 * -errno on a network error
1291 *
1292 * on error, the caller should assume the target node is "dead"
1293 *
1294 */
1295
1296static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to)
1297{
1298 struct dlm_ctxt *dlm = mle->dlm;
1299 struct dlm_master_request request;
1300 int ret, response=0, resend;
1301
1302 memset(&request, 0, sizeof(request));
1303 request.node_idx = dlm->node_num;
1304
1305 BUG_ON(mle->type == DLM_MLE_MIGRATION);
1306
1307 if (mle->type != DLM_MLE_MASTER) {
1308 request.namelen = mle->u.name.len;
1309 memcpy(request.name, mle->u.name.name, request.namelen);
1310 } else {
1311 request.namelen = mle->u.res->lockname.len;
1312 memcpy(request.name, mle->u.res->lockname.name,
1313 request.namelen);
1314 }
1315
1316again:
1317 ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request,
1318 sizeof(request), to, &response);
1319 if (ret < 0) {
1320 if (ret == -ESRCH) {
1321 /* should never happen */
1322 mlog(ML_ERROR, "TCP stack not ready!\n");
1323 BUG();
1324 } else if (ret == -EINVAL) {
1325 mlog(ML_ERROR, "bad args passed to o2net!\n");
1326 BUG();
1327 } else if (ret == -ENOMEM) {
1328 mlog(ML_ERROR, "out of memory while trying to send "
1329 "network message! retrying\n");
1330 /* this is totally crude */
1331 msleep(50);
1332 goto again;
1333 } else if (!dlm_is_host_down(ret)) {
1334 /* not a network error. bad. */
1335 mlog_errno(ret);
1336 mlog(ML_ERROR, "unhandled error!");
1337 BUG();
1338 }
1339 /* all other errors should be network errors,
1340 * and likely indicate node death */
1341 mlog(ML_ERROR, "link to %d went down!\n", to);
1342 goto out;
1343 }
1344
1345 ret = 0;
1346 resend = 0;
1347 spin_lock(&mle->spinlock);
1348 switch (response) {
1349 case DLM_MASTER_RESP_YES:
1350 set_bit(to, mle->response_map);
1351 mlog(0, "node %u is the master, response=YES\n", to);
1352 mle->master = to;
1353 break;
1354 case DLM_MASTER_RESP_NO:
1355 mlog(0, "node %u not master, response=NO\n", to);
1356 set_bit(to, mle->response_map);
1357 break;
1358 case DLM_MASTER_RESP_MAYBE:
1359 mlog(0, "node %u not master, response=MAYBE\n", to);
1360 set_bit(to, mle->response_map);
1361 set_bit(to, mle->maybe_map);
1362 break;
1363 case DLM_MASTER_RESP_ERROR:
1364 mlog(0, "node %u hit an error, resending\n", to);
1365 resend = 1;
1366 response = 0;
1367 break;
1368 default:
1369 mlog(ML_ERROR, "bad response! %u\n", response);
1370 BUG();
1371 }
1372 spin_unlock(&mle->spinlock);
1373 if (resend) {
1374 /* this is also totally crude */
1375 msleep(50);
1376 goto again;
1377 }
1378
1379out:
1380 return ret;
1381}
1382
1383/*
1384 * locks that can be taken here:
1385 * dlm->spinlock
1386 * res->spinlock
1387 * mle->spinlock
1388 * dlm->master_list
1389 *
1390 * if possible, TRIM THIS DOWN!!!
1391 */
1392int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data)
1393{
1394 u8 response = DLM_MASTER_RESP_MAYBE;
1395 struct dlm_ctxt *dlm = data;
9c6510a5 1396 struct dlm_lock_resource *res = NULL;
6714d8e8
KH		 1397 struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
1398 struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
1399 char *name;
a3d33291 1400 unsigned int namelen, hash;
6714d8e8
KH		 1401 int found, ret;
1402 int set_maybe;
9c6510a5 1403 int dispatch_assert = 0;
6714d8e8
KH		 1404
1405 if (!dlm_grab(dlm))
1406 return DLM_MASTER_RESP_NO;
1407
1408 if (!dlm_domain_fully_joined(dlm)) {
1409 response = DLM_MASTER_RESP_NO;
1410 goto send_response;
1411 }
1412
1413 name = request->name;
1414 namelen = request->namelen;
a3d33291 1415 hash = dlm_lockid_hash(name, namelen);
6714d8e8
KH		 1416
1417 if (namelen > DLM_LOCKID_NAME_MAX) {
1418 response = DLM_IVBUFLEN;
1419 goto send_response;
1420 }
1421
1422way_up_top:
1423 spin_lock(&dlm->spinlock);
a3d33291 1424 res = __dlm_lookup_lockres(dlm, name, namelen, hash);
6714d8e8
KH		 1425 if (res) {
1426 spin_unlock(&dlm->spinlock);
1427
1428 /* take care of the easy cases up front */
1429 spin_lock(&res->spinlock);
1430 if (res->state & DLM_LOCK_RES_RECOVERING) {
1431 spin_unlock(&res->spinlock);
1432 mlog(0, "returning DLM_MASTER_RESP_ERROR since res is "
1433 "being recovered\n");
1434 response = DLM_MASTER_RESP_ERROR;
1435 if (mle)
1436 kmem_cache_free(dlm_mle_cache, mle);
1437 goto send_response;
1438 }
1439
1440 if (res->owner == dlm->node_num) {
6714d8e8
KH		 1441 spin_unlock(&res->spinlock);
1442 // mlog(0, "this node is the master\n");
1443 response = DLM_MASTER_RESP_YES;
1444 if (mle)
1445 kmem_cache_free(dlm_mle_cache, mle);
1446
1447 /* this node is the owner.
1448 * there is some extra work that needs to
1449 * happen now. the requesting node has
1450 * caused all nodes up to this one to
1451 * create mles. this node now needs to
1452 * go back and clean those up. */
9c6510a5 1453 dispatch_assert = 1;
6714d8e8
KH		 1454 goto send_response;
1455 } else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1456 spin_unlock(&res->spinlock);
1457 // mlog(0, "node %u is the master\n", res->owner);
1458 response = DLM_MASTER_RESP_NO;
1459 if (mle)
1460 kmem_cache_free(dlm_mle_cache, mle);
1461 goto send_response;
1462 }
1463
1464 /* ok, there is no owner. either this node is
1465 * being blocked, or it is actively trying to
1466 * master this lock. */
1467 if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1468 mlog(ML_ERROR, "lock with no owner should be "
1469 "in-progress!\n");
1470 BUG();
1471 }
1472
1473 // mlog(0, "lockres is in progress...\n");
1474 spin_lock(&dlm->master_lock);
1475 found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1476 if (!found) {
1477 mlog(ML_ERROR, "no mle found for this lock!\n");
1478 BUG();
1479 }
1480 set_maybe = 1;
1481 spin_lock(&tmpmle->spinlock);
1482 if (tmpmle->type == DLM_MLE_BLOCK) {
1483 // mlog(0, "this node is waiting for "
1484 // "lockres to be mastered\n");
1485 response = DLM_MASTER_RESP_NO;
1486 } else if (tmpmle->type == DLM_MLE_MIGRATION) {
1487 mlog(0, "node %u is master, but trying to migrate to "
1488 "node %u.\n", tmpmle->master, tmpmle->new_master);
1489 if (tmpmle->master == dlm->node_num) {
1490 response = DLM_MASTER_RESP_YES;
1491 mlog(ML_ERROR, "no owner on lockres, but this "
1492 "node is trying to migrate it to %u?!\n",
1493 tmpmle->new_master);
1494 BUG();
1495 } else {
1496 /* the real master can respond on its own */
1497 response = DLM_MASTER_RESP_NO;
1498 }
1499 } else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) {
1500 set_maybe = 0;
9c6510a5 1501 if (tmpmle->master == dlm->node_num) {
6714d8e8 1502 response = DLM_MASTER_RESP_YES;
9c6510a5
KH		 1503 /* this node will be the owner.
1504 * go back and clean the mles on any
1505 * other nodes */
1506 dispatch_assert = 1;
1507 } else
6714d8e8
KH		 1508 response = DLM_MASTER_RESP_NO;
1509 } else {
1510 // mlog(0, "this node is attempting to "
1511 // "master lockres\n");
1512 response = DLM_MASTER_RESP_MAYBE;
1513 }
1514 if (set_maybe)
1515 set_bit(request->node_idx, tmpmle->maybe_map);
1516 spin_unlock(&tmpmle->spinlock);
1517
1518 spin_unlock(&dlm->master_lock);
1519 spin_unlock(&res->spinlock);
1520
1521 /* keep the mle attached to heartbeat events */
1522 dlm_put_mle(tmpmle);
1523 if (mle)
1524 kmem_cache_free(dlm_mle_cache, mle);
1525 goto send_response;
1526 }
1527
1528 /*
1529 * lockres doesn't exist on this node
1530 * if there is an MLE_BLOCK, return NO
1531 * if there is an MLE_MASTER, return MAYBE
1532 * otherwise, add an MLE_BLOCK, return NO
1533 */
1534 spin_lock(&dlm->master_lock);
1535 found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1536 if (!found) {
1537 /* this lockid has never been seen on this node yet */
1538 // mlog(0, "no mle found\n");
1539 if (!mle) {
1540 spin_unlock(&dlm->master_lock);
1541 spin_unlock(&dlm->spinlock);
1542
1543 mle = (struct dlm_master_list_entry *)
1544 kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
1545 if (!mle) {
6714d8e8 1546 response = DLM_MASTER_RESP_ERROR;
9c6510a5 1547 mlog_errno(-ENOMEM);
6714d8e8
KH		 1548 goto send_response;
1549 }
6714d8e8
KH		 1550 goto way_up_top;
1551 }
1552
1553 // mlog(0, "this is second time thru, already allocated, "
1554 // "add the block.\n");
41b8c8a1 1555 dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
6714d8e8
KH		 1556 set_bit(request->node_idx, mle->maybe_map);
1557 list_add(&mle->list, &dlm->master_list);
1558 response = DLM_MASTER_RESP_NO;
1559 } else {
1560 // mlog(0, "mle was found\n");
1561 set_maybe = 1;
1562 spin_lock(&tmpmle->spinlock);
9c6510a5
KH		 1563 if (tmpmle->master == dlm->node_num) {
1564 mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n");
1565 BUG();
1566 }
6714d8e8
KH		 1567 if (tmpmle->type == DLM_MLE_BLOCK)
1568 response = DLM_MASTER_RESP_NO;
1569 else if (tmpmle->type == DLM_MLE_MIGRATION) {
1570 mlog(0, "migration mle was found (%u->%u)\n",
1571 tmpmle->master, tmpmle->new_master);
6714d8e8
KH		 1572 /* real master can respond on its own */
1573 response = DLM_MASTER_RESP_NO;
9c6510a5
KH		 1574 } else
1575 response = DLM_MASTER_RESP_MAYBE;
6714d8e8
KH		 1576 if (set_maybe)
1577 set_bit(request->node_idx, tmpmle->maybe_map);
1578 spin_unlock(&tmpmle->spinlock);
1579 }
1580 spin_unlock(&dlm->master_lock);
1581 spin_unlock(&dlm->spinlock);
1582
1583 if (found) {
1584 /* keep the mle attached to heartbeat events */
1585 dlm_put_mle(tmpmle);
1586 }
1587send_response:
9c6510a5
KH		 1588
1589 if (dispatch_assert) {
1590 if (response != DLM_MASTER_RESP_YES)
1591 mlog(ML_ERROR, "invalid response %d\n", response);
1592 if (!res) {
1593 mlog(ML_ERROR, "bad lockres while trying to assert!\n");
1594 BUG();
1595 }
1596 mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
1597 dlm->node_num, res->lockname.len, res->lockname.name);
1598 ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
1599 DLM_ASSERT_MASTER_MLE_CLEANUP);
1600 if (ret < 0) {
1601 mlog(ML_ERROR, "failed to dispatch assert master work\n");
1602 response = DLM_MASTER_RESP_ERROR;
1603 }
1604 }
1605
6714d8e8
KH		 1606 dlm_put(dlm);
1607 return response;
1608}
1609
1610/*
1611 * DLM_ASSERT_MASTER_MSG
1612 */
1613
1614
1615/*
1616 * NOTE: this can be used for debugging
1617 * can periodically run all locks owned by this node
1618 * and re-assert across the cluster...
1619 */
1620static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
1621 unsigned int namelen, void *nodemap,
1622 u32 flags)
1623{
1624 struct dlm_assert_master assert;
1625 int to, tmpret;
1626 struct dlm_node_iter iter;
1627 int ret = 0;
9c6510a5 1628 int reassert;
6714d8e8
KH		 1629
1630 BUG_ON(namelen > O2NM_MAX_NAME_LEN);
9c6510a5
KH		 1631again:
1632 reassert = 0;
6714d8e8
KH		 1633
1634 /* note that if this nodemap is empty, it returns 0 */
1635 dlm_node_iter_init(nodemap, &iter);
1636 while ((to = dlm_node_iter_next(&iter)) >= 0) {
1637 int r = 0;
a9ee4c8a
KH		 1638 struct dlm_master_list_entry *mle = NULL;
1639
6714d8e8
KH		 1640 mlog(0, "sending assert master to %d (%.*s)\n", to,
1641 namelen, lockname);
1642 memset(&assert, 0, sizeof(assert));
1643 assert.node_idx = dlm->node_num;
1644 assert.namelen = namelen;
1645 memcpy(assert.name, lockname, namelen);
1646 assert.flags = cpu_to_be32(flags);
1647
1648 tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
1649 &assert, sizeof(assert), to, &r);
1650 if (tmpret < 0) {
1651 mlog(ML_ERROR, "assert_master returned %d!\n", tmpret);
1652 if (!dlm_is_host_down(tmpret)) {
1653 mlog(ML_ERROR, "unhandled error!\n");
1654 BUG();
1655 }
1656 /* a node died. finish out the rest of the nodes. */
1657 mlog(ML_ERROR, "link to %d went down!\n", to);
1658 /* any nonzero status return will do */
1659 ret = tmpret;
1660 } else if (r < 0) {
1661 /* ok, something horribly messed. kill thyself. */
1662 mlog(ML_ERROR,"during assert master of %.*s to %u, "
1663 "got %d.\n", namelen, lockname, to, r);
a9ee4c8a
KH		 1664 spin_lock(&dlm->spinlock);
1665 spin_lock(&dlm->master_lock);
1666 if (dlm_find_mle(dlm, &mle, (char *)lockname,
1667 namelen)) {
1668 dlm_print_one_mle(mle);
1669 __dlm_put_mle(mle);
1670 }
1671 spin_unlock(&dlm->master_lock);
1672 spin_unlock(&dlm->spinlock);
6714d8e8 1673 BUG();
9c6510a5
KH		 1674 } else if (r == EAGAIN) {
1675 mlog(0, "%.*s: node %u create mles on other "
1676 "nodes and requests a re-assert\n",
1677 namelen, lockname, to);
1678 reassert = 1;
6714d8e8
KH		 1679 }
1680 }
1681
9c6510a5
KH		 1682 if (reassert)
1683 goto again;
1684
6714d8e8
KH		 1685 return ret;
1686}
1687
1688/*
1689 * locks that can be taken here:
1690 * dlm->spinlock
1691 * res->spinlock
1692 * mle->spinlock
1693 * dlm->master_list
1694 *
1695 * if possible, TRIM THIS DOWN!!!
1696 */
1697int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data)
1698{
1699 struct dlm_ctxt *dlm = data;
1700 struct dlm_master_list_entry *mle = NULL;
1701 struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
1702 struct dlm_lock_resource *res = NULL;
1703 char *name;
a3d33291 1704 unsigned int namelen, hash;
6714d8e8 1705 u32 flags;
9c6510a5
KH		 1706 int master_request = 0;
1707 int ret = 0;
6714d8e8
KH		 1708
1709 if (!dlm_grab(dlm))
1710 return 0;
1711
1712 name = assert->name;
1713 namelen = assert->namelen;
a3d33291 1714 hash = dlm_lockid_hash(name, namelen);
6714d8e8
KH		 1715 flags = be32_to_cpu(assert->flags);
1716
1717 if (namelen > DLM_LOCKID_NAME_MAX) {
1718 mlog(ML_ERROR, "Invalid name length!");
1719 goto done;
1720 }
1721
1722 spin_lock(&dlm->spinlock);
1723
1724 if (flags)
1725 mlog(0, "assert_master with flags: %u\n", flags);
1726
1727 /* find the MLE */
1728 spin_lock(&dlm->master_lock);
1729 if (!dlm_find_mle(dlm, &mle, name, namelen)) {
1730 /* not an error, could be master just re-asserting */
1731 mlog(0, "just got an assert_master from %u, but no "
1732 "MLE for it! (%.*s)\n", assert->node_idx,
1733 namelen, name);
1734 } else {
1735 int bit = find_next_bit (mle->maybe_map, O2NM_MAX_NODES, 0);
1736 if (bit >= O2NM_MAX_NODES) {
1737 /* not necessarily an error, though less likely.
1738 * could be master just re-asserting. */
aa852354 1739 mlog(0, "no bits set in the maybe_map, but %u "
6714d8e8
KH		 1740 "is asserting! (%.*s)\n", assert->node_idx,
1741 namelen, name);
1742 } else if (bit != assert->node_idx) {
1743 if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1744 mlog(0, "master %u was found, %u should "
1745 "back off\n", assert->node_idx, bit);
1746 } else {
1747 /* with the fix for bug 569, a higher node
1748 * number winning the mastery will respond
1749 * YES to mastery requests, but this node
1750 * had no way of knowing. let it pass. */
aa852354 1751 mlog(0, "%u is the lowest node, "
6714d8e8
KH		 1752 "%u is asserting. (%.*s) %u must "
1753 "have begun after %u won.\n", bit,
1754 assert->node_idx, namelen, name, bit,
1755 assert->node_idx);
1756 }
1757 }
2d1a868c
KH		 1758 if (mle->type == DLM_MLE_MIGRATION) {
1759 if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1760 mlog(0, "%s:%.*s: got cleanup assert"
1761 " from %u for migration\n",
1762 dlm->name, namelen, name,
1763 assert->node_idx);
1764 } else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) {
1765 mlog(0, "%s:%.*s: got unrelated assert"
1766 " from %u for migration, ignoring\n",
1767 dlm->name, namelen, name,
1768 assert->node_idx);
1769 __dlm_put_mle(mle);
1770 spin_unlock(&dlm->master_lock);
1771 spin_unlock(&dlm->spinlock);
1772 goto done;
1773 }
1774 }
6714d8e8
KH		 1775 }
1776 spin_unlock(&dlm->master_lock);
1777
1778 /* ok everything checks out with the MLE
1779 * now check to see if there is a lockres */
a3d33291 1780 res = __dlm_lookup_lockres(dlm, name, namelen, hash);
6714d8e8
KH		 1781 if (res) {
1782 spin_lock(&res->spinlock);
1783 if (res->state & DLM_LOCK_RES_RECOVERING) {
1784 mlog(ML_ERROR, "%u asserting but %.*s is "
1785 "RECOVERING!\n", assert->node_idx, namelen, name);
1786 goto kill;
1787 }
1788 if (!mle) {
dc2ed195
KH		 1789 if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN &&
1790 res->owner != assert->node_idx) {
6714d8e8
KH		 1791 mlog(ML_ERROR, "assert_master from "
1792 "%u, but current owner is "
1793 "%u! (%.*s)\n",
1794 assert->node_idx, res->owner,
1795 namelen, name);
1796 goto kill;
1797 }
1798 } else if (mle->type != DLM_MLE_MIGRATION) {
1799 if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1800 /* owner is just re-asserting */
1801 if (res->owner == assert->node_idx) {
1802 mlog(0, "owner %u re-asserting on "
1803 "lock %.*s\n", assert->node_idx,
1804 namelen, name);
1805 goto ok;
1806 }
1807 mlog(ML_ERROR, "got assert_master from "
1808 "node %u, but %u is the owner! "
1809 "(%.*s)\n", assert->node_idx,
1810 res->owner, namelen, name);
1811 goto kill;
1812 }
1813 if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1814 mlog(ML_ERROR, "got assert from %u, but lock "
1815 "with no owner should be "
1816 "in-progress! (%.*s)\n",
1817 assert->node_idx,
1818 namelen, name);
1819 goto kill;
1820 }
1821 } else /* mle->type == DLM_MLE_MIGRATION */ {
1822 /* should only be getting an assert from new master */
1823 if (assert->node_idx != mle->new_master) {
1824 mlog(ML_ERROR, "got assert from %u, but "
1825 "new master is %u, and old master "
1826 "was %u (%.*s)\n",
1827 assert->node_idx, mle->new_master,
1828 mle->master, namelen, name);
1829 goto kill;
1830 }
1831
1832 }
1833ok:
1834 spin_unlock(&res->spinlock);
1835 }
1836 spin_unlock(&dlm->spinlock);
1837
1838 // mlog(0, "woo! got an assert_master from node %u!\n",
1839 // assert->node_idx);
1840 if (mle) {
9c6510a5
KH		 1841 int extra_ref = 0;
1842 int nn = -1;
a2bf0477 1843 int rr, err = 0;
6714d8e8
KH		 1844
1845 spin_lock(&mle->spinlock);
9c6510a5
KH		 1846 if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
1847 extra_ref = 1;
1848 else {
1849 /* MASTER mle: if any bits set in the response map
1850 * then the calling node needs to re-assert to clear
1851 * up nodes that this node contacted */
1852 while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
1853 nn+1)) < O2NM_MAX_NODES) {
1854 if (nn != dlm->node_num && nn != assert->node_idx)
1855 master_request = 1;
1856 }
1857 }
6714d8e8
KH		 1858 mle->master = assert->node_idx;
1859 atomic_set(&mle->woken, 1);
1860 wake_up(&mle->wq);
1861 spin_unlock(&mle->spinlock);
1862
a2bf0477 1863 if (res) {
6714d8e8 1864 spin_lock(&res->spinlock);
a2bf0477
KH		 1865 if (mle->type == DLM_MLE_MIGRATION) {
1866 mlog(0, "finishing off migration of lockres %.*s, "
1867 "from %u to %u\n",
1868 res->lockname.len, res->lockname.name,
1869 dlm->node_num, mle->new_master);
1870 res->state &= ~DLM_LOCK_RES_MIGRATING;
1871 dlm_change_lockres_owner(dlm, res, mle->new_master);
1872 BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
1873 } else {
1874 dlm_change_lockres_owner(dlm, res, mle->master);
1875 }
6714d8e8
KH		 1876 spin_unlock(&res->spinlock);
1877 }
a2bf0477
KH		 1878
1879 /* master is known, detach if not already detached.
1880 * ensures that only one assert_master call will happen
1881 * on this mle. */
1882 spin_lock(&dlm->spinlock);
1883 spin_lock(&dlm->master_lock);
1884
1885 rr = atomic_read(&mle->mle_refs.refcount);
1886 if (mle->inuse > 0) {
1887 if (extra_ref && rr < 3)
1888 err = 1;
1889 else if (!extra_ref && rr < 2)
1890 err = 1;
1891 } else {
1892 if (extra_ref && rr < 2)
1893 err = 1;
1894 else if (!extra_ref && rr < 1)
1895 err = 1;
1896 }
1897 if (err) {
1898 mlog(ML_ERROR, "%s:%.*s: got assert master from %u "
1899 "that will mess up this node, refs=%d, extra=%d, "
1900 "inuse=%d\n", dlm->name, namelen, name,
1901 assert->node_idx, rr, extra_ref, mle->inuse);
1902 dlm_print_one_mle(mle);
1903 }
1904 list_del_init(&mle->list);
1905 __dlm_mle_detach_hb_events(dlm, mle);
1906 __dlm_put_mle(mle);
6714d8e8
KH		 1907 if (extra_ref) {
1908 /* the assert master message now balances the extra
1909 * ref given by the master / migration request message.
1910 * if this is the last put, it will be removed
1911 * from the list. */
a2bf0477
KH		 1912 __dlm_put_mle(mle);
1913 }
1914 spin_unlock(&dlm->master_lock);
1915 spin_unlock(&dlm->spinlock);
1916 } else if (res) {
1917 if (res->owner != assert->node_idx) {
1918 mlog(0, "assert_master from %u, but current "
1919 "owner is %u (%.*s), no mle\n", assert->node_idx,
1920 res->owner, namelen, name);
6714d8e8
KH		 1921 }
1922 }
1923
1924done:
9c6510a5 1925 ret = 0;
6714d8e8
KH		 1926 if (res)
1927 dlm_lockres_put(res);
1928 dlm_put(dlm);
9c6510a5
KH		 1929 if (master_request) {
1930 mlog(0, "need to tell master to reassert\n");
1931 ret = EAGAIN; // positive. negative would shoot down the node.
1932 }
1933 return ret;
6714d8e8
KH		 1934
1935kill:
1936 /* kill the caller! */
a9ee4c8a
KH		 1937 mlog(ML_ERROR, "Bad message received from another node. Dumping state "
1938 "and killing the other node now! This node is OK and can continue.\n");
1939 __dlm_print_one_lock_resource(res);
6714d8e8
KH		 1940 spin_unlock(&res->spinlock);
1941 spin_unlock(&dlm->spinlock);
1942 dlm_lockres_put(res);
6714d8e8
KH		 1943 dlm_put(dlm);
1944 return -EINVAL;
1945}
1946
1947int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
1948 struct dlm_lock_resource *res,
1949 int ignore_higher, u8 request_from, u32 flags)
1950{
1951 struct dlm_work_item *item;
1952 item = kcalloc(1, sizeof(*item), GFP_KERNEL);
1953 if (!item)
1954 return -ENOMEM;
1955
1956
1957 /* queue up work for dlm_assert_master_worker */
1958 dlm_grab(dlm); /* get an extra ref for the work item */
1959 dlm_init_work_item(dlm, item, dlm_assert_master_worker, NULL);
1960 item->u.am.lockres = res; /* already have a ref */
1961 /* can optionally ignore node numbers higher than this node */
1962 item->u.am.ignore_higher = ignore_higher;
1963 item->u.am.request_from = request_from;
1964 item->u.am.flags = flags;
1965
9c6510a5
KH		 1966 if (ignore_higher)
1967 mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
1968 res->lockname.name);
1969
6714d8e8
KH		 1970 spin_lock(&dlm->work_lock);
1971 list_add_tail(&item->list, &dlm->work_list);
1972 spin_unlock(&dlm->work_lock);
1973
1974 schedule_work(&dlm->dispatched_work);
1975 return 0;
1976}
1977
1978static void dlm_assert_master_worker(struct dlm_work_item *item, void *data)
1979{
1980 struct dlm_ctxt *dlm = data;
1981 int ret = 0;
1982 struct dlm_lock_resource *res;
1983 unsigned long nodemap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1984 int ignore_higher;
1985 int bit;
1986 u8 request_from;
1987 u32 flags;
1988
1989 dlm = item->dlm;
1990 res = item->u.am.lockres;
1991 ignore_higher = item->u.am.ignore_higher;
1992 request_from = item->u.am.request_from;
1993 flags = item->u.am.flags;
1994
1995 spin_lock(&dlm->spinlock);
1996 memcpy(nodemap, dlm->domain_map, sizeof(nodemap));
1997 spin_unlock(&dlm->spinlock);
1998
1999 clear_bit(dlm->node_num, nodemap);
2000 if (ignore_higher) {
2001 /* if is this just to clear up mles for nodes below
2002 * this node, do not send the message to the original
2003 * caller or any node number higher than this */
2004 clear_bit(request_from, nodemap);
2005 bit = dlm->node_num;
2006 while (1) {
2007 bit = find_next_bit(nodemap, O2NM_MAX_NODES,
2008 bit+1);
2009 if (bit >= O2NM_MAX_NODES)
2010 break;
2011 clear_bit(bit, nodemap);
2012 }
2013 }
2014
2015 /* this call now finishes out the nodemap
2016 * even if one or more nodes die */
2017 mlog(0, "worker about to master %.*s here, this=%u\n",
2018 res->lockname.len, res->lockname.name, dlm->node_num);
2019 ret = dlm_do_assert_master(dlm, res->lockname.name,
2020 res->lockname.len,
2021 nodemap, flags);
2022 if (ret < 0) {
2023 /* no need to restart, we are done */
2024 mlog_errno(ret);
2025 }
2026
2027 dlm_lockres_put(res);
2028
2029 mlog(0, "finished with dlm_assert_master_worker\n");
2030}
2031
c03872f5
KH		 2032/* SPECIAL CASE for the $RECOVERY lock used by the recovery thread.
2033 * We cannot wait for node recovery to complete to begin mastering this
2034 * lockres because this lockres is used to kick off recovery! ;-)
2035 * So, do a pre-check on all living nodes to see if any of those nodes
2036 * think that $RECOVERY is currently mastered by a dead node. If so,
2037 * we wait a short time to allow that node to get notified by its own
2038 * heartbeat stack, then check again. All $RECOVERY lock resources
2039 * mastered by dead nodes are purged when the hearbeat callback is
2040 * fired, so we can know for sure that it is safe to continue once
2041 * the node returns a live node or no node. */
2042static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
2043 struct dlm_lock_resource *res)
2044{
2045 struct dlm_node_iter iter;
2046 int nodenum;
2047 int ret = 0;
2048 u8 master = DLM_LOCK_RES_OWNER_UNKNOWN;
2049
2050 spin_lock(&dlm->spinlock);
2051 dlm_node_iter_init(dlm->domain_map, &iter);
2052 spin_unlock(&dlm->spinlock);
2053
2054 while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
2055 /* do not send to self */
2056 if (nodenum == dlm->node_num)
2057 continue;
2058 ret = dlm_do_master_requery(dlm, res, nodenum, &master);
2059 if (ret < 0) {
2060 mlog_errno(ret);
2061 if (!dlm_is_host_down(ret))
2062 BUG();
2063 /* host is down, so answer for that node would be
2064 * DLM_LOCK_RES_OWNER_UNKNOWN. continue. */
2065 }
2066
2067 if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
2068 /* check to see if this master is in the recovery map */
2069 spin_lock(&dlm->spinlock);
2070 if (test_bit(master, dlm->recovery_map)) {
2071 mlog(ML_NOTICE, "%s: node %u has not seen "
2072 "node %u go down yet, and thinks the "
2073 "dead node is mastering the recovery "
2074 "lock. must wait.\n", dlm->name,
2075 nodenum, master);
2076 ret = -EAGAIN;
2077 }
2078 spin_unlock(&dlm->spinlock);
2079 mlog(0, "%s: reco lock master is %u\n", dlm->name,
2080 master);
2081 break;
2082 }
2083 }
2084 return ret;
2085}
2086
6714d8e8
KH		 2087
2088/*
2089 * DLM_MIGRATE_LOCKRES
2090 */
2091
2092
2093int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
2094 u8 target)
2095{
2096 struct dlm_master_list_entry *mle = NULL;
2097 struct dlm_master_list_entry *oldmle = NULL;
2098 struct dlm_migratable_lockres *mres = NULL;
2099 int ret = -EINVAL;
2100 const char *name;
2101 unsigned int namelen;
2102 int mle_added = 0;
2103 struct list_head *queue, *iter;
2104 int i;
2105 struct dlm_lock *lock;
2106 int empty = 1;
2107
2108 if (!dlm_grab(dlm))
2109 return -EINVAL;
2110
2111 name = res->lockname.name;
2112 namelen = res->lockname.len;
2113
2114 mlog(0, "migrating %.*s to %u\n", namelen, name, target);
2115
2116 /*
2117 * ensure this lockres is a proper candidate for migration
2118 */
2119 spin_lock(&res->spinlock);
2120 if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
2121 mlog(0, "cannot migrate lockres with unknown owner!\n");
2122 spin_unlock(&res->spinlock);
2123 goto leave;
2124 }
2125 if (res->owner != dlm->node_num) {
2126 mlog(0, "cannot migrate lockres this node doesn't own!\n");
2127 spin_unlock(&res->spinlock);
2128 goto leave;
2129 }
2130 mlog(0, "checking queues...\n");
2131 queue = &res->granted;
2132 for (i=0; i<3; i++) {
2133 list_for_each(iter, queue) {
2134 lock = list_entry (iter, struct dlm_lock, list);
2135 empty = 0;
2136 if (lock->ml.node == dlm->node_num) {
2137 mlog(0, "found a lock owned by this node "
2138 "still on the %s queue! will not "
2139 "migrate this lockres\n",
2140 i==0 ? "granted" :
2141 (i==1 ? "converting" : "blocked"));
2142 spin_unlock(&res->spinlock);
2143 ret = -ENOTEMPTY;
2144 goto leave;
2145 }
2146 }
2147 queue++;
2148 }
2149 mlog(0, "all locks on this lockres are nonlocal. continuing\n");
2150 spin_unlock(&res->spinlock);
2151
2152 /* no work to do */
2153 if (empty) {
2154 mlog(0, "no locks were found on this lockres! done!\n");
2155 ret = 0;
2156 goto leave;
2157 }
2158
2159 /*
2160 * preallocate up front
2161 * if this fails, abort
2162 */
2163
2164 ret = -ENOMEM;
2165 mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_KERNEL);
2166 if (!mres) {
2167 mlog_errno(ret);
2168 goto leave;
2169 }
2170
2171 mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
2172 GFP_KERNEL);
2173 if (!mle) {
2174 mlog_errno(ret);
2175 goto leave;
2176 }
2177 ret = 0;
2178
2179 /*
2180 * find a node to migrate the lockres to
2181 */
2182
2183 mlog(0, "picking a migration node\n");
2184 spin_lock(&dlm->spinlock);
2185 /* pick a new node */
2186 if (!test_bit(target, dlm->domain_map) ||
2187 target >= O2NM_MAX_NODES) {
2188 target = dlm_pick_migration_target(dlm, res);
2189 }
2190 mlog(0, "node %u chosen for migration\n", target);
2191
2192 if (target >= O2NM_MAX_NODES ||
2193 !test_bit(target, dlm->domain_map)) {
2194 /* target chosen is not alive */
2195 ret = -EINVAL;
2196 }
2197
2198 if (ret) {
2199 spin_unlock(&dlm->spinlock);
2200 goto fail;
2201 }
2202
2203 mlog(0, "continuing with target = %u\n", target);
2204
2205 /*
2206 * clear any existing master requests and
2207 * add the migration mle to the list
2208 */
2209 spin_lock(&dlm->master_lock);
2210 ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name,
2211 namelen, target, dlm->node_num);
2212 spin_unlock(&dlm->master_lock);
2213 spin_unlock(&dlm->spinlock);
2214
2215 if (ret == -EEXIST) {
2216 mlog(0, "another process is already migrating it\n");
2217 goto fail;
2218 }
2219 mle_added = 1;
2220
2221 /*
2222 * set the MIGRATING flag and flush asts
2223 * if we fail after this we need to re-dirty the lockres
2224 */
2225 if (dlm_mark_lockres_migrating(dlm, res, target) < 0) {
2226 mlog(ML_ERROR, "tried to migrate %.*s to %u, but "
2227 "the target went down.\n", res->lockname.len,
2228 res->lockname.name, target);
2229 spin_lock(&res->spinlock);
2230 res->state &= ~DLM_LOCK_RES_MIGRATING;
2231 spin_unlock(&res->spinlock);
2232 ret = -EINVAL;
2233 }
2234
2235fail:
2236 if (oldmle) {
2237 /* master is known, detach if not already detached */
2238 dlm_mle_detach_hb_events(dlm, oldmle);
2239 dlm_put_mle(oldmle);
2240 }
2241
2242 if (ret < 0) {
2243 if (mle_added) {
2244 dlm_mle_detach_hb_events(dlm, mle);
2245 dlm_put_mle(mle);
2246 } else if (mle) {
2247 kmem_cache_free(dlm_mle_cache, mle);
2248 }
2249 goto leave;
2250 }
2251
2252 /*
2253 * at this point, we have a migration target, an mle
2254 * in the master list, and the MIGRATING flag set on
2255 * the lockres
2256 */
2257
2258
2259 /* get an extra reference on the mle.
2260 * otherwise the assert_master from the new
2261 * master will destroy this.
2262 * also, make sure that all callers of dlm_get_mle
2263 * take both dlm->spinlock and dlm->master_lock */
2264 spin_lock(&dlm->spinlock);
2265 spin_lock(&dlm->master_lock);
a2bf0477 2266 dlm_get_mle_inuse(mle);
6714d8e8
KH		 2267 spin_unlock(&dlm->master_lock);
2268 spin_unlock(&dlm->spinlock);
2269
2270 /* notify new node and send all lock state */
2271 /* call send_one_lockres with migration flag.
2272 * this serves as notice to the target node that a
2273 * migration is starting. */
2274 ret = dlm_send_one_lockres(dlm, res, mres, target,
2275 DLM_MRES_MIGRATION);
2276
2277 if (ret < 0) {
2278 mlog(0, "migration to node %u failed with %d\n",
2279 target, ret);
2280 /* migration failed, detach and clean up mle */
2281 dlm_mle_detach_hb_events(dlm, mle);
2282 dlm_put_mle(mle);
a2bf0477
KH		 2283 dlm_put_mle_inuse(mle);
2284 spin_lock(&res->spinlock);
2285 res->state &= ~DLM_LOCK_RES_MIGRATING;
2286 spin_unlock(&res->spinlock);
6714d8e8
KH		 2287 goto leave;
2288 }
2289
2290 /* at this point, the target sends a message to all nodes,
2291 * (using dlm_do_migrate_request). this node is skipped since
2292 * we had to put an mle in the list to begin the process. this
2293 * node now waits for target to do an assert master. this node
2294 * will be the last one notified, ensuring that the migration
2295 * is complete everywhere. if the target dies while this is
2296 * going on, some nodes could potentially see the target as the
2297 * master, so it is important that my recovery finds the migration
2298 * mle and sets the master to UNKNONWN. */
2299
2300
2301 /* wait for new node to assert master */
2302 while (1) {
2303 ret = wait_event_interruptible_timeout(mle->wq,
2304 (atomic_read(&mle->woken) == 1),
2305 msecs_to_jiffies(5000));
2306
2307 if (ret >= 0) {
2308 if (atomic_read(&mle->woken) == 1 ||
2309 res->owner == target)
2310 break;
2311
2312 mlog(0, "timed out during migration\n");
e2faea4c
KH		 2313 /* avoid hang during shutdown when migrating lockres
2314 * to a node which also goes down */
2315 if (dlm_is_node_dead(dlm, target)) {
aa852354
KH		 2316 mlog(0, "%s:%.*s: expected migration "
2317 "target %u is no longer up, restarting\n",
e2faea4c
KH		 2318 dlm->name, res->lockname.len,
2319 res->lockname.name, target);
2320 ret = -ERESTARTSYS;
2321 }
6714d8e8
KH		 2322 }
2323 if (ret == -ERESTARTSYS) {
2324 /* migration failed, detach and clean up mle */
2325 dlm_mle_detach_hb_events(dlm, mle);
2326 dlm_put_mle(mle);
a2bf0477
KH		 2327 dlm_put_mle_inuse(mle);
2328 spin_lock(&res->spinlock);
2329 res->state &= ~DLM_LOCK_RES_MIGRATING;
2330 spin_unlock(&res->spinlock);
6714d8e8
KH		 2331 goto leave;
2332 }
2333 /* TODO: if node died: stop, clean up, return error */
2334 }
2335
2336 /* all done, set the owner, clear the flag */
2337 spin_lock(&res->spinlock);
2338 dlm_set_lockres_owner(dlm, res, target);
2339 res->state &= ~DLM_LOCK_RES_MIGRATING;
2340 dlm_remove_nonlocal_locks(dlm, res);
2341 spin_unlock(&res->spinlock);
2342 wake_up(&res->wq);
2343
2344 /* master is known, detach if not already detached */
2345 dlm_mle_detach_hb_events(dlm, mle);
a2bf0477 2346 dlm_put_mle_inuse(mle);
6714d8e8
KH		 2347 ret = 0;
2348
2349 dlm_lockres_calc_usage(dlm, res);
2350
2351leave:
2352 /* re-dirty the lockres if we failed */
2353 if (ret < 0)
2354 dlm_kick_thread(dlm, res);
2355
2356 /* TODO: cleanup */
2357 if (mres)
2358 free_page((unsigned long)mres);
2359
2360 dlm_put(dlm);
2361
2362 mlog(0, "returning %d\n", ret);
2363 return ret;
2364}
2365EXPORT_SYMBOL_GPL(dlm_migrate_lockres);
2366
2367int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock)
2368{
2369 int ret;
2370 spin_lock(&dlm->ast_lock);
2371 spin_lock(&lock->spinlock);
2372 ret = (list_empty(&lock->bast_list) && !lock->bast_pending);
2373 spin_unlock(&lock->spinlock);
2374 spin_unlock(&dlm->ast_lock);
2375 return ret;
2376}
2377
2378static int dlm_migration_can_proceed(struct dlm_ctxt *dlm,
2379 struct dlm_lock_resource *res,
2380 u8 mig_target)
2381{
2382 int can_proceed;
2383 spin_lock(&res->spinlock);
2384 can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING);
2385 spin_unlock(&res->spinlock);
2386
2387 /* target has died, so make the caller break out of the
2388 * wait_event, but caller must recheck the domain_map */
2389 spin_lock(&dlm->spinlock);
2390 if (!test_bit(mig_target, dlm->domain_map))
2391 can_proceed = 1;
2392 spin_unlock(&dlm->spinlock);
2393 return can_proceed;
2394}
2395
2396int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
2397{
2398 int ret;
2399 spin_lock(&res->spinlock);
2400 ret = !!(res->state & DLM_LOCK_RES_DIRTY);
2401 spin_unlock(&res->spinlock);
2402 return ret;
2403}
2404
2405
2406static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
2407 struct dlm_lock_resource *res,
2408 u8 target)
2409{
2410 int ret = 0;
2411
2412 mlog(0, "dlm_mark_lockres_migrating: %.*s, from %u to %u\n",
2413 res->lockname.len, res->lockname.name, dlm->node_num,
2414 target);
2415 /* need to set MIGRATING flag on lockres. this is done by
2416 * ensuring that all asts have been flushed for this lockres. */
2417 spin_lock(&res->spinlock);
2418 BUG_ON(res->migration_pending);
2419 res->migration_pending = 1;
2420 /* strategy is to reserve an extra ast then release
2421 * it below, letting the release do all of the work */
2422 __dlm_lockres_reserve_ast(res);
2423 spin_unlock(&res->spinlock);
2424
2425 /* now flush all the pending asts.. hang out for a bit */
2426 dlm_kick_thread(dlm, res);
2427 wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
2428 dlm_lockres_release_ast(dlm, res);
2429
2430 mlog(0, "about to wait on migration_wq, dirty=%s\n",
2431 res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no");
2432 /* if the extra ref we just put was the final one, this
2433 * will pass thru immediately. otherwise, we need to wait
2434 * for the last ast to finish. */
2435again:
2436 ret = wait_event_interruptible_timeout(dlm->migration_wq,
2437 dlm_migration_can_proceed(dlm, res, target),
2438 msecs_to_jiffies(1000));
2439 if (ret < 0) {
2440 mlog(0, "woken again: migrating? %s, dead? %s\n",
2441 res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2442 test_bit(target, dlm->domain_map) ? "no":"yes");
2443 } else {
2444 mlog(0, "all is well: migrating? %s, dead? %s\n",
2445 res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2446 test_bit(target, dlm->domain_map) ? "no":"yes");
2447 }
2448 if (!dlm_migration_can_proceed(dlm, res, target)) {
2449 mlog(0, "trying again...\n");
2450 goto again;
2451 }
2452
2453 /* did the target go down or die? */
2454 spin_lock(&dlm->spinlock);
2455 if (!test_bit(target, dlm->domain_map)) {
2456 mlog(ML_ERROR, "aha. migration target %u just went down\n",
2457 target);
2458 ret = -EHOSTDOWN;
2459 }
2460 spin_unlock(&dlm->spinlock);
2461
2462 /*
2463 * at this point:
2464 *
2465 * o the DLM_LOCK_RES_MIGRATING flag is set
2466 * o there are no pending asts on this lockres
2467 * o all processes trying to reserve an ast on this
2468 * lockres must wait for the MIGRATING flag to clear
2469 */
2470 return ret;
2471}
2472
2473/* last step in the migration process.
2474 * original master calls this to free all of the dlm_lock
2475 * structures that used to be for other nodes. */
2476static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
2477 struct dlm_lock_resource *res)
2478{
2479 struct list_head *iter, *iter2;
2480 struct list_head *queue = &res->granted;
2481 int i;
2482 struct dlm_lock *lock;
2483
2484 assert_spin_locked(&res->spinlock);
2485
2486 BUG_ON(res->owner == dlm->node_num);
2487
2488 for (i=0; i<3; i++) {
2489 list_for_each_safe(iter, iter2, queue) {
2490 lock = list_entry (iter, struct dlm_lock, list);
2491 if (lock->ml.node != dlm->node_num) {
2492 mlog(0, "putting lock for node %u\n",
2493 lock->ml.node);
2494 /* be extra careful */
2495 BUG_ON(!list_empty(&lock->ast_list));
2496 BUG_ON(!list_empty(&lock->bast_list));
2497 BUG_ON(lock->ast_pending);
2498 BUG_ON(lock->bast_pending);
2499 list_del_init(&lock->list);
2500 dlm_lock_put(lock);
2501 }
2502 }
2503 queue++;
2504 }
2505}
2506
2507/* for now this is not too intelligent. we will
2508 * need stats to make this do the right thing.
2509 * this just finds the first lock on one of the
2510 * queues and uses that node as the target. */
2511static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
2512 struct dlm_lock_resource *res)
2513{
2514 int i;
2515 struct list_head *queue = &res->granted;
2516 struct list_head *iter;
2517 struct dlm_lock *lock;
2518 int nodenum;
2519
2520 assert_spin_locked(&dlm->spinlock);
2521
2522 spin_lock(&res->spinlock);
2523 for (i=0; i<3; i++) {
2524 list_for_each(iter, queue) {
2525 /* up to the caller to make sure this node
2526 * is alive */
2527 lock = list_entry (iter, struct dlm_lock, list);
2528 if (lock->ml.node != dlm->node_num) {
2529 spin_unlock(&res->spinlock);
2530 return lock->ml.node;
2531 }
2532 }
2533 queue++;
2534 }
2535 spin_unlock(&res->spinlock);
2536 mlog(0, "have not found a suitable target yet! checking domain map\n");
2537
2538 /* ok now we're getting desperate. pick anyone alive. */
2539 nodenum = -1;
2540 while (1) {
2541 nodenum = find_next_bit(dlm->domain_map,
2542 O2NM_MAX_NODES, nodenum+1);
2543 mlog(0, "found %d in domain map\n", nodenum);
2544 if (nodenum >= O2NM_MAX_NODES)
2545 break;
2546 if (nodenum != dlm->node_num) {
2547 mlog(0, "picking %d\n", nodenum);
2548 return nodenum;
2549 }
2550 }
2551
2552 mlog(0, "giving up. no master to migrate to\n");
2553 return DLM_LOCK_RES_OWNER_UNKNOWN;
2554}
2555
2556
2557
2558/* this is called by the new master once all lockres
2559 * data has been received */
2560static int dlm_do_migrate_request(struct dlm_ctxt *dlm,
2561 struct dlm_lock_resource *res,
2562 u8 master, u8 new_master,
2563 struct dlm_node_iter *iter)
2564{
2565 struct dlm_migrate_request migrate;
2566 int ret, status = 0;
2567 int nodenum;
2568
2569 memset(&migrate, 0, sizeof(migrate));
2570 migrate.namelen = res->lockname.len;
2571 memcpy(migrate.name, res->lockname.name, migrate.namelen);
2572 migrate.new_master = new_master;
2573 migrate.master = master;
2574
2575 ret = 0;
2576
2577 /* send message to all nodes, except the master and myself */
2578 while ((nodenum = dlm_node_iter_next(iter)) >= 0) {
2579 if (nodenum == master ||
2580 nodenum == new_master)
2581 continue;
2582
2583 ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
2584 &migrate, sizeof(migrate), nodenum,
2585 &status);
2586 if (ret < 0)
2587 mlog_errno(ret);
2588 else if (status < 0) {
2589 mlog(0, "migrate request (node %u) returned %d!\n",
2590 nodenum, status);
2591 ret = status;
2592 }
2593 }
2594
2595 if (ret < 0)
2596 mlog_errno(ret);
2597
2598 mlog(0, "returning ret=%d\n", ret);
2599 return ret;
2600}
2601
2602
2603/* if there is an existing mle for this lockres, we now know who the master is.
2604 * (the one who sent us *this* message) we can clear it up right away.
2605 * since the process that put the mle on the list still has a reference to it,
2606 * we can unhash it now, set the master and wake the process. as a result,
2607 * we will have no mle in the list to start with. now we can add an mle for
2608 * the migration and this should be the only one found for those scanning the
2609 * list. */
2610int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data)
2611{
2612 struct dlm_ctxt *dlm = data;
2613 struct dlm_lock_resource *res = NULL;
2614 struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
2615 struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
2616 const char *name;
a3d33291 2617 unsigned int namelen, hash;
6714d8e8
KH		 2618 int ret = 0;
2619
2620 if (!dlm_grab(dlm))
2621 return -EINVAL;
2622
2623 name = migrate->name;
2624 namelen = migrate->namelen;
a3d33291 2625 hash = dlm_lockid_hash(name, namelen);
6714d8e8
KH		 2626
2627 /* preallocate.. if this fails, abort */
2628 mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
2629 GFP_KERNEL);
2630
2631 if (!mle) {
2632 ret = -ENOMEM;
2633 goto leave;
2634 }
2635
2636 /* check for pre-existing lock */
2637 spin_lock(&dlm->spinlock);
a3d33291 2638 res = __dlm_lookup_lockres(dlm, name, namelen, hash);
6714d8e8
KH		 2639 spin_lock(&dlm->master_lock);
2640
2641 if (res) {
2642 spin_lock(&res->spinlock);
2643 if (res->state & DLM_LOCK_RES_RECOVERING) {
2644 /* if all is working ok, this can only mean that we got
2645 * a migrate request from a node that we now see as
2646 * dead. what can we do here? drop it to the floor? */
2647 spin_unlock(&res->spinlock);
2648 mlog(ML_ERROR, "Got a migrate request, but the "
2649 "lockres is marked as recovering!");
2650 kmem_cache_free(dlm_mle_cache, mle);
2651 ret = -EINVAL; /* need a better solution */
2652 goto unlock;
2653 }
2654 res->state |= DLM_LOCK_RES_MIGRATING;
2655 spin_unlock(&res->spinlock);
2656 }
2657
2658 /* ignore status. only nonzero status would BUG. */
2659 ret = dlm_add_migration_mle(dlm, res, mle, &oldmle,
2660 name, namelen,
2661 migrate->new_master,
2662 migrate->master);
2663
2664unlock:
2665 spin_unlock(&dlm->master_lock);
2666 spin_unlock(&dlm->spinlock);
2667
2668 if (oldmle) {
2669 /* master is known, detach if not already detached */
2670 dlm_mle_detach_hb_events(dlm, oldmle);
2671 dlm_put_mle(oldmle);
2672 }
2673
2674 if (res)
2675 dlm_lockres_put(res);
2676leave:
2677 dlm_put(dlm);
2678 return ret;
2679}
2680
2681/* must be holding dlm->spinlock and dlm->master_lock
2682 * when adding a migration mle, we can clear any other mles
2683 * in the master list because we know with certainty that
2684 * the master is "master". so we remove any old mle from
2685 * the list after setting it's master field, and then add
2686 * the new migration mle. this way we can hold with the rule
2687 * of having only one mle for a given lock name at all times. */
2688static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
2689 struct dlm_lock_resource *res,
2690 struct dlm_master_list_entry *mle,
2691 struct dlm_master_list_entry **oldmle,
2692 const char *name, unsigned int namelen,
2693 u8 new_master, u8 master)
2694{
2695 int found;
2696 int ret = 0;
2697
2698 *oldmle = NULL;
2699
2700 mlog_entry_void();
2701
2702 assert_spin_locked(&dlm->spinlock);
2703 assert_spin_locked(&dlm->master_lock);
2704
2705 /* caller is responsible for any ref taken here on oldmle */
2706 found = dlm_find_mle(dlm, oldmle, (char *)name, namelen);
2707 if (found) {
2708 struct dlm_master_list_entry *tmp = *oldmle;
2709 spin_lock(&tmp->spinlock);
2710 if (tmp->type == DLM_MLE_MIGRATION) {
2711 if (master == dlm->node_num) {
2712 /* ah another process raced me to it */
2713 mlog(0, "tried to migrate %.*s, but some "
2714 "process beat me to it\n",
2715 namelen, name);
2716 ret = -EEXIST;
2717 } else {
2718 /* bad. 2 NODES are trying to migrate! */
2719 mlog(ML_ERROR, "migration error mle: "
2720 "master=%u new_master=%u // request: "
2721 "master=%u new_master=%u // "
2722 "lockres=%.*s\n",
2723 tmp->master, tmp->new_master,
2724 master, new_master,
2725 namelen, name);
2726 BUG();
2727 }
2728 } else {
2729 /* this is essentially what assert_master does */
2730 tmp->master = master;
2731 atomic_set(&tmp->woken, 1);
2732 wake_up(&tmp->wq);
2733 /* remove it from the list so that only one
2734 * mle will be found */
2735 list_del_init(&tmp->list);
da01ad05 2736 __dlm_mle_detach_hb_events(dlm, mle);
6714d8e8
KH		 2737 }
2738 spin_unlock(&tmp->spinlock);
2739 }
2740
2741 /* now add a migration mle to the tail of the list */
2742 dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen);
2743 mle->new_master = new_master;
2744 mle->master = master;
2745 /* do this for consistency with other mle types */
2746 set_bit(new_master, mle->maybe_map);
2747 list_add(&mle->list, &dlm->master_list);
2748
2749 return ret;
2750}
2751
2752
2753void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
2754{
2755 struct list_head *iter, *iter2;
2756 struct dlm_master_list_entry *mle;
2757 struct dlm_lock_resource *res;
a3d33291 2758 unsigned int hash;
6714d8e8
KH		 2759
2760 mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
2761top:
2762 assert_spin_locked(&dlm->spinlock);
2763
2764 /* clean the master list */
2765 spin_lock(&dlm->master_lock);
2766 list_for_each_safe(iter, iter2, &dlm->master_list) {
2767 mle = list_entry(iter, struct dlm_master_list_entry, list);
2768
2769 BUG_ON(mle->type != DLM_MLE_BLOCK &&
2770 mle->type != DLM_MLE_MASTER &&
2771 mle->type != DLM_MLE_MIGRATION);
2772
2773 /* MASTER mles are initiated locally. the waiting
2774 * process will notice the node map change
2775 * shortly. let that happen as normal. */
2776 if (mle->type == DLM_MLE_MASTER)
2777 continue;
2778
2779
2780 /* BLOCK mles are initiated by other nodes.
2781 * need to clean up if the dead node would have
2782 * been the master. */
2783 if (mle->type == DLM_MLE_BLOCK) {
2784 int bit;
2785
2786 spin_lock(&mle->spinlock);
2787 bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
2788 if (bit != dead_node) {
2789 mlog(0, "mle found, but dead node %u would "
2790 "not have been master\n", dead_node);
2791 spin_unlock(&mle->spinlock);
2792 } else {
2793 /* must drop the refcount by one since the
2794 * assert_master will never arrive. this
2795 * may result in the mle being unlinked and
2796 * freed, but there may still be a process
2797 * waiting in the dlmlock path which is fine. */
2798 mlog(ML_ERROR, "node %u was expected master\n",
2799 dead_node);
2800 atomic_set(&mle->woken, 1);
2801 spin_unlock(&mle->spinlock);
2802 wake_up(&mle->wq);
f671c09b
KH		 2803 /* do not need events any longer, so detach
2804 * from heartbeat */
2805 __dlm_mle_detach_hb_events(dlm, mle);
6714d8e8
KH		 2806 __dlm_put_mle(mle);
2807 }
2808 continue;
2809 }
2810
2811 /* everything else is a MIGRATION mle */
2812
2813 /* the rule for MIGRATION mles is that the master
2814 * becomes UNKNOWN if *either* the original or
2815 * the new master dies. all UNKNOWN lockreses
2816 * are sent to whichever node becomes the recovery
2817 * master. the new master is responsible for
2818 * determining if there is still a master for
2819 * this lockres, or if he needs to take over
2820 * mastery. either way, this node should expect
2821 * another message to resolve this. */
2822 if (mle->master != dead_node &&
2823 mle->new_master != dead_node)
2824 continue;
2825
2826 /* if we have reached this point, this mle needs to
2827 * be removed from the list and freed. */
2828
2829 /* remove from the list early. NOTE: unlinking
2830 * list_head while in list_for_each_safe */
da01ad05 2831 __dlm_mle_detach_hb_events(dlm, mle);
6714d8e8
KH		 2832 spin_lock(&mle->spinlock);
2833 list_del_init(&mle->list);
2834 atomic_set(&mle->woken, 1);
2835 spin_unlock(&mle->spinlock);
2836 wake_up(&mle->wq);
2837
aa852354
KH		 2838 mlog(0, "%s: node %u died during migration from "
2839 "%u to %u!\n", dlm->name, dead_node,
6714d8e8
KH		 2840 mle->master, mle->new_master);
2841 /* if there is a lockres associated with this
2842 * mle, find it and set its owner to UNKNOWN */
a3d33291 2843 hash = dlm_lockid_hash(mle->u.name.name, mle->u.name.len);
6714d8e8 2844 res = __dlm_lookup_lockres(dlm, mle->u.name.name,
a3d33291 2845 mle->u.name.len, hash);
6714d8e8
KH		 2846 if (res) {
2847 /* unfortunately if we hit this rare case, our
2848 * lock ordering is messed. we need to drop
2849 * the master lock so that we can take the
2850 * lockres lock, meaning that we will have to
2851 * restart from the head of list. */
2852 spin_unlock(&dlm->master_lock);
2853
2854 /* move lockres onto recovery list */
2855 spin_lock(&res->spinlock);
2856 dlm_set_lockres_owner(dlm, res,
2857 DLM_LOCK_RES_OWNER_UNKNOWN);
2858 dlm_move_lockres_to_recovery_list(dlm, res);
2859 spin_unlock(&res->spinlock);
2860 dlm_lockres_put(res);
2861
f671c09b
KH		 2862 /* about to get rid of mle, detach from heartbeat */
2863 __dlm_mle_detach_hb_events(dlm, mle);
2864
6714d8e8
KH		 2865 /* dump the mle */
2866 spin_lock(&dlm->master_lock);
2867 __dlm_put_mle(mle);
2868 spin_unlock(&dlm->master_lock);
2869
2870 /* restart */
2871 goto top;
2872 }
2873
2874 /* this may be the last reference */
2875 __dlm_put_mle(mle);
2876 }
2877 spin_unlock(&dlm->master_lock);
2878}
2879
2880
2881int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
2882 u8 old_master)
2883{
2884 struct dlm_node_iter iter;
2885 int ret = 0;
2886
2887 spin_lock(&dlm->spinlock);
2888 dlm_node_iter_init(dlm->domain_map, &iter);
2889 clear_bit(old_master, iter.node_map);
2890 clear_bit(dlm->node_num, iter.node_map);
2891 spin_unlock(&dlm->spinlock);
2892
2893 mlog(0, "now time to do a migrate request to other nodes\n");
2894 ret = dlm_do_migrate_request(dlm, res, old_master,
2895 dlm->node_num, &iter);
2896 if (ret < 0) {
2897 mlog_errno(ret);
2898 goto leave;
2899 }
2900
2901 mlog(0, "doing assert master of %.*s to all except the original node\n",
2902 res->lockname.len, res->lockname.name);
2903 /* this call now finishes out the nodemap
2904 * even if one or more nodes die */
2905 ret = dlm_do_assert_master(dlm, res->lockname.name,
2906 res->lockname.len, iter.node_map,
2907 DLM_ASSERT_MASTER_FINISH_MIGRATION);
2908 if (ret < 0) {
2909 /* no longer need to retry. all living nodes contacted. */
2910 mlog_errno(ret);
2911 ret = 0;
2912 }
2913
2914 memset(iter.node_map, 0, sizeof(iter.node_map));
2915 set_bit(old_master, iter.node_map);
2916 mlog(0, "doing assert master of %.*s back to %u\n",
2917 res->lockname.len, res->lockname.name, old_master);
2918 ret = dlm_do_assert_master(dlm, res->lockname.name,
2919 res->lockname.len, iter.node_map,
2920 DLM_ASSERT_MASTER_FINISH_MIGRATION);
2921 if (ret < 0) {
2922 mlog(0, "assert master to original master failed "
2923 "with %d.\n", ret);
2924 /* the only nonzero status here would be because of
2925 * a dead original node. we're done. */
2926 ret = 0;
2927 }
2928
2929 /* all done, set the owner, clear the flag */
2930 spin_lock(&res->spinlock);
2931 dlm_set_lockres_owner(dlm, res, dlm->node_num);
2932 res->state &= ~DLM_LOCK_RES_MIGRATING;
2933 spin_unlock(&res->spinlock);
2934 /* re-dirty it on the new master */
2935 dlm_kick_thread(dlm, res);
2936 wake_up(&res->wq);
2937leave:
2938 return ret;
2939}
2940
2941/*
2942 * LOCKRES AST REFCOUNT
2943 * this is integral to migration
2944 */
2945
2946/* for future intent to call an ast, reserve one ahead of time.
2947 * this should be called only after waiting on the lockres
2948 * with dlm_wait_on_lockres, and while still holding the
2949 * spinlock after the call. */
2950void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res)
2951{
2952 assert_spin_locked(&res->spinlock);
2953 if (res->state & DLM_LOCK_RES_MIGRATING) {
2954 __dlm_print_one_lock_resource(res);
2955 }
2956 BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
2957
2958 atomic_inc(&res->asts_reserved);
2959}
2960
2961/*
2962 * used to drop the reserved ast, either because it went unused,
2963 * or because the ast/bast was actually called.
2964 *
2965 * also, if there is a pending migration on this lockres,
2966 * and this was the last pending ast on the lockres,
2967 * atomically set the MIGRATING flag before we drop the lock.
2968 * this is how we ensure that migration can proceed with no
2969 * asts in progress. note that it is ok if the state of the
2970 * queues is such that a lock should be granted in the future
2971 * or that a bast should be fired, because the new master will
2972 * shuffle the lists on this lockres as soon as it is migrated.
2973 */
2974void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
2975 struct dlm_lock_resource *res)
2976{
2977 if (!atomic_dec_and_lock(&res->asts_reserved, &res->spinlock))
2978 return;
2979
2980 if (!res->migration_pending) {
2981 spin_unlock(&res->spinlock);
2982 return;
2983 }
2984
2985 BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
2986 res->migration_pending = 0;
2987 res->state |= DLM_LOCK_RES_MIGRATING;
2988 spin_unlock(&res->spinlock);
2989 wake_up(&res->wq);
2990 wake_up(&dlm->migration_wq);
2991}
Linux 6.x block layer and io_uring tree(s)