aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/dlm/dlmmaster.c
diff options
context:
space:
mode:
authorKurt Hackel <kurt.hackel@oracle.com>2005-12-15 17:31:23 -0500
committerJoel Becker <joel.becker@oracle.com>2006-01-03 14:45:47 -0500
commit6714d8e86bf443f6f7af50f9d432025649f091f5 (patch)
tree2c484bd1894a90cad7020869c7054f192d3bf34d /fs/ocfs2/dlm/dlmmaster.c
parent98211489d4147e41b11703e4245846d60b3acce4 (diff)
[PATCH] OCFS2: The Second Oracle Cluster Filesystem
A distributed lock manager built with the cluster file system use case in mind. The OCFS2 dlm exposes a VMS style API, though things have been simplified internally. The only lock levels implemented currently are NLMODE, PRMODE and EXMODE. Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Diffstat (limited to 'fs/ocfs2/dlm/dlmmaster.c')
-rw-r--r--fs/ocfs2/dlm/dlmmaster.c2666
1 files changed, 2666 insertions, 0 deletions
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
new file mode 100644
index 000000000000..047279546b4f
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -0,0 +1,2666 @@
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"
51
52#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
53#include "cluster/masklog.h"
54
55enum dlm_mle_type {
56 DLM_MLE_BLOCK,
57 DLM_MLE_MASTER,
58 DLM_MLE_MIGRATION
59};
60
61struct dlm_lock_name
62{
63 u8 len;
64 u8 name[DLM_LOCKID_NAME_MAX];
65};
66
67struct dlm_master_list_entry
68{
69 struct list_head list;
70 struct list_head hb_events;
71 struct dlm_ctxt *dlm;
72 spinlock_t spinlock;
73 wait_queue_head_t wq;
74 atomic_t woken;
75 struct kref mle_refs;
76 unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
77 unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
78 unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
79 unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
80 u8 master;
81 u8 new_master;
82 enum dlm_mle_type type;
83 struct o2hb_callback_func mle_hb_up;
84 struct o2hb_callback_func mle_hb_down;
85 union {
86 struct dlm_lock_resource *res;
87 struct dlm_lock_name name;
88 } u;
89};
90
91static void dlm_mle_node_down(struct dlm_ctxt *dlm,
92 struct dlm_master_list_entry *mle,
93 struct o2nm_node *node,
94 int idx);
95static void dlm_mle_node_up(struct dlm_ctxt *dlm,
96 struct dlm_master_list_entry *mle,
97 struct o2nm_node *node,
98 int idx);
99
100static void dlm_assert_master_worker(struct dlm_work_item *item, void *data);
101static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
102 unsigned int namelen, void *nodemap,
103 u32 flags);
104
105static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
106 struct dlm_master_list_entry *mle,
107 const char *name,
108 unsigned int namelen)
109{
110 struct dlm_lock_resource *res;
111
112 if (dlm != mle->dlm)
113 return 0;
114
115 if (mle->type == DLM_MLE_BLOCK ||
116 mle->type == DLM_MLE_MIGRATION) {
117 if (namelen != mle->u.name.len ||
118 memcmp(name, mle->u.name.name, namelen)!=0)
119 return 0;
120 } else {
121 res = mle->u.res;
122 if (namelen != res->lockname.len ||
123 memcmp(res->lockname.name, name, namelen) != 0)
124 return 0;
125 }
126 return 1;
127}
128
129#if 0
130/* Code here is included but defined out as it aids debugging */
131
132void dlm_print_one_mle(struct dlm_master_list_entry *mle)
133{
134 int i = 0, refs;
135 char *type;
136 char attached;
137 u8 master;
138 unsigned int namelen;
139 const char *name;
140 struct kref *k;
141
142 k = &mle->mle_refs;
143 if (mle->type == DLM_MLE_BLOCK)
144 type = "BLK";
145 else if (mle->type == DLM_MLE_MASTER)
146 type = "MAS";
147 else
148 type = "MIG";
149 refs = atomic_read(&k->refcount);
150 master = mle->master;
151 attached = (list_empty(&mle->hb_events) ? 'N' : 'Y');
152
153 if (mle->type != DLM_MLE_MASTER) {
154 namelen = mle->u.name.len;
155 name = mle->u.name.name;
156 } else {
157 namelen = mle->u.res->lockname.len;
158 name = mle->u.res->lockname.name;
159 }
160
161 mlog(ML_NOTICE, " #%3d: %3s %3d %3u %3u %c (%d)%.*s\n",
162 i, type, refs, master, mle->new_master, attached,
163 namelen, namelen, name);
164}
165
166static void dlm_dump_mles(struct dlm_ctxt *dlm)
167{
168 struct dlm_master_list_entry *mle;
169 struct list_head *iter;
170
171 mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name);
172 mlog(ML_NOTICE, " ####: type refs owner new events? lockname nodemap votemap respmap maybemap\n");
173 spin_lock(&dlm->master_lock);
174 list_for_each(iter, &dlm->master_list) {
175 mle = list_entry(iter, struct dlm_master_list_entry, list);
176 dlm_print_one_mle(mle);
177 }
178 spin_unlock(&dlm->master_lock);
179}
180
181extern spinlock_t dlm_domain_lock;
182extern struct list_head dlm_domains;
183
184int dlm_dump_all_mles(const char __user *data, unsigned int len)
185{
186 struct list_head *iter;
187 struct dlm_ctxt *dlm;
188
189 spin_lock(&dlm_domain_lock);
190 list_for_each(iter, &dlm_domains) {
191 dlm = list_entry (iter, struct dlm_ctxt, list);
192 mlog(ML_NOTICE, "found dlm: %p, name=%s\n", dlm, dlm->name);
193 dlm_dump_mles(dlm);
194 }
195 spin_unlock(&dlm_domain_lock);
196 return len;
197}
198EXPORT_SYMBOL_GPL(dlm_dump_all_mles);
199
200#endif /* 0 */
201
202
203static kmem_cache_t *dlm_mle_cache = NULL;
204
205
206static void dlm_mle_release(struct kref *kref);
207static void dlm_init_mle(struct dlm_master_list_entry *mle,
208 enum dlm_mle_type type,
209 struct dlm_ctxt *dlm,
210 struct dlm_lock_resource *res,
211 const char *name,
212 unsigned int namelen);
213static void dlm_put_mle(struct dlm_master_list_entry *mle);
214static void __dlm_put_mle(struct dlm_master_list_entry *mle);
215static int dlm_find_mle(struct dlm_ctxt *dlm,
216 struct dlm_master_list_entry **mle,
217 char *name, unsigned int namelen);
218
219static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to);
220
221
222static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
223 struct dlm_lock_resource *res,
224 struct dlm_master_list_entry *mle,
225 int *blocked);
226static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
227 struct dlm_lock_resource *res,
228 struct dlm_master_list_entry *mle,
229 int blocked);
230static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
231 struct dlm_lock_resource *res,
232 struct dlm_master_list_entry *mle,
233 struct dlm_master_list_entry **oldmle,
234 const char *name, unsigned int namelen,
235 u8 new_master, u8 master);
236
237static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
238 struct dlm_lock_resource *res);
239static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
240 struct dlm_lock_resource *res);
241static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
242 struct dlm_lock_resource *res,
243 u8 target);
244
245
246int dlm_is_host_down(int errno)
247{
248 switch (errno) {
249 case -EBADF:
250 case -ECONNREFUSED:
251 case -ENOTCONN:
252 case -ECONNRESET:
253 case -EPIPE:
254 case -EHOSTDOWN:
255 case -EHOSTUNREACH:
256 case -ETIMEDOUT:
257 case -ECONNABORTED:
258 case -ENETDOWN:
259 case -ENETUNREACH:
260 case -ENETRESET:
261 case -ESHUTDOWN:
262 case -ENOPROTOOPT:
263 case -EINVAL: /* if returned from our tcp code,
264 this means there is no socket */
265 return 1;
266 }
267 return 0;
268}
269
270
271/*
272 * MASTER LIST FUNCTIONS
273 */
274
275
276/*
277 * regarding master list entries and heartbeat callbacks:
278 *
279 * in order to avoid sleeping and allocation that occurs in
280 * heartbeat, master list entries are simply attached to the
281 * dlm's established heartbeat callbacks. the mle is attached
282 * when it is created, and since the dlm->spinlock is held at
283 * that time, any heartbeat event will be properly discovered
284 * by the mle. the mle needs to be detached from the
285 * dlm->mle_hb_events list as soon as heartbeat events are no
286 * longer useful to the mle, and before the mle is freed.
287 *
288 * as a general rule, heartbeat events are no longer needed by
289 * the mle once an "answer" regarding the lock master has been
290 * received.
291 */
292static inline void __dlm_mle_attach_hb_events(struct dlm_ctxt *dlm,
293 struct dlm_master_list_entry *mle)
294{
295 assert_spin_locked(&dlm->spinlock);
296
297 list_add_tail(&mle->hb_events, &dlm->mle_hb_events);
298}
299
300
301static inline void __dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
302 struct dlm_master_list_entry *mle)
303{
304 if (!list_empty(&mle->hb_events))
305 list_del_init(&mle->hb_events);
306}
307
308
309static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
310 struct dlm_master_list_entry *mle)
311{
312 spin_lock(&dlm->spinlock);
313 __dlm_mle_detach_hb_events(dlm, mle);
314 spin_unlock(&dlm->spinlock);
315}
316
317/* remove from list and free */
318static void __dlm_put_mle(struct dlm_master_list_entry *mle)
319{
320 struct dlm_ctxt *dlm;
321 dlm = mle->dlm;
322
323 assert_spin_locked(&dlm->spinlock);
324 assert_spin_locked(&dlm->master_lock);
325 BUG_ON(!atomic_read(&mle->mle_refs.refcount));
326
327 kref_put(&mle->mle_refs, dlm_mle_release);
328}
329
330
331/* must not have any spinlocks coming in */
332static void dlm_put_mle(struct dlm_master_list_entry *mle)
333{
334 struct dlm_ctxt *dlm;
335 dlm = mle->dlm;
336
337 spin_lock(&dlm->spinlock);
338 spin_lock(&dlm->master_lock);
339 __dlm_put_mle(mle);
340 spin_unlock(&dlm->master_lock);
341 spin_unlock(&dlm->spinlock);
342}
343
344static inline void dlm_get_mle(struct dlm_master_list_entry *mle)
345{
346 kref_get(&mle->mle_refs);
347}
348
349static void dlm_init_mle(struct dlm_master_list_entry *mle,
350 enum dlm_mle_type type,
351 struct dlm_ctxt *dlm,
352 struct dlm_lock_resource *res,
353 const char *name,
354 unsigned int namelen)
355{
356 assert_spin_locked(&dlm->spinlock);
357
358 mle->dlm = dlm;
359 mle->type = type;
360 INIT_LIST_HEAD(&mle->list);
361 INIT_LIST_HEAD(&mle->hb_events);
362 memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
363 spin_lock_init(&mle->spinlock);
364 init_waitqueue_head(&mle->wq);
365 atomic_set(&mle->woken, 0);
366 kref_init(&mle->mle_refs);
367 memset(mle->response_map, 0, sizeof(mle->response_map));
368 mle->master = O2NM_MAX_NODES;
369 mle->new_master = O2NM_MAX_NODES;
370
371 if (mle->type == DLM_MLE_MASTER) {
372 BUG_ON(!res);
373 mle->u.res = res;
374 } else if (mle->type == DLM_MLE_BLOCK) {
375 BUG_ON(!name);
376 memcpy(mle->u.name.name, name, namelen);
377 mle->u.name.len = namelen;
378 } else /* DLM_MLE_MIGRATION */ {
379 BUG_ON(!name);
380 memcpy(mle->u.name.name, name, namelen);
381 mle->u.name.len = namelen;
382 }
383
384 /* copy off the node_map and register hb callbacks on our copy */
385 memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map));
386 memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map));
387 clear_bit(dlm->node_num, mle->vote_map);
388 clear_bit(dlm->node_num, mle->node_map);
389
390 /* attach the mle to the domain node up/down events */
391 __dlm_mle_attach_hb_events(dlm, mle);
392}
393
394
395/* returns 1 if found, 0 if not */
396static int dlm_find_mle(struct dlm_ctxt *dlm,
397 struct dlm_master_list_entry **mle,
398 char *name, unsigned int namelen)
399{
400 struct dlm_master_list_entry *tmpmle;
401 struct list_head *iter;
402
403 assert_spin_locked(&dlm->master_lock);
404
405 list_for_each(iter, &dlm->master_list) {
406 tmpmle = list_entry(iter, struct dlm_master_list_entry, list);
407 if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
408 continue;
409 dlm_get_mle(tmpmle);
410 *mle = tmpmle;
411 return 1;
412 }
413 return 0;
414}
415
416void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up)
417{
418 struct dlm_master_list_entry *mle;
419 struct list_head *iter;
420
421 assert_spin_locked(&dlm->spinlock);
422
423 list_for_each(iter, &dlm->mle_hb_events) {
424 mle = list_entry(iter, struct dlm_master_list_entry,
425 hb_events);
426 if (node_up)
427 dlm_mle_node_up(dlm, mle, NULL, idx);
428 else
429 dlm_mle_node_down(dlm, mle, NULL, idx);
430 }
431}
432
433static void dlm_mle_node_down(struct dlm_ctxt *dlm,
434 struct dlm_master_list_entry *mle,
435 struct o2nm_node *node, int idx)
436{
437 spin_lock(&mle->spinlock);
438
439 if (!test_bit(idx, mle->node_map))
440 mlog(0, "node %u already removed from nodemap!\n", idx);
441 else
442 clear_bit(idx, mle->node_map);
443
444 spin_unlock(&mle->spinlock);
445}
446
447static void dlm_mle_node_up(struct dlm_ctxt *dlm,
448 struct dlm_master_list_entry *mle,
449 struct o2nm_node *node, int idx)
450{
451 spin_lock(&mle->spinlock);
452
453 if (test_bit(idx, mle->node_map))
454 mlog(0, "node %u already in node map!\n", idx);
455 else
456 set_bit(idx, mle->node_map);
457
458 spin_unlock(&mle->spinlock);
459}
460
461
462int dlm_init_mle_cache(void)
463{
464 dlm_mle_cache = kmem_cache_create("dlm_mle_cache",
465 sizeof(struct dlm_master_list_entry),
466 0, SLAB_HWCACHE_ALIGN,
467 NULL, NULL);
468 if (dlm_mle_cache == NULL)
469 return -ENOMEM;
470 return 0;
471}
472
473void dlm_destroy_mle_cache(void)
474{
475 if (dlm_mle_cache)
476 kmem_cache_destroy(dlm_mle_cache);
477}
478
479static void dlm_mle_release(struct kref *kref)
480{
481 struct dlm_master_list_entry *mle;
482 struct dlm_ctxt *dlm;
483
484 mlog_entry_void();
485
486 mle = container_of(kref, struct dlm_master_list_entry, mle_refs);
487 dlm = mle->dlm;
488
489 if (mle->type != DLM_MLE_MASTER) {
490 mlog(0, "calling mle_release for %.*s, type %d\n",
491 mle->u.name.len, mle->u.name.name, mle->type);
492 } else {
493 mlog(0, "calling mle_release for %.*s, type %d\n",
494 mle->u.res->lockname.len,
495 mle->u.res->lockname.name, mle->type);
496 }
497 assert_spin_locked(&dlm->spinlock);
498 assert_spin_locked(&dlm->master_lock);
499
500 /* remove from list if not already */
501 if (!list_empty(&mle->list))
502 list_del_init(&mle->list);
503
504 /* detach the mle from the domain node up/down events */
505 __dlm_mle_detach_hb_events(dlm, mle);
506
507 /* NOTE: kfree under spinlock here.
508 * if this is bad, we can move this to a freelist. */
509 kmem_cache_free(dlm_mle_cache, mle);
510}
511
512
513/*
514 * LOCK RESOURCE FUNCTIONS
515 */
516
517static void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
518 struct dlm_lock_resource *res,
519 u8 owner)
520{
521 assert_spin_locked(&res->spinlock);
522
523 mlog_entry("%.*s, %u\n", res->lockname.len, res->lockname.name, owner);
524
525 if (owner == dlm->node_num)
526 atomic_inc(&dlm->local_resources);
527 else if (owner == DLM_LOCK_RES_OWNER_UNKNOWN)
528 atomic_inc(&dlm->unknown_resources);
529 else
530 atomic_inc(&dlm->remote_resources);
531
532 res->owner = owner;
533}
534
535void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
536 struct dlm_lock_resource *res, u8 owner)
537{
538 assert_spin_locked(&res->spinlock);
539
540 if (owner == res->owner)
541 return;
542
543 if (res->owner == dlm->node_num)
544 atomic_dec(&dlm->local_resources);
545 else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN)
546 atomic_dec(&dlm->unknown_resources);
547 else
548 atomic_dec(&dlm->remote_resources);
549
550 dlm_set_lockres_owner(dlm, res, owner);
551}
552
553
554static void dlm_lockres_release(struct kref *kref)
555{
556 struct dlm_lock_resource *res;
557
558 res = container_of(kref, struct dlm_lock_resource, refs);
559
560 /* This should not happen -- all lockres' have a name
561 * associated with them at init time. */
562 BUG_ON(!res->lockname.name);
563
564 mlog(0, "destroying lockres %.*s\n", res->lockname.len,
565 res->lockname.name);
566
567 /* By the time we're ready to blow this guy away, we shouldn't
568 * be on any lists. */
569 BUG_ON(!list_empty(&res->list));
570 BUG_ON(!list_empty(&res->granted));
571 BUG_ON(!list_empty(&res->converting));
572 BUG_ON(!list_empty(&res->blocked));
573 BUG_ON(!list_empty(&res->dirty));
574 BUG_ON(!list_empty(&res->recovering));
575 BUG_ON(!list_empty(&res->purge));
576
577 kfree(res->lockname.name);
578
579 kfree(res);
580}
581
582void dlm_lockres_get(struct dlm_lock_resource *res)
583{
584 kref_get(&res->refs);
585}
586
587void dlm_lockres_put(struct dlm_lock_resource *res)
588{
589 kref_put(&res->refs, dlm_lockres_release);
590}
591
592static void dlm_init_lockres(struct dlm_ctxt *dlm,
593 struct dlm_lock_resource *res,
594 const char *name, unsigned int namelen)
595{
596 char *qname;
597
598 /* If we memset here, we lose our reference to the kmalloc'd
599 * res->lockname.name, so be sure to init every field
600 * correctly! */
601
602 qname = (char *) res->lockname.name;
603 memcpy(qname, name, namelen);
604
605 res->lockname.len = namelen;
606 res->lockname.hash = full_name_hash(name, namelen);
607
608 init_waitqueue_head(&res->wq);
609 spin_lock_init(&res->spinlock);
610 INIT_LIST_HEAD(&res->list);
611 INIT_LIST_HEAD(&res->granted);
612 INIT_LIST_HEAD(&res->converting);
613 INIT_LIST_HEAD(&res->blocked);
614 INIT_LIST_HEAD(&res->dirty);
615 INIT_LIST_HEAD(&res->recovering);
616 INIT_LIST_HEAD(&res->purge);
617 atomic_set(&res->asts_reserved, 0);
618 res->migration_pending = 0;
619
620 kref_init(&res->refs);
621
622 /* just for consistency */
623 spin_lock(&res->spinlock);
624 dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
625 spin_unlock(&res->spinlock);
626
627 res->state = DLM_LOCK_RES_IN_PROGRESS;
628
629 res->last_used = 0;
630
631 memset(res->lvb, 0, DLM_LVB_LEN);
632}
633
634struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
635 const char *name,
636 unsigned int namelen)
637{
638 struct dlm_lock_resource *res;
639
640 res = kmalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
641 if (!res)
642 return NULL;
643
644 res->lockname.name = kmalloc(namelen, GFP_KERNEL);
645 if (!res->lockname.name) {
646 kfree(res);
647 return NULL;
648 }
649
650 dlm_init_lockres(dlm, res, name, namelen);
651 return res;
652}
653
654/*
655 * lookup a lock resource by name.
656 * may already exist in the hashtable.
657 * lockid is null terminated
658 *
659 * if not, allocate enough for the lockres and for
660 * the temporary structure used in doing the mastering.
661 *
662 * also, do a lookup in the dlm->master_list to see
663 * if another node has begun mastering the same lock.
664 * if so, there should be a block entry in there
665 * for this name, and we should *not* attempt to master
666 * the lock here. need to wait around for that node
667 * to assert_master (or die).
668 *
669 */
670struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
671 const char *lockid,
672 int flags)
673{
674 struct dlm_lock_resource *tmpres=NULL, *res=NULL;
675 struct dlm_master_list_entry *mle = NULL;
676 struct dlm_master_list_entry *alloc_mle = NULL;
677 int blocked = 0;
678 int ret, nodenum;
679 struct dlm_node_iter iter;
680 unsigned int namelen;
681 int tries = 0;
682
683 BUG_ON(!lockid);
684
685 namelen = strlen(lockid);
686
687 mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
688
689lookup:
690 spin_lock(&dlm->spinlock);
691 tmpres = __dlm_lookup_lockres(dlm, lockid, namelen);
692 if (tmpres) {
693 spin_unlock(&dlm->spinlock);
694 mlog(0, "found in hash!\n");
695 if (res)
696 dlm_lockres_put(res);
697 res = tmpres;
698 goto leave;
699 }
700
701 if (!res) {
702 spin_unlock(&dlm->spinlock);
703 mlog(0, "allocating a new resource\n");
704 /* nothing found and we need to allocate one. */
705 alloc_mle = (struct dlm_master_list_entry *)
706 kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
707 if (!alloc_mle)
708 goto leave;
709 res = dlm_new_lockres(dlm, lockid, namelen);
710 if (!res)
711 goto leave;
712 goto lookup;
713 }
714
715 mlog(0, "no lockres found, allocated our own: %p\n", res);
716
717 if (flags & LKM_LOCAL) {
718 /* caller knows it's safe to assume it's not mastered elsewhere
719 * DONE! return right away */
720 spin_lock(&res->spinlock);
721 dlm_change_lockres_owner(dlm, res, dlm->node_num);
722 __dlm_insert_lockres(dlm, res);
723 spin_unlock(&res->spinlock);
724 spin_unlock(&dlm->spinlock);
725 /* lockres still marked IN_PROGRESS */
726 goto wake_waiters;
727 }
728
729 /* check master list to see if another node has started mastering it */
730 spin_lock(&dlm->master_lock);
731
732 /* if we found a block, wait for lock to be mastered by another node */
733 blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen);
734 if (blocked) {
735 if (mle->type == DLM_MLE_MASTER) {
736 mlog(ML_ERROR, "master entry for nonexistent lock!\n");
737 BUG();
738 } else if (mle->type == DLM_MLE_MIGRATION) {
739 /* migration is in progress! */
740 /* the good news is that we now know the
741 * "current" master (mle->master). */
742
743 spin_unlock(&dlm->master_lock);
744 assert_spin_locked(&dlm->spinlock);
745
746 /* set the lockres owner and hash it */
747 spin_lock(&res->spinlock);
748 dlm_set_lockres_owner(dlm, res, mle->master);
749 __dlm_insert_lockres(dlm, res);
750 spin_unlock(&res->spinlock);
751 spin_unlock(&dlm->spinlock);
752
753 /* master is known, detach */
754 dlm_mle_detach_hb_events(dlm, mle);
755 dlm_put_mle(mle);
756 mle = NULL;
757 goto wake_waiters;
758 }
759 } else {
760 /* go ahead and try to master lock on this node */
761 mle = alloc_mle;
762 /* make sure this does not get freed below */
763 alloc_mle = NULL;
764 dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
765 set_bit(dlm->node_num, mle->maybe_map);
766 list_add(&mle->list, &dlm->master_list);
767 }
768
769 /* at this point there is either a DLM_MLE_BLOCK or a
770 * DLM_MLE_MASTER on the master list, so it's safe to add the
771 * lockres to the hashtable. anyone who finds the lock will
772 * still have to wait on the IN_PROGRESS. */
773
774 /* finally add the lockres to its hash bucket */
775 __dlm_insert_lockres(dlm, res);
776 /* get an extra ref on the mle in case this is a BLOCK
777 * if so, the creator of the BLOCK may try to put the last
778 * ref at this time in the assert master handler, so we
779 * need an extra one to keep from a bad ptr deref. */
780 dlm_get_mle(mle);
781 spin_unlock(&dlm->master_lock);
782 spin_unlock(&dlm->spinlock);
783
784 /* must wait for lock to be mastered elsewhere */
785 if (blocked)
786 goto wait;
787
788redo_request:
789 ret = -EINVAL;
790 dlm_node_iter_init(mle->vote_map, &iter);
791 while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
792 ret = dlm_do_master_request(mle, nodenum);
793 if (ret < 0)
794 mlog_errno(ret);
795 if (mle->master != O2NM_MAX_NODES) {
796 /* found a master ! */
797 break;
798 }
799 }
800
801wait:
802 /* keep going until the response map includes all nodes */
803 ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
804 if (ret < 0) {
805 mlog(0, "%s:%.*s: node map changed, redo the "
806 "master request now, blocked=%d\n",
807 dlm->name, res->lockname.len,
808 res->lockname.name, blocked);
809 if (++tries > 20) {
810 mlog(ML_ERROR, "%s:%.*s: spinning on "
811 "dlm_wait_for_lock_mastery, blocked=%d\n",
812 dlm->name, res->lockname.len,
813 res->lockname.name, blocked);
814 dlm_print_one_lock_resource(res);
815 /* dlm_print_one_mle(mle); */
816 tries = 0;
817 }
818 goto redo_request;
819 }
820
821 mlog(0, "lockres mastered by %u\n", res->owner);
822 /* make sure we never continue without this */
823 BUG_ON(res->owner == O2NM_MAX_NODES);
824
825 /* master is known, detach if not already detached */
826 dlm_mle_detach_hb_events(dlm, mle);
827 dlm_put_mle(mle);
828 /* put the extra ref */
829 dlm_put_mle(mle);
830
831wake_waiters:
832 spin_lock(&res->spinlock);
833 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
834 spin_unlock(&res->spinlock);
835 wake_up(&res->wq);
836
837leave:
838 /* need to free the unused mle */
839 if (alloc_mle)
840 kmem_cache_free(dlm_mle_cache, alloc_mle);
841
842 return res;
843}
844
845
846#define DLM_MASTERY_TIMEOUT_MS 5000
847
848static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
849 struct dlm_lock_resource *res,
850 struct dlm_master_list_entry *mle,
851 int *blocked)
852{
853 u8 m;
854 int ret, bit;
855 int map_changed, voting_done;
856 int assert, sleep;
857
858recheck:
859 ret = 0;
860 assert = 0;
861
862 /* check if another node has already become the owner */
863 spin_lock(&res->spinlock);
864 if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
865 spin_unlock(&res->spinlock);
866 goto leave;
867 }
868 spin_unlock(&res->spinlock);
869
870 spin_lock(&mle->spinlock);
871 m = mle->master;
872 map_changed = (memcmp(mle->vote_map, mle->node_map,
873 sizeof(mle->vote_map)) != 0);
874 voting_done = (memcmp(mle->vote_map, mle->response_map,
875 sizeof(mle->vote_map)) == 0);
876
877 /* restart if we hit any errors */
878 if (map_changed) {
879 int b;
880 mlog(0, "%s: %.*s: node map changed, restarting\n",
881 dlm->name, res->lockname.len, res->lockname.name);
882 ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked);
883 b = (mle->type == DLM_MLE_BLOCK);
884 if ((*blocked && !b) || (!*blocked && b)) {
885 mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
886 dlm->name, res->lockname.len, res->lockname.name,
887 *blocked, b);
888 *blocked = b;
889 }
890 spin_unlock(&mle->spinlock);
891 if (ret < 0) {
892 mlog_errno(ret);
893 goto leave;
894 }
895 mlog(0, "%s:%.*s: restart lock mastery succeeded, "
896 "rechecking now\n", dlm->name, res->lockname.len,
897 res->lockname.name);
898 goto recheck;
899 }
900
901 if (m != O2NM_MAX_NODES) {
902 /* another node has done an assert!
903 * all done! */
904 sleep = 0;
905 } else {
906 sleep = 1;
907 /* have all nodes responded? */
908 if (voting_done && !*blocked) {
909 bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
910 if (dlm->node_num <= bit) {
911 /* my node number is lowest.
912 * now tell other nodes that I am
913 * mastering this. */
914 mle->master = dlm->node_num;
915 assert = 1;
916 sleep = 0;
917 }
918 /* if voting is done, but we have not received
919 * an assert master yet, we must sleep */
920 }
921 }
922
923 spin_unlock(&mle->spinlock);
924
925 /* sleep if we haven't finished voting yet */
926 if (sleep) {
927 unsigned long timeo = msecs_to_jiffies(DLM_MASTERY_TIMEOUT_MS);
928
929 /*
930 if (atomic_read(&mle->mle_refs.refcount) < 2)
931 mlog(ML_ERROR, "mle (%p) refs=%d, name=%.*s\n", mle,
932 atomic_read(&mle->mle_refs.refcount),
933 res->lockname.len, res->lockname.name);
934 */
935 atomic_set(&mle->woken, 0);
936 (void)wait_event_timeout(mle->wq,
937 (atomic_read(&mle->woken) == 1),
938 timeo);
939 if (res->owner == O2NM_MAX_NODES) {
940 mlog(0, "waiting again\n");
941 goto recheck;
942 }
943 mlog(0, "done waiting, master is %u\n", res->owner);
944 ret = 0;
945 goto leave;
946 }
947
948 ret = 0; /* done */
949 if (assert) {
950 m = dlm->node_num;
951 mlog(0, "about to master %.*s here, this=%u\n",
952 res->lockname.len, res->lockname.name, m);
953 ret = dlm_do_assert_master(dlm, res->lockname.name,
954 res->lockname.len, mle->vote_map, 0);
955 if (ret) {
956 /* This is a failure in the network path,
957 * not in the response to the assert_master
958 * (any nonzero response is a BUG on this node).
959 * Most likely a socket just got disconnected
960 * due to node death. */
961 mlog_errno(ret);
962 }
963 /* no longer need to restart lock mastery.
964 * all living nodes have been contacted. */
965 ret = 0;
966 }
967
968 /* set the lockres owner */
969 spin_lock(&res->spinlock);
970 dlm_change_lockres_owner(dlm, res, m);
971 spin_unlock(&res->spinlock);
972
973leave:
974 return ret;
975}
976
977struct dlm_bitmap_diff_iter
978{
979 int curnode;
980 unsigned long *orig_bm;
981 unsigned long *cur_bm;
982 unsigned long diff_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
983};
984
985enum dlm_node_state_change
986{
987 NODE_DOWN = -1,
988 NODE_NO_CHANGE = 0,
989 NODE_UP
990};
991
992static void dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter *iter,
993 unsigned long *orig_bm,
994 unsigned long *cur_bm)
995{
996 unsigned long p1, p2;
997 int i;
998
999 iter->curnode = -1;
1000 iter->orig_bm = orig_bm;
1001 iter->cur_bm = cur_bm;
1002
1003 for (i = 0; i < BITS_TO_LONGS(O2NM_MAX_NODES); i++) {
1004 p1 = *(iter->orig_bm + i);
1005 p2 = *(iter->cur_bm + i);
1006 iter->diff_bm[i] = (p1 & ~p2) | (p2 & ~p1);
1007 }
1008}
1009
1010static int dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter *iter,
1011 enum dlm_node_state_change *state)
1012{
1013 int bit;
1014
1015 if (iter->curnode >= O2NM_MAX_NODES)
1016 return -ENOENT;
1017
1018 bit = find_next_bit(iter->diff_bm, O2NM_MAX_NODES,
1019 iter->curnode+1);
1020 if (bit >= O2NM_MAX_NODES) {
1021 iter->curnode = O2NM_MAX_NODES;
1022 return -ENOENT;
1023 }
1024
1025 /* if it was there in the original then this node died */
1026 if (test_bit(bit, iter->orig_bm))
1027 *state = NODE_DOWN;
1028 else
1029 *state = NODE_UP;
1030
1031 iter->curnode = bit;
1032 return bit;
1033}
1034
1035
1036static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
1037 struct dlm_lock_resource *res,
1038 struct dlm_master_list_entry *mle,
1039 int blocked)
1040{
1041 struct dlm_bitmap_diff_iter bdi;
1042 enum dlm_node_state_change sc;
1043 int node;
1044 int ret = 0;
1045
1046 mlog(0, "something happened such that the "
1047 "master process may need to be restarted!\n");
1048
1049 assert_spin_locked(&mle->spinlock);
1050
1051 dlm_bitmap_diff_iter_init(&bdi, mle->vote_map, mle->node_map);
1052 node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1053 while (node >= 0) {
1054 if (sc == NODE_UP) {
1055 /* a node came up. easy. might not even need
1056 * to talk to it if its node number is higher
1057 * or if we are already blocked. */
1058 mlog(0, "node up! %d\n", node);
1059 if (blocked)
1060 goto next;
1061
1062 if (node > dlm->node_num) {
1063 mlog(0, "node > this node. skipping.\n");
1064 goto next;
1065 }
1066
1067 /* redo the master request, but only for the new node */
1068 mlog(0, "sending request to new node\n");
1069 clear_bit(node, mle->response_map);
1070 set_bit(node, mle->vote_map);
1071 } else {
1072 mlog(ML_ERROR, "node down! %d\n", node);
1073
1074 /* if the node wasn't involved in mastery skip it,
1075 * but clear it out from the maps so that it will
1076 * not affect mastery of this lockres */
1077 clear_bit(node, mle->response_map);
1078 clear_bit(node, mle->vote_map);
1079 if (!test_bit(node, mle->maybe_map))
1080 goto next;
1081
1082 /* if we're already blocked on lock mastery, and the
1083 * dead node wasn't the expected master, or there is
1084 * another node in the maybe_map, keep waiting */
1085 if (blocked) {
1086 int lowest = find_next_bit(mle->maybe_map,
1087 O2NM_MAX_NODES, 0);
1088
1089 /* act like it was never there */
1090 clear_bit(node, mle->maybe_map);
1091
1092 if (node != lowest)
1093 goto next;
1094
1095 mlog(ML_ERROR, "expected master %u died while "
1096 "this node was blocked waiting on it!\n",
1097 node);
1098 lowest = find_next_bit(mle->maybe_map,
1099 O2NM_MAX_NODES,
1100 lowest+1);
1101 if (lowest < O2NM_MAX_NODES) {
1102 mlog(0, "still blocked. waiting "
1103 "on %u now\n", lowest);
1104 goto next;
1105 }
1106
1107 /* mle is an MLE_BLOCK, but there is now
1108 * nothing left to block on. we need to return
1109 * all the way back out and try again with
1110 * an MLE_MASTER. dlm_do_local_recovery_cleanup
1111 * has already run, so the mle refcount is ok */
1112 mlog(0, "no longer blocking. we can "
1113 "try to master this here\n");
1114 mle->type = DLM_MLE_MASTER;
1115 memset(mle->maybe_map, 0,
1116 sizeof(mle->maybe_map));
1117 memset(mle->response_map, 0,
1118 sizeof(mle->maybe_map));
1119 memcpy(mle->vote_map, mle->node_map,
1120 sizeof(mle->node_map));
1121 mle->u.res = res;
1122 set_bit(dlm->node_num, mle->maybe_map);
1123
1124 ret = -EAGAIN;
1125 goto next;
1126 }
1127
1128 clear_bit(node, mle->maybe_map);
1129 if (node > dlm->node_num)
1130 goto next;
1131
1132 mlog(0, "dead node in map!\n");
1133 /* yuck. go back and re-contact all nodes
1134 * in the vote_map, removing this node. */
1135 memset(mle->response_map, 0,
1136 sizeof(mle->response_map));
1137 }
1138 ret = -EAGAIN;
1139next:
1140 node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1141 }
1142 return ret;
1143}
1144
1145
1146/*
1147 * DLM_MASTER_REQUEST_MSG
1148 *
1149 * returns: 0 on success,
1150 * -errno on a network error
1151 *
1152 * on error, the caller should assume the target node is "dead"
1153 *
1154 */
1155
1156static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to)
1157{
1158 struct dlm_ctxt *dlm = mle->dlm;
1159 struct dlm_master_request request;
1160 int ret, response=0, resend;
1161
1162 memset(&request, 0, sizeof(request));
1163 request.node_idx = dlm->node_num;
1164
1165 BUG_ON(mle->type == DLM_MLE_MIGRATION);
1166
1167 if (mle->type != DLM_MLE_MASTER) {
1168 request.namelen = mle->u.name.len;
1169 memcpy(request.name, mle->u.name.name, request.namelen);
1170 } else {
1171 request.namelen = mle->u.res->lockname.len;
1172 memcpy(request.name, mle->u.res->lockname.name,
1173 request.namelen);
1174 }
1175
1176again:
1177 ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request,
1178 sizeof(request), to, &response);
1179 if (ret < 0) {
1180 if (ret == -ESRCH) {
1181 /* should never happen */
1182 mlog(ML_ERROR, "TCP stack not ready!\n");
1183 BUG();
1184 } else if (ret == -EINVAL) {
1185 mlog(ML_ERROR, "bad args passed to o2net!\n");
1186 BUG();
1187 } else if (ret == -ENOMEM) {
1188 mlog(ML_ERROR, "out of memory while trying to send "
1189 "network message! retrying\n");
1190 /* this is totally crude */
1191 msleep(50);
1192 goto again;
1193 } else if (!dlm_is_host_down(ret)) {
1194 /* not a network error. bad. */
1195 mlog_errno(ret);
1196 mlog(ML_ERROR, "unhandled error!");
1197 BUG();
1198 }
1199 /* all other errors should be network errors,
1200 * and likely indicate node death */
1201 mlog(ML_ERROR, "link to %d went down!\n", to);
1202 goto out;
1203 }
1204
1205 ret = 0;
1206 resend = 0;
1207 spin_lock(&mle->spinlock);
1208 switch (response) {
1209 case DLM_MASTER_RESP_YES:
1210 set_bit(to, mle->response_map);
1211 mlog(0, "node %u is the master, response=YES\n", to);
1212 mle->master = to;
1213 break;
1214 case DLM_MASTER_RESP_NO:
1215 mlog(0, "node %u not master, response=NO\n", to);
1216 set_bit(to, mle->response_map);
1217 break;
1218 case DLM_MASTER_RESP_MAYBE:
1219 mlog(0, "node %u not master, response=MAYBE\n", to);
1220 set_bit(to, mle->response_map);
1221 set_bit(to, mle->maybe_map);
1222 break;
1223 case DLM_MASTER_RESP_ERROR:
1224 mlog(0, "node %u hit an error, resending\n", to);
1225 resend = 1;
1226 response = 0;
1227 break;
1228 default:
1229 mlog(ML_ERROR, "bad response! %u\n", response);
1230 BUG();
1231 }
1232 spin_unlock(&mle->spinlock);
1233 if (resend) {
1234 /* this is also totally crude */
1235 msleep(50);
1236 goto again;
1237 }
1238
1239out:
1240 return ret;
1241}
1242
1243/*
1244 * locks that can be taken here:
1245 * dlm->spinlock
1246 * res->spinlock
1247 * mle->spinlock
1248 * dlm->master_list
1249 *
1250 * if possible, TRIM THIS DOWN!!!
1251 */
1252int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data)
1253{
1254 u8 response = DLM_MASTER_RESP_MAYBE;
1255 struct dlm_ctxt *dlm = data;
1256 struct dlm_lock_resource *res;
1257 struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
1258 struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
1259 char *name;
1260 unsigned int namelen;
1261 int found, ret;
1262 int set_maybe;
1263
1264 if (!dlm_grab(dlm))
1265 return DLM_MASTER_RESP_NO;
1266
1267 if (!dlm_domain_fully_joined(dlm)) {
1268 response = DLM_MASTER_RESP_NO;
1269 goto send_response;
1270 }
1271
1272 name = request->name;
1273 namelen = request->namelen;
1274
1275 if (namelen > DLM_LOCKID_NAME_MAX) {
1276 response = DLM_IVBUFLEN;
1277 goto send_response;
1278 }
1279
1280way_up_top:
1281 spin_lock(&dlm->spinlock);
1282 res = __dlm_lookup_lockres(dlm, name, namelen);
1283 if (res) {
1284 spin_unlock(&dlm->spinlock);
1285
1286 /* take care of the easy cases up front */
1287 spin_lock(&res->spinlock);
1288 if (res->state & DLM_LOCK_RES_RECOVERING) {
1289 spin_unlock(&res->spinlock);
1290 mlog(0, "returning DLM_MASTER_RESP_ERROR since res is "
1291 "being recovered\n");
1292 response = DLM_MASTER_RESP_ERROR;
1293 if (mle)
1294 kmem_cache_free(dlm_mle_cache, mle);
1295 goto send_response;
1296 }
1297
1298 if (res->owner == dlm->node_num) {
1299 u32 flags = DLM_ASSERT_MASTER_MLE_CLEANUP;
1300 spin_unlock(&res->spinlock);
1301 // mlog(0, "this node is the master\n");
1302 response = DLM_MASTER_RESP_YES;
1303 if (mle)
1304 kmem_cache_free(dlm_mle_cache, mle);
1305
1306 /* this node is the owner.
1307 * there is some extra work that needs to
1308 * happen now. the requesting node has
1309 * caused all nodes up to this one to
1310 * create mles. this node now needs to
1311 * go back and clean those up. */
1312 mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
1313 dlm->node_num, res->lockname.len, res->lockname.name);
1314 ret = dlm_dispatch_assert_master(dlm, res, 1,
1315 request->node_idx,
1316 flags);
1317 if (ret < 0) {
1318 mlog(ML_ERROR, "failed to dispatch assert "
1319 "master work\n");
1320 response = DLM_MASTER_RESP_ERROR;
1321 }
1322 goto send_response;
1323 } else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1324 spin_unlock(&res->spinlock);
1325 // mlog(0, "node %u is the master\n", res->owner);
1326 response = DLM_MASTER_RESP_NO;
1327 if (mle)
1328 kmem_cache_free(dlm_mle_cache, mle);
1329 goto send_response;
1330 }
1331
1332 /* ok, there is no owner. either this node is
1333 * being blocked, or it is actively trying to
1334 * master this lock. */
1335 if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1336 mlog(ML_ERROR, "lock with no owner should be "
1337 "in-progress!\n");
1338 BUG();
1339 }
1340
1341 // mlog(0, "lockres is in progress...\n");
1342 spin_lock(&dlm->master_lock);
1343 found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1344 if (!found) {
1345 mlog(ML_ERROR, "no mle found for this lock!\n");
1346 BUG();
1347 }
1348 set_maybe = 1;
1349 spin_lock(&tmpmle->spinlock);
1350 if (tmpmle->type == DLM_MLE_BLOCK) {
1351 // mlog(0, "this node is waiting for "
1352 // "lockres to be mastered\n");
1353 response = DLM_MASTER_RESP_NO;
1354 } else if (tmpmle->type == DLM_MLE_MIGRATION) {
1355 mlog(0, "node %u is master, but trying to migrate to "
1356 "node %u.\n", tmpmle->master, tmpmle->new_master);
1357 if (tmpmle->master == dlm->node_num) {
1358 response = DLM_MASTER_RESP_YES;
1359 mlog(ML_ERROR, "no owner on lockres, but this "
1360 "node is trying to migrate it to %u?!\n",
1361 tmpmle->new_master);
1362 BUG();
1363 } else {
1364 /* the real master can respond on its own */
1365 response = DLM_MASTER_RESP_NO;
1366 }
1367 } else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) {
1368 set_maybe = 0;
1369 if (tmpmle->master == dlm->node_num)
1370 response = DLM_MASTER_RESP_YES;
1371 else
1372 response = DLM_MASTER_RESP_NO;
1373 } else {
1374 // mlog(0, "this node is attempting to "
1375 // "master lockres\n");
1376 response = DLM_MASTER_RESP_MAYBE;
1377 }
1378 if (set_maybe)
1379 set_bit(request->node_idx, tmpmle->maybe_map);
1380 spin_unlock(&tmpmle->spinlock);
1381
1382 spin_unlock(&dlm->master_lock);
1383 spin_unlock(&res->spinlock);
1384
1385 /* keep the mle attached to heartbeat events */
1386 dlm_put_mle(tmpmle);
1387 if (mle)
1388 kmem_cache_free(dlm_mle_cache, mle);
1389 goto send_response;
1390 }
1391
1392 /*
1393 * lockres doesn't exist on this node
1394 * if there is an MLE_BLOCK, return NO
1395 * if there is an MLE_MASTER, return MAYBE
1396 * otherwise, add an MLE_BLOCK, return NO
1397 */
1398 spin_lock(&dlm->master_lock);
1399 found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1400 if (!found) {
1401 /* this lockid has never been seen on this node yet */
1402 // mlog(0, "no mle found\n");
1403 if (!mle) {
1404 spin_unlock(&dlm->master_lock);
1405 spin_unlock(&dlm->spinlock);
1406
1407 mle = (struct dlm_master_list_entry *)
1408 kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL);
1409 if (!mle) {
1410 // bad bad bad... this sucks.
1411 response = DLM_MASTER_RESP_ERROR;
1412 goto send_response;
1413 }
1414 spin_lock(&dlm->spinlock);
1415 dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL,
1416 name, namelen);
1417 spin_unlock(&dlm->spinlock);
1418 goto way_up_top;
1419 }
1420
1421 // mlog(0, "this is second time thru, already allocated, "
1422 // "add the block.\n");
1423 set_bit(request->node_idx, mle->maybe_map);
1424 list_add(&mle->list, &dlm->master_list);
1425 response = DLM_MASTER_RESP_NO;
1426 } else {
1427 // mlog(0, "mle was found\n");
1428 set_maybe = 1;
1429 spin_lock(&tmpmle->spinlock);
1430 if (tmpmle->type == DLM_MLE_BLOCK)
1431 response = DLM_MASTER_RESP_NO;
1432 else if (tmpmle->type == DLM_MLE_MIGRATION) {
1433 mlog(0, "migration mle was found (%u->%u)\n",
1434 tmpmle->master, tmpmle->new_master);
1435 if (tmpmle->master == dlm->node_num) {
1436 mlog(ML_ERROR, "no lockres, but migration mle "
1437 "says that this node is master!\n");
1438 BUG();
1439 }
1440 /* real master can respond on its own */
1441 response = DLM_MASTER_RESP_NO;
1442 } else {
1443 if (tmpmle->master == dlm->node_num) {
1444 response = DLM_MASTER_RESP_YES;
1445 set_maybe = 0;
1446 } else
1447 response = DLM_MASTER_RESP_MAYBE;
1448 }
1449 if (set_maybe)
1450 set_bit(request->node_idx, tmpmle->maybe_map);
1451 spin_unlock(&tmpmle->spinlock);
1452 }
1453 spin_unlock(&dlm->master_lock);
1454 spin_unlock(&dlm->spinlock);
1455
1456 if (found) {
1457 /* keep the mle attached to heartbeat events */
1458 dlm_put_mle(tmpmle);
1459 }
1460send_response:
1461 dlm_put(dlm);
1462 return response;
1463}
1464
1465/*
1466 * DLM_ASSERT_MASTER_MSG
1467 */
1468
1469
1470/*
1471 * NOTE: this can be used for debugging
1472 * can periodically run all locks owned by this node
1473 * and re-assert across the cluster...
1474 */
1475static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname,
1476 unsigned int namelen, void *nodemap,
1477 u32 flags)
1478{
1479 struct dlm_assert_master assert;
1480 int to, tmpret;
1481 struct dlm_node_iter iter;
1482 int ret = 0;
1483
1484 BUG_ON(namelen > O2NM_MAX_NAME_LEN);
1485
1486 /* note that if this nodemap is empty, it returns 0 */
1487 dlm_node_iter_init(nodemap, &iter);
1488 while ((to = dlm_node_iter_next(&iter)) >= 0) {
1489 int r = 0;
1490 mlog(0, "sending assert master to %d (%.*s)\n", to,
1491 namelen, lockname);
1492 memset(&assert, 0, sizeof(assert));
1493 assert.node_idx = dlm->node_num;
1494 assert.namelen = namelen;
1495 memcpy(assert.name, lockname, namelen);
1496 assert.flags = cpu_to_be32(flags);
1497
1498 tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
1499 &assert, sizeof(assert), to, &r);
1500 if (tmpret < 0) {
1501 mlog(ML_ERROR, "assert_master returned %d!\n", tmpret);
1502 if (!dlm_is_host_down(tmpret)) {
1503 mlog(ML_ERROR, "unhandled error!\n");
1504 BUG();
1505 }
1506 /* a node died. finish out the rest of the nodes. */
1507 mlog(ML_ERROR, "link to %d went down!\n", to);
1508 /* any nonzero status return will do */
1509 ret = tmpret;
1510 } else if (r < 0) {
1511 /* ok, something horribly messed. kill thyself. */
1512 mlog(ML_ERROR,"during assert master of %.*s to %u, "
1513 "got %d.\n", namelen, lockname, to, r);
1514 dlm_dump_lock_resources(dlm);
1515 BUG();
1516 }
1517 }
1518
1519 return ret;
1520}
1521
1522/*
1523 * locks that can be taken here:
1524 * dlm->spinlock
1525 * res->spinlock
1526 * mle->spinlock
1527 * dlm->master_list
1528 *
1529 * if possible, TRIM THIS DOWN!!!
1530 */
1531int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data)
1532{
1533 struct dlm_ctxt *dlm = data;
1534 struct dlm_master_list_entry *mle = NULL;
1535 struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
1536 struct dlm_lock_resource *res = NULL;
1537 char *name;
1538 unsigned int namelen;
1539 u32 flags;
1540
1541 if (!dlm_grab(dlm))
1542 return 0;
1543
1544 name = assert->name;
1545 namelen = assert->namelen;
1546 flags = be32_to_cpu(assert->flags);
1547
1548 if (namelen > DLM_LOCKID_NAME_MAX) {
1549 mlog(ML_ERROR, "Invalid name length!");
1550 goto done;
1551 }
1552
1553 spin_lock(&dlm->spinlock);
1554
1555 if (flags)
1556 mlog(0, "assert_master with flags: %u\n", flags);
1557
1558 /* find the MLE */
1559 spin_lock(&dlm->master_lock);
1560 if (!dlm_find_mle(dlm, &mle, name, namelen)) {
1561 /* not an error, could be master just re-asserting */
1562 mlog(0, "just got an assert_master from %u, but no "
1563 "MLE for it! (%.*s)\n", assert->node_idx,
1564 namelen, name);
1565 } else {
1566 int bit = find_next_bit (mle->maybe_map, O2NM_MAX_NODES, 0);
1567 if (bit >= O2NM_MAX_NODES) {
1568 /* not necessarily an error, though less likely.
1569 * could be master just re-asserting. */
1570 mlog(ML_ERROR, "no bits set in the maybe_map, but %u "
1571 "is asserting! (%.*s)\n", assert->node_idx,
1572 namelen, name);
1573 } else if (bit != assert->node_idx) {
1574 if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1575 mlog(0, "master %u was found, %u should "
1576 "back off\n", assert->node_idx, bit);
1577 } else {
1578 /* with the fix for bug 569, a higher node
1579 * number winning the mastery will respond
1580 * YES to mastery requests, but this node
1581 * had no way of knowing. let it pass. */
1582 mlog(ML_ERROR, "%u is the lowest node, "
1583 "%u is asserting. (%.*s) %u must "
1584 "have begun after %u won.\n", bit,
1585 assert->node_idx, namelen, name, bit,
1586 assert->node_idx);
1587 }
1588 }
1589 }
1590 spin_unlock(&dlm->master_lock);
1591
1592 /* ok everything checks out with the MLE
1593 * now check to see if there is a lockres */
1594 res = __dlm_lookup_lockres(dlm, name, namelen);
1595 if (res) {
1596 spin_lock(&res->spinlock);
1597 if (res->state & DLM_LOCK_RES_RECOVERING) {
1598 mlog(ML_ERROR, "%u asserting but %.*s is "
1599 "RECOVERING!\n", assert->node_idx, namelen, name);
1600 goto kill;
1601 }
1602 if (!mle) {
1603 if (res->owner != assert->node_idx) {
1604 mlog(ML_ERROR, "assert_master from "
1605 "%u, but current owner is "
1606 "%u! (%.*s)\n",
1607 assert->node_idx, res->owner,
1608 namelen, name);
1609 goto kill;
1610 }
1611 } else if (mle->type != DLM_MLE_MIGRATION) {
1612 if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1613 /* owner is just re-asserting */
1614 if (res->owner == assert->node_idx) {
1615 mlog(0, "owner %u re-asserting on "
1616 "lock %.*s\n", assert->node_idx,
1617 namelen, name);
1618 goto ok;
1619 }
1620 mlog(ML_ERROR, "got assert_master from "
1621 "node %u, but %u is the owner! "
1622 "(%.*s)\n", assert->node_idx,
1623 res->owner, namelen, name);
1624 goto kill;
1625 }
1626 if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1627 mlog(ML_ERROR, "got assert from %u, but lock "
1628 "with no owner should be "
1629 "in-progress! (%.*s)\n",
1630 assert->node_idx,
1631 namelen, name);
1632 goto kill;
1633 }
1634 } else /* mle->type == DLM_MLE_MIGRATION */ {
1635 /* should only be getting an assert from new master */
1636 if (assert->node_idx != mle->new_master) {
1637 mlog(ML_ERROR, "got assert from %u, but "
1638 "new master is %u, and old master "
1639 "was %u (%.*s)\n",
1640 assert->node_idx, mle->new_master,
1641 mle->master, namelen, name);
1642 goto kill;
1643 }
1644
1645 }
1646ok:
1647 spin_unlock(&res->spinlock);
1648 }
1649 spin_unlock(&dlm->spinlock);
1650
1651 // mlog(0, "woo! got an assert_master from node %u!\n",
1652 // assert->node_idx);
1653 if (mle) {
1654 int extra_ref;
1655
1656 spin_lock(&mle->spinlock);
1657 extra_ref = !!(mle->type == DLM_MLE_BLOCK
1658 || mle->type == DLM_MLE_MIGRATION);
1659 mle->master = assert->node_idx;
1660 atomic_set(&mle->woken, 1);
1661 wake_up(&mle->wq);
1662 spin_unlock(&mle->spinlock);
1663
1664 if (mle->type == DLM_MLE_MIGRATION && res) {
1665 mlog(0, "finishing off migration of lockres %.*s, "
1666 "from %u to %u\n",
1667 res->lockname.len, res->lockname.name,
1668 dlm->node_num, mle->new_master);
1669 spin_lock(&res->spinlock);
1670 res->state &= ~DLM_LOCK_RES_MIGRATING;
1671 dlm_change_lockres_owner(dlm, res, mle->new_master);
1672 BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
1673 spin_unlock(&res->spinlock);
1674 }
1675 /* master is known, detach if not already detached */
1676 dlm_mle_detach_hb_events(dlm, mle);
1677 dlm_put_mle(mle);
1678
1679 if (extra_ref) {
1680 /* the assert master message now balances the extra
1681 * ref given by the master / migration request message.
1682 * if this is the last put, it will be removed
1683 * from the list. */
1684 dlm_put_mle(mle);
1685 }
1686 }
1687
1688done:
1689 if (res)
1690 dlm_lockres_put(res);
1691 dlm_put(dlm);
1692 return 0;
1693
1694kill:
1695 /* kill the caller! */
1696 spin_unlock(&res->spinlock);
1697 spin_unlock(&dlm->spinlock);
1698 dlm_lockres_put(res);
1699 mlog(ML_ERROR, "Bad message received from another node. Dumping state "
1700 "and killing the other node now! This node is OK and can continue.\n");
1701 dlm_dump_lock_resources(dlm);
1702 dlm_put(dlm);
1703 return -EINVAL;
1704}
1705
1706int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
1707 struct dlm_lock_resource *res,
1708 int ignore_higher, u8 request_from, u32 flags)
1709{
1710 struct dlm_work_item *item;
1711 item = kcalloc(1, sizeof(*item), GFP_KERNEL);
1712 if (!item)
1713 return -ENOMEM;
1714
1715
1716 /* queue up work for dlm_assert_master_worker */
1717 dlm_grab(dlm); /* get an extra ref for the work item */
1718 dlm_init_work_item(dlm, item, dlm_assert_master_worker, NULL);
1719 item->u.am.lockres = res; /* already have a ref */
1720 /* can optionally ignore node numbers higher than this node */
1721 item->u.am.ignore_higher = ignore_higher;
1722 item->u.am.request_from = request_from;
1723 item->u.am.flags = flags;
1724
1725 spin_lock(&dlm->work_lock);
1726 list_add_tail(&item->list, &dlm->work_list);
1727 spin_unlock(&dlm->work_lock);
1728
1729 schedule_work(&dlm->dispatched_work);
1730 return 0;
1731}
1732
1733static void dlm_assert_master_worker(struct dlm_work_item *item, void *data)
1734{
1735 struct dlm_ctxt *dlm = data;
1736 int ret = 0;
1737 struct dlm_lock_resource *res;
1738 unsigned long nodemap[BITS_TO_LONGS(O2NM_MAX_NODES)];
1739 int ignore_higher;
1740 int bit;
1741 u8 request_from;
1742 u32 flags;
1743
1744 dlm = item->dlm;
1745 res = item->u.am.lockres;
1746 ignore_higher = item->u.am.ignore_higher;
1747 request_from = item->u.am.request_from;
1748 flags = item->u.am.flags;
1749
1750 spin_lock(&dlm->spinlock);
1751 memcpy(nodemap, dlm->domain_map, sizeof(nodemap));
1752 spin_unlock(&dlm->spinlock);
1753
1754 clear_bit(dlm->node_num, nodemap);
1755 if (ignore_higher) {
1756 /* if is this just to clear up mles for nodes below
1757 * this node, do not send the message to the original
1758 * caller or any node number higher than this */
1759 clear_bit(request_from, nodemap);
1760 bit = dlm->node_num;
1761 while (1) {
1762 bit = find_next_bit(nodemap, O2NM_MAX_NODES,
1763 bit+1);
1764 if (bit >= O2NM_MAX_NODES)
1765 break;
1766 clear_bit(bit, nodemap);
1767 }
1768 }
1769
1770 /* this call now finishes out the nodemap
1771 * even if one or more nodes die */
1772 mlog(0, "worker about to master %.*s here, this=%u\n",
1773 res->lockname.len, res->lockname.name, dlm->node_num);
1774 ret = dlm_do_assert_master(dlm, res->lockname.name,
1775 res->lockname.len,
1776 nodemap, flags);
1777 if (ret < 0) {
1778 /* no need to restart, we are done */
1779 mlog_errno(ret);
1780 }
1781
1782 dlm_lockres_put(res);
1783
1784 mlog(0, "finished with dlm_assert_master_worker\n");
1785}
1786
1787
1788/*
1789 * DLM_MIGRATE_LOCKRES
1790 */
1791
1792
1793int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
1794 u8 target)
1795{
1796 struct dlm_master_list_entry *mle = NULL;
1797 struct dlm_master_list_entry *oldmle = NULL;
1798 struct dlm_migratable_lockres *mres = NULL;
1799 int ret = -EINVAL;
1800 const char *name;
1801 unsigned int namelen;
1802 int mle_added = 0;
1803 struct list_head *queue, *iter;
1804 int i;
1805 struct dlm_lock *lock;
1806 int empty = 1;
1807
1808 if (!dlm_grab(dlm))
1809 return -EINVAL;
1810
1811 name = res->lockname.name;
1812 namelen = res->lockname.len;
1813
1814 mlog(0, "migrating %.*s to %u\n", namelen, name, target);
1815
1816 /*
1817 * ensure this lockres is a proper candidate for migration
1818 */
1819 spin_lock(&res->spinlock);
1820 if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
1821 mlog(0, "cannot migrate lockres with unknown owner!\n");
1822 spin_unlock(&res->spinlock);
1823 goto leave;
1824 }
1825 if (res->owner != dlm->node_num) {
1826 mlog(0, "cannot migrate lockres this node doesn't own!\n");
1827 spin_unlock(&res->spinlock);
1828 goto leave;
1829 }
1830 mlog(0, "checking queues...\n");
1831 queue = &res->granted;
1832 for (i=0; i<3; i++) {
1833 list_for_each(iter, queue) {
1834 lock = list_entry (iter, struct dlm_lock, list);
1835 empty = 0;
1836 if (lock->ml.node == dlm->node_num) {
1837 mlog(0, "found a lock owned by this node "
1838 "still on the %s queue! will not "
1839 "migrate this lockres\n",
1840 i==0 ? "granted" :
1841 (i==1 ? "converting" : "blocked"));
1842 spin_unlock(&res->spinlock);
1843 ret = -ENOTEMPTY;
1844 goto leave;
1845 }
1846 }
1847 queue++;
1848 }
1849 mlog(0, "all locks on this lockres are nonlocal. continuing\n");
1850 spin_unlock(&res->spinlock);
1851
1852 /* no work to do */
1853 if (empty) {
1854 mlog(0, "no locks were found on this lockres! done!\n");
1855 ret = 0;
1856 goto leave;
1857 }
1858
1859 /*
1860 * preallocate up front
1861 * if this fails, abort
1862 */
1863
1864 ret = -ENOMEM;
1865 mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_KERNEL);
1866 if (!mres) {
1867 mlog_errno(ret);
1868 goto leave;
1869 }
1870
1871 mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
1872 GFP_KERNEL);
1873 if (!mle) {
1874 mlog_errno(ret);
1875 goto leave;
1876 }
1877 ret = 0;
1878
1879 /*
1880 * find a node to migrate the lockres to
1881 */
1882
1883 mlog(0, "picking a migration node\n");
1884 spin_lock(&dlm->spinlock);
1885 /* pick a new node */
1886 if (!test_bit(target, dlm->domain_map) ||
1887 target >= O2NM_MAX_NODES) {
1888 target = dlm_pick_migration_target(dlm, res);
1889 }
1890 mlog(0, "node %u chosen for migration\n", target);
1891
1892 if (target >= O2NM_MAX_NODES ||
1893 !test_bit(target, dlm->domain_map)) {
1894 /* target chosen is not alive */
1895 ret = -EINVAL;
1896 }
1897
1898 if (ret) {
1899 spin_unlock(&dlm->spinlock);
1900 goto fail;
1901 }
1902
1903 mlog(0, "continuing with target = %u\n", target);
1904
1905 /*
1906 * clear any existing master requests and
1907 * add the migration mle to the list
1908 */
1909 spin_lock(&dlm->master_lock);
1910 ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name,
1911 namelen, target, dlm->node_num);
1912 spin_unlock(&dlm->master_lock);
1913 spin_unlock(&dlm->spinlock);
1914
1915 if (ret == -EEXIST) {
1916 mlog(0, "another process is already migrating it\n");
1917 goto fail;
1918 }
1919 mle_added = 1;
1920
1921 /*
1922 * set the MIGRATING flag and flush asts
1923 * if we fail after this we need to re-dirty the lockres
1924 */
1925 if (dlm_mark_lockres_migrating(dlm, res, target) < 0) {
1926 mlog(ML_ERROR, "tried to migrate %.*s to %u, but "
1927 "the target went down.\n", res->lockname.len,
1928 res->lockname.name, target);
1929 spin_lock(&res->spinlock);
1930 res->state &= ~DLM_LOCK_RES_MIGRATING;
1931 spin_unlock(&res->spinlock);
1932 ret = -EINVAL;
1933 }
1934
1935fail:
1936 if (oldmle) {
1937 /* master is known, detach if not already detached */
1938 dlm_mle_detach_hb_events(dlm, oldmle);
1939 dlm_put_mle(oldmle);
1940 }
1941
1942 if (ret < 0) {
1943 if (mle_added) {
1944 dlm_mle_detach_hb_events(dlm, mle);
1945 dlm_put_mle(mle);
1946 } else if (mle) {
1947 kmem_cache_free(dlm_mle_cache, mle);
1948 }
1949 goto leave;
1950 }
1951
1952 /*
1953 * at this point, we have a migration target, an mle
1954 * in the master list, and the MIGRATING flag set on
1955 * the lockres
1956 */
1957
1958
1959 /* get an extra reference on the mle.
1960 * otherwise the assert_master from the new
1961 * master will destroy this.
1962 * also, make sure that all callers of dlm_get_mle
1963 * take both dlm->spinlock and dlm->master_lock */
1964 spin_lock(&dlm->spinlock);
1965 spin_lock(&dlm->master_lock);
1966 dlm_get_mle(mle);
1967 spin_unlock(&dlm->master_lock);
1968 spin_unlock(&dlm->spinlock);
1969
1970 /* notify new node and send all lock state */
1971 /* call send_one_lockres with migration flag.
1972 * this serves as notice to the target node that a
1973 * migration is starting. */
1974 ret = dlm_send_one_lockres(dlm, res, mres, target,
1975 DLM_MRES_MIGRATION);
1976
1977 if (ret < 0) {
1978 mlog(0, "migration to node %u failed with %d\n",
1979 target, ret);
1980 /* migration failed, detach and clean up mle */
1981 dlm_mle_detach_hb_events(dlm, mle);
1982 dlm_put_mle(mle);
1983 dlm_put_mle(mle);
1984 goto leave;
1985 }
1986
1987 /* at this point, the target sends a message to all nodes,
1988 * (using dlm_do_migrate_request). this node is skipped since
1989 * we had to put an mle in the list to begin the process. this
1990 * node now waits for target to do an assert master. this node
1991 * will be the last one notified, ensuring that the migration
1992 * is complete everywhere. if the target dies while this is
1993 * going on, some nodes could potentially see the target as the
1994 * master, so it is important that my recovery finds the migration
1995 * mle and sets the master to UNKNONWN. */
1996
1997
1998 /* wait for new node to assert master */
1999 while (1) {
2000 ret = wait_event_interruptible_timeout(mle->wq,
2001 (atomic_read(&mle->woken) == 1),
2002 msecs_to_jiffies(5000));
2003
2004 if (ret >= 0) {
2005 if (atomic_read(&mle->woken) == 1 ||
2006 res->owner == target)
2007 break;
2008
2009 mlog(0, "timed out during migration\n");
2010 }
2011 if (ret == -ERESTARTSYS) {
2012 /* migration failed, detach and clean up mle */
2013 dlm_mle_detach_hb_events(dlm, mle);
2014 dlm_put_mle(mle);
2015 dlm_put_mle(mle);
2016 goto leave;
2017 }
2018 /* TODO: if node died: stop, clean up, return error */
2019 }
2020
2021 /* all done, set the owner, clear the flag */
2022 spin_lock(&res->spinlock);
2023 dlm_set_lockres_owner(dlm, res, target);
2024 res->state &= ~DLM_LOCK_RES_MIGRATING;
2025 dlm_remove_nonlocal_locks(dlm, res);
2026 spin_unlock(&res->spinlock);
2027 wake_up(&res->wq);
2028
2029 /* master is known, detach if not already detached */
2030 dlm_mle_detach_hb_events(dlm, mle);
2031 dlm_put_mle(mle);
2032 ret = 0;
2033
2034 dlm_lockres_calc_usage(dlm, res);
2035
2036leave:
2037 /* re-dirty the lockres if we failed */
2038 if (ret < 0)
2039 dlm_kick_thread(dlm, res);
2040
2041 /* TODO: cleanup */
2042 if (mres)
2043 free_page((unsigned long)mres);
2044
2045 dlm_put(dlm);
2046
2047 mlog(0, "returning %d\n", ret);
2048 return ret;
2049}
2050EXPORT_SYMBOL_GPL(dlm_migrate_lockres);
2051
2052int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock)
2053{
2054 int ret;
2055 spin_lock(&dlm->ast_lock);
2056 spin_lock(&lock->spinlock);
2057 ret = (list_empty(&lock->bast_list) && !lock->bast_pending);
2058 spin_unlock(&lock->spinlock);
2059 spin_unlock(&dlm->ast_lock);
2060 return ret;
2061}
2062
2063static int dlm_migration_can_proceed(struct dlm_ctxt *dlm,
2064 struct dlm_lock_resource *res,
2065 u8 mig_target)
2066{
2067 int can_proceed;
2068 spin_lock(&res->spinlock);
2069 can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING);
2070 spin_unlock(&res->spinlock);
2071
2072 /* target has died, so make the caller break out of the
2073 * wait_event, but caller must recheck the domain_map */
2074 spin_lock(&dlm->spinlock);
2075 if (!test_bit(mig_target, dlm->domain_map))
2076 can_proceed = 1;
2077 spin_unlock(&dlm->spinlock);
2078 return can_proceed;
2079}
2080
2081int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
2082{
2083 int ret;
2084 spin_lock(&res->spinlock);
2085 ret = !!(res->state & DLM_LOCK_RES_DIRTY);
2086 spin_unlock(&res->spinlock);
2087 return ret;
2088}
2089
2090
2091static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
2092 struct dlm_lock_resource *res,
2093 u8 target)
2094{
2095 int ret = 0;
2096
2097 mlog(0, "dlm_mark_lockres_migrating: %.*s, from %u to %u\n",
2098 res->lockname.len, res->lockname.name, dlm->node_num,
2099 target);
2100 /* need to set MIGRATING flag on lockres. this is done by
2101 * ensuring that all asts have been flushed for this lockres. */
2102 spin_lock(&res->spinlock);
2103 BUG_ON(res->migration_pending);
2104 res->migration_pending = 1;
2105 /* strategy is to reserve an extra ast then release
2106 * it below, letting the release do all of the work */
2107 __dlm_lockres_reserve_ast(res);
2108 spin_unlock(&res->spinlock);
2109
2110 /* now flush all the pending asts.. hang out for a bit */
2111 dlm_kick_thread(dlm, res);
2112 wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
2113 dlm_lockres_release_ast(dlm, res);
2114
2115 mlog(0, "about to wait on migration_wq, dirty=%s\n",
2116 res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no");
2117 /* if the extra ref we just put was the final one, this
2118 * will pass thru immediately. otherwise, we need to wait
2119 * for the last ast to finish. */
2120again:
2121 ret = wait_event_interruptible_timeout(dlm->migration_wq,
2122 dlm_migration_can_proceed(dlm, res, target),
2123 msecs_to_jiffies(1000));
2124 if (ret < 0) {
2125 mlog(0, "woken again: migrating? %s, dead? %s\n",
2126 res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2127 test_bit(target, dlm->domain_map) ? "no":"yes");
2128 } else {
2129 mlog(0, "all is well: migrating? %s, dead? %s\n",
2130 res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2131 test_bit(target, dlm->domain_map) ? "no":"yes");
2132 }
2133 if (!dlm_migration_can_proceed(dlm, res, target)) {
2134 mlog(0, "trying again...\n");
2135 goto again;
2136 }
2137
2138 /* did the target go down or die? */
2139 spin_lock(&dlm->spinlock);
2140 if (!test_bit(target, dlm->domain_map)) {
2141 mlog(ML_ERROR, "aha. migration target %u just went down\n",
2142 target);
2143 ret = -EHOSTDOWN;
2144 }
2145 spin_unlock(&dlm->spinlock);
2146
2147 /*
2148 * at this point:
2149 *
2150 * o the DLM_LOCK_RES_MIGRATING flag is set
2151 * o there are no pending asts on this lockres
2152 * o all processes trying to reserve an ast on this
2153 * lockres must wait for the MIGRATING flag to clear
2154 */
2155 return ret;
2156}
2157
2158/* last step in the migration process.
2159 * original master calls this to free all of the dlm_lock
2160 * structures that used to be for other nodes. */
2161static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
2162 struct dlm_lock_resource *res)
2163{
2164 struct list_head *iter, *iter2;
2165 struct list_head *queue = &res->granted;
2166 int i;
2167 struct dlm_lock *lock;
2168
2169 assert_spin_locked(&res->spinlock);
2170
2171 BUG_ON(res->owner == dlm->node_num);
2172
2173 for (i=0; i<3; i++) {
2174 list_for_each_safe(iter, iter2, queue) {
2175 lock = list_entry (iter, struct dlm_lock, list);
2176 if (lock->ml.node != dlm->node_num) {
2177 mlog(0, "putting lock for node %u\n",
2178 lock->ml.node);
2179 /* be extra careful */
2180 BUG_ON(!list_empty(&lock->ast_list));
2181 BUG_ON(!list_empty(&lock->bast_list));
2182 BUG_ON(lock->ast_pending);
2183 BUG_ON(lock->bast_pending);
2184 list_del_init(&lock->list);
2185 dlm_lock_put(lock);
2186 }
2187 }
2188 queue++;
2189 }
2190}
2191
2192/* for now this is not too intelligent. we will
2193 * need stats to make this do the right thing.
2194 * this just finds the first lock on one of the
2195 * queues and uses that node as the target. */
2196static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
2197 struct dlm_lock_resource *res)
2198{
2199 int i;
2200 struct list_head *queue = &res->granted;
2201 struct list_head *iter;
2202 struct dlm_lock *lock;
2203 int nodenum;
2204
2205 assert_spin_locked(&dlm->spinlock);
2206
2207 spin_lock(&res->spinlock);
2208 for (i=0; i<3; i++) {
2209 list_for_each(iter, queue) {
2210 /* up to the caller to make sure this node
2211 * is alive */
2212 lock = list_entry (iter, struct dlm_lock, list);
2213 if (lock->ml.node != dlm->node_num) {
2214 spin_unlock(&res->spinlock);
2215 return lock->ml.node;
2216 }
2217 }
2218 queue++;
2219 }
2220 spin_unlock(&res->spinlock);
2221 mlog(0, "have not found a suitable target yet! checking domain map\n");
2222
2223 /* ok now we're getting desperate. pick anyone alive. */
2224 nodenum = -1;
2225 while (1) {
2226 nodenum = find_next_bit(dlm->domain_map,
2227 O2NM_MAX_NODES, nodenum+1);
2228 mlog(0, "found %d in domain map\n", nodenum);
2229 if (nodenum >= O2NM_MAX_NODES)
2230 break;
2231 if (nodenum != dlm->node_num) {
2232 mlog(0, "picking %d\n", nodenum);
2233 return nodenum;
2234 }
2235 }
2236
2237 mlog(0, "giving up. no master to migrate to\n");
2238 return DLM_LOCK_RES_OWNER_UNKNOWN;
2239}
2240
2241
2242
2243/* this is called by the new master once all lockres
2244 * data has been received */
2245static int dlm_do_migrate_request(struct dlm_ctxt *dlm,
2246 struct dlm_lock_resource *res,
2247 u8 master, u8 new_master,
2248 struct dlm_node_iter *iter)
2249{
2250 struct dlm_migrate_request migrate;
2251 int ret, status = 0;
2252 int nodenum;
2253
2254 memset(&migrate, 0, sizeof(migrate));
2255 migrate.namelen = res->lockname.len;
2256 memcpy(migrate.name, res->lockname.name, migrate.namelen);
2257 migrate.new_master = new_master;
2258 migrate.master = master;
2259
2260 ret = 0;
2261
2262 /* send message to all nodes, except the master and myself */
2263 while ((nodenum = dlm_node_iter_next(iter)) >= 0) {
2264 if (nodenum == master ||
2265 nodenum == new_master)
2266 continue;
2267
2268 ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
2269 &migrate, sizeof(migrate), nodenum,
2270 &status);
2271 if (ret < 0)
2272 mlog_errno(ret);
2273 else if (status < 0) {
2274 mlog(0, "migrate request (node %u) returned %d!\n",
2275 nodenum, status);
2276 ret = status;
2277 }
2278 }
2279
2280 if (ret < 0)
2281 mlog_errno(ret);
2282
2283 mlog(0, "returning ret=%d\n", ret);
2284 return ret;
2285}
2286
2287
2288/* if there is an existing mle for this lockres, we now know who the master is.
2289 * (the one who sent us *this* message) we can clear it up right away.
2290 * since the process that put the mle on the list still has a reference to it,
2291 * we can unhash it now, set the master and wake the process. as a result,
2292 * we will have no mle in the list to start with. now we can add an mle for
2293 * the migration and this should be the only one found for those scanning the
2294 * list. */
2295int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data)
2296{
2297 struct dlm_ctxt *dlm = data;
2298 struct dlm_lock_resource *res = NULL;
2299 struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
2300 struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
2301 const char *name;
2302 unsigned int namelen;
2303 int ret = 0;
2304
2305 if (!dlm_grab(dlm))
2306 return -EINVAL;
2307
2308 name = migrate->name;
2309 namelen = migrate->namelen;
2310
2311 /* preallocate.. if this fails, abort */
2312 mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache,
2313 GFP_KERNEL);
2314
2315 if (!mle) {
2316 ret = -ENOMEM;
2317 goto leave;
2318 }
2319
2320 /* check for pre-existing lock */
2321 spin_lock(&dlm->spinlock);
2322 res = __dlm_lookup_lockres(dlm, name, namelen);
2323 spin_lock(&dlm->master_lock);
2324
2325 if (res) {
2326 spin_lock(&res->spinlock);
2327 if (res->state & DLM_LOCK_RES_RECOVERING) {
2328 /* if all is working ok, this can only mean that we got
2329 * a migrate request from a node that we now see as
2330 * dead. what can we do here? drop it to the floor? */
2331 spin_unlock(&res->spinlock);
2332 mlog(ML_ERROR, "Got a migrate request, but the "
2333 "lockres is marked as recovering!");
2334 kmem_cache_free(dlm_mle_cache, mle);
2335 ret = -EINVAL; /* need a better solution */
2336 goto unlock;
2337 }
2338 res->state |= DLM_LOCK_RES_MIGRATING;
2339 spin_unlock(&res->spinlock);
2340 }
2341
2342 /* ignore status. only nonzero status would BUG. */
2343 ret = dlm_add_migration_mle(dlm, res, mle, &oldmle,
2344 name, namelen,
2345 migrate->new_master,
2346 migrate->master);
2347
2348unlock:
2349 spin_unlock(&dlm->master_lock);
2350 spin_unlock(&dlm->spinlock);
2351
2352 if (oldmle) {
2353 /* master is known, detach if not already detached */
2354 dlm_mle_detach_hb_events(dlm, oldmle);
2355 dlm_put_mle(oldmle);
2356 }
2357
2358 if (res)
2359 dlm_lockres_put(res);
2360leave:
2361 dlm_put(dlm);
2362 return ret;
2363}
2364
2365/* must be holding dlm->spinlock and dlm->master_lock
2366 * when adding a migration mle, we can clear any other mles
2367 * in the master list because we know with certainty that
2368 * the master is "master". so we remove any old mle from
2369 * the list after setting it's master field, and then add
2370 * the new migration mle. this way we can hold with the rule
2371 * of having only one mle for a given lock name at all times. */
2372static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
2373 struct dlm_lock_resource *res,
2374 struct dlm_master_list_entry *mle,
2375 struct dlm_master_list_entry **oldmle,
2376 const char *name, unsigned int namelen,
2377 u8 new_master, u8 master)
2378{
2379 int found;
2380 int ret = 0;
2381
2382 *oldmle = NULL;
2383
2384 mlog_entry_void();
2385
2386 assert_spin_locked(&dlm->spinlock);
2387 assert_spin_locked(&dlm->master_lock);
2388
2389 /* caller is responsible for any ref taken here on oldmle */
2390 found = dlm_find_mle(dlm, oldmle, (char *)name, namelen);
2391 if (found) {
2392 struct dlm_master_list_entry *tmp = *oldmle;
2393 spin_lock(&tmp->spinlock);
2394 if (tmp->type == DLM_MLE_MIGRATION) {
2395 if (master == dlm->node_num) {
2396 /* ah another process raced me to it */
2397 mlog(0, "tried to migrate %.*s, but some "
2398 "process beat me to it\n",
2399 namelen, name);
2400 ret = -EEXIST;
2401 } else {
2402 /* bad. 2 NODES are trying to migrate! */
2403 mlog(ML_ERROR, "migration error mle: "
2404 "master=%u new_master=%u // request: "
2405 "master=%u new_master=%u // "
2406 "lockres=%.*s\n",
2407 tmp->master, tmp->new_master,
2408 master, new_master,
2409 namelen, name);
2410 BUG();
2411 }
2412 } else {
2413 /* this is essentially what assert_master does */
2414 tmp->master = master;
2415 atomic_set(&tmp->woken, 1);
2416 wake_up(&tmp->wq);
2417 /* remove it from the list so that only one
2418 * mle will be found */
2419 list_del_init(&tmp->list);
2420 }
2421 spin_unlock(&tmp->spinlock);
2422 }
2423
2424 /* now add a migration mle to the tail of the list */
2425 dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen);
2426 mle->new_master = new_master;
2427 mle->master = master;
2428 /* do this for consistency with other mle types */
2429 set_bit(new_master, mle->maybe_map);
2430 list_add(&mle->list, &dlm->master_list);
2431
2432 return ret;
2433}
2434
2435
2436void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
2437{
2438 struct list_head *iter, *iter2;
2439 struct dlm_master_list_entry *mle;
2440 struct dlm_lock_resource *res;
2441
2442 mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
2443top:
2444 assert_spin_locked(&dlm->spinlock);
2445
2446 /* clean the master list */
2447 spin_lock(&dlm->master_lock);
2448 list_for_each_safe(iter, iter2, &dlm->master_list) {
2449 mle = list_entry(iter, struct dlm_master_list_entry, list);
2450
2451 BUG_ON(mle->type != DLM_MLE_BLOCK &&
2452 mle->type != DLM_MLE_MASTER &&
2453 mle->type != DLM_MLE_MIGRATION);
2454
2455 /* MASTER mles are initiated locally. the waiting
2456 * process will notice the node map change
2457 * shortly. let that happen as normal. */
2458 if (mle->type == DLM_MLE_MASTER)
2459 continue;
2460
2461
2462 /* BLOCK mles are initiated by other nodes.
2463 * need to clean up if the dead node would have
2464 * been the master. */
2465 if (mle->type == DLM_MLE_BLOCK) {
2466 int bit;
2467
2468 spin_lock(&mle->spinlock);
2469 bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
2470 if (bit != dead_node) {
2471 mlog(0, "mle found, but dead node %u would "
2472 "not have been master\n", dead_node);
2473 spin_unlock(&mle->spinlock);
2474 } else {
2475 /* must drop the refcount by one since the
2476 * assert_master will never arrive. this
2477 * may result in the mle being unlinked and
2478 * freed, but there may still be a process
2479 * waiting in the dlmlock path which is fine. */
2480 mlog(ML_ERROR, "node %u was expected master\n",
2481 dead_node);
2482 atomic_set(&mle->woken, 1);
2483 spin_unlock(&mle->spinlock);
2484 wake_up(&mle->wq);
2485 /* final put will take care of list removal */
2486 __dlm_put_mle(mle);
2487 }
2488 continue;
2489 }
2490
2491 /* everything else is a MIGRATION mle */
2492
2493 /* the rule for MIGRATION mles is that the master
2494 * becomes UNKNOWN if *either* the original or
2495 * the new master dies. all UNKNOWN lockreses
2496 * are sent to whichever node becomes the recovery
2497 * master. the new master is responsible for
2498 * determining if there is still a master for
2499 * this lockres, or if he needs to take over
2500 * mastery. either way, this node should expect
2501 * another message to resolve this. */
2502 if (mle->master != dead_node &&
2503 mle->new_master != dead_node)
2504 continue;
2505
2506 /* if we have reached this point, this mle needs to
2507 * be removed from the list and freed. */
2508
2509 /* remove from the list early. NOTE: unlinking
2510 * list_head while in list_for_each_safe */
2511 spin_lock(&mle->spinlock);
2512 list_del_init(&mle->list);
2513 atomic_set(&mle->woken, 1);
2514 spin_unlock(&mle->spinlock);
2515 wake_up(&mle->wq);
2516
2517 mlog(0, "node %u died during migration from "
2518 "%u to %u!\n", dead_node,
2519 mle->master, mle->new_master);
2520 /* if there is a lockres associated with this
2521 * mle, find it and set its owner to UNKNOWN */
2522 res = __dlm_lookup_lockres(dlm, mle->u.name.name,
2523 mle->u.name.len);
2524 if (res) {
2525 /* unfortunately if we hit this rare case, our
2526 * lock ordering is messed. we need to drop
2527 * the master lock so that we can take the
2528 * lockres lock, meaning that we will have to
2529 * restart from the head of list. */
2530 spin_unlock(&dlm->master_lock);
2531
2532 /* move lockres onto recovery list */
2533 spin_lock(&res->spinlock);
2534 dlm_set_lockres_owner(dlm, res,
2535 DLM_LOCK_RES_OWNER_UNKNOWN);
2536 dlm_move_lockres_to_recovery_list(dlm, res);
2537 spin_unlock(&res->spinlock);
2538 dlm_lockres_put(res);
2539
2540 /* dump the mle */
2541 spin_lock(&dlm->master_lock);
2542 __dlm_put_mle(mle);
2543 spin_unlock(&dlm->master_lock);
2544
2545 /* restart */
2546 goto top;
2547 }
2548
2549 /* this may be the last reference */
2550 __dlm_put_mle(mle);
2551 }
2552 spin_unlock(&dlm->master_lock);
2553}
2554
2555
2556int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
2557 u8 old_master)
2558{
2559 struct dlm_node_iter iter;
2560 int ret = 0;
2561
2562 spin_lock(&dlm->spinlock);
2563 dlm_node_iter_init(dlm->domain_map, &iter);
2564 clear_bit(old_master, iter.node_map);
2565 clear_bit(dlm->node_num, iter.node_map);
2566 spin_unlock(&dlm->spinlock);
2567
2568 mlog(0, "now time to do a migrate request to other nodes\n");
2569 ret = dlm_do_migrate_request(dlm, res, old_master,
2570 dlm->node_num, &iter);
2571 if (ret < 0) {
2572 mlog_errno(ret);
2573 goto leave;
2574 }
2575
2576 mlog(0, "doing assert master of %.*s to all except the original node\n",
2577 res->lockname.len, res->lockname.name);
2578 /* this call now finishes out the nodemap
2579 * even if one or more nodes die */
2580 ret = dlm_do_assert_master(dlm, res->lockname.name,
2581 res->lockname.len, iter.node_map,
2582 DLM_ASSERT_MASTER_FINISH_MIGRATION);
2583 if (ret < 0) {
2584 /* no longer need to retry. all living nodes contacted. */
2585 mlog_errno(ret);
2586 ret = 0;
2587 }
2588
2589 memset(iter.node_map, 0, sizeof(iter.node_map));
2590 set_bit(old_master, iter.node_map);
2591 mlog(0, "doing assert master of %.*s back to %u\n",
2592 res->lockname.len, res->lockname.name, old_master);
2593 ret = dlm_do_assert_master(dlm, res->lockname.name,
2594 res->lockname.len, iter.node_map,
2595 DLM_ASSERT_MASTER_FINISH_MIGRATION);
2596 if (ret < 0) {
2597 mlog(0, "assert master to original master failed "
2598 "with %d.\n", ret);
2599 /* the only nonzero status here would be because of
2600 * a dead original node. we're done. */
2601 ret = 0;
2602 }
2603
2604 /* all done, set the owner, clear the flag */
2605 spin_lock(&res->spinlock);
2606 dlm_set_lockres_owner(dlm, res, dlm->node_num);
2607 res->state &= ~DLM_LOCK_RES_MIGRATING;
2608 spin_unlock(&res->spinlock);
2609 /* re-dirty it on the new master */
2610 dlm_kick_thread(dlm, res);
2611 wake_up(&res->wq);
2612leave:
2613 return ret;
2614}
2615
2616/*
2617 * LOCKRES AST REFCOUNT
2618 * this is integral to migration
2619 */
2620
2621/* for future intent to call an ast, reserve one ahead of time.
2622 * this should be called only after waiting on the lockres
2623 * with dlm_wait_on_lockres, and while still holding the
2624 * spinlock after the call. */
2625void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res)
2626{
2627 assert_spin_locked(&res->spinlock);
2628 if (res->state & DLM_LOCK_RES_MIGRATING) {
2629 __dlm_print_one_lock_resource(res);
2630 }
2631 BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
2632
2633 atomic_inc(&res->asts_reserved);
2634}
2635
2636/*
2637 * used to drop the reserved ast, either because it went unused,
2638 * or because the ast/bast was actually called.
2639 *
2640 * also, if there is a pending migration on this lockres,
2641 * and this was the last pending ast on the lockres,
2642 * atomically set the MIGRATING flag before we drop the lock.
2643 * this is how we ensure that migration can proceed with no
2644 * asts in progress. note that it is ok if the state of the
2645 * queues is such that a lock should be granted in the future
2646 * or that a bast should be fired, because the new master will
2647 * shuffle the lists on this lockres as soon as it is migrated.
2648 */
2649void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
2650 struct dlm_lock_resource *res)
2651{
2652 if (!atomic_dec_and_lock(&res->asts_reserved, &res->spinlock))
2653 return;
2654
2655 if (!res->migration_pending) {
2656 spin_unlock(&res->spinlock);
2657 return;
2658 }
2659
2660 BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
2661 res->migration_pending = 0;
2662 res->state |= DLM_LOCK_RES_MIGRATING;
2663 spin_unlock(&res->spinlock);
2664 wake_up(&res->wq);
2665 wake_up(&dlm->migration_wq);
2666}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-14 06:57:38 -0500