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/init.h>
34 #include <linux/sysctl.h>
35 #include <linux/random.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/spinlock.h>
40 #include <linux/delay.h>
41 
42 
43 #include "cluster/heartbeat.h"
44 #include "cluster/nodemanager.h"
45 #include "cluster/tcp.h"
46 
47 #include "dlmapi.h"
48 #include "dlmcommon.h"
49 #include "dlmdomain.h"
50 #include "dlmdebug.h"
51 
52 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
53 #include "cluster/masklog.h"
54 
55 static void dlm_mle_node_down(struct dlm_ctxt *dlm,
56 			      struct dlm_master_list_entry *mle,
57 			      struct o2nm_node *node,
58 			      int idx);
59 static void dlm_mle_node_up(struct dlm_ctxt *dlm,
60 			    struct dlm_master_list_entry *mle,
61 			    struct o2nm_node *node,
62 			    int idx);
63 
64 static void dlm_assert_master_worker(struct dlm_work_item *item, void *data);
65 static int dlm_do_assert_master(struct dlm_ctxt *dlm,
66 				struct dlm_lock_resource *res,
67 				void *nodemap, u32 flags);
68 static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data);
69 
dlm_mle_equal(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle,const char * name,unsigned int namelen)70 static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
71 				struct dlm_master_list_entry *mle,
72 				const char *name,
73 				unsigned int namelen)
74 {
75 	if (dlm != mle->dlm)
76 		return 0;
77 
78 	if (namelen != mle->mnamelen ||
79 	    memcmp(name, mle->mname, namelen) != 0)
80 		return 0;
81 
82 	return 1;
83 }
84 
85 static struct kmem_cache *dlm_lockres_cache = NULL;
86 static struct kmem_cache *dlm_lockname_cache = NULL;
87 static struct kmem_cache *dlm_mle_cache = NULL;
88 
89 static void dlm_mle_release(struct kref *kref);
90 static void dlm_init_mle(struct dlm_master_list_entry *mle,
91 			enum dlm_mle_type type,
92 			struct dlm_ctxt *dlm,
93 			struct dlm_lock_resource *res,
94 			const char *name,
95 			unsigned int namelen);
96 static void dlm_put_mle(struct dlm_master_list_entry *mle);
97 static void __dlm_put_mle(struct dlm_master_list_entry *mle);
98 static int dlm_find_mle(struct dlm_ctxt *dlm,
99 			struct dlm_master_list_entry **mle,
100 			char *name, unsigned int namelen);
101 
102 static int dlm_do_master_request(struct dlm_lock_resource *res,
103 				 struct dlm_master_list_entry *mle, int to);
104 
105 
106 static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
107 				     struct dlm_lock_resource *res,
108 				     struct dlm_master_list_entry *mle,
109 				     int *blocked);
110 static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
111 				    struct dlm_lock_resource *res,
112 				    struct dlm_master_list_entry *mle,
113 				    int blocked);
114 static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
115 				 struct dlm_lock_resource *res,
116 				 struct dlm_master_list_entry *mle,
117 				 struct dlm_master_list_entry **oldmle,
118 				 const char *name, unsigned int namelen,
119 				 u8 new_master, u8 master);
120 
121 static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
122 				    struct dlm_lock_resource *res);
123 static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
124 				      struct dlm_lock_resource *res);
125 static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
126 				       struct dlm_lock_resource *res,
127 				       u8 target);
128 static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
129 				       struct dlm_lock_resource *res);
130 
131 
dlm_is_host_down(int errno)132 int dlm_is_host_down(int errno)
133 {
134 	switch (errno) {
135 		case -EBADF:
136 		case -ECONNREFUSED:
137 		case -ENOTCONN:
138 		case -ECONNRESET:
139 		case -EPIPE:
140 		case -EHOSTDOWN:
141 		case -EHOSTUNREACH:
142 		case -ETIMEDOUT:
143 		case -ECONNABORTED:
144 		case -ENETDOWN:
145 		case -ENETUNREACH:
146 		case -ENETRESET:
147 		case -ESHUTDOWN:
148 		case -ENOPROTOOPT:
149 		case -EINVAL:   /* if returned from our tcp code,
150 				   this means there is no socket */
151 			return 1;
152 	}
153 	return 0;
154 }
155 
156 
157 /*
158  * MASTER LIST FUNCTIONS
159  */
160 
161 
162 /*
163  * regarding master list entries and heartbeat callbacks:
164  *
165  * in order to avoid sleeping and allocation that occurs in
166  * heartbeat, master list entries are simply attached to the
167  * dlm's established heartbeat callbacks.  the mle is attached
168  * when it is created, and since the dlm->spinlock is held at
169  * that time, any heartbeat event will be properly discovered
170  * by the mle.  the mle needs to be detached from the
171  * dlm->mle_hb_events list as soon as heartbeat events are no
172  * longer useful to the mle, and before the mle is freed.
173  *
174  * as a general rule, heartbeat events are no longer needed by
175  * the mle once an "answer" regarding the lock master has been
176  * received.
177  */
__dlm_mle_attach_hb_events(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)178 static inline void __dlm_mle_attach_hb_events(struct dlm_ctxt *dlm,
179 					      struct dlm_master_list_entry *mle)
180 {
181 	assert_spin_locked(&dlm->spinlock);
182 
183 	list_add_tail(&mle->hb_events, &dlm->mle_hb_events);
184 }
185 
186 
__dlm_mle_detach_hb_events(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)187 static inline void __dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
188 					      struct dlm_master_list_entry *mle)
189 {
190 	if (!list_empty(&mle->hb_events))
191 		list_del_init(&mle->hb_events);
192 }
193 
194 
dlm_mle_detach_hb_events(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)195 static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
196 					    struct dlm_master_list_entry *mle)
197 {
198 	spin_lock(&dlm->spinlock);
199 	__dlm_mle_detach_hb_events(dlm, mle);
200 	spin_unlock(&dlm->spinlock);
201 }
202 
dlm_get_mle_inuse(struct dlm_master_list_entry * mle)203 static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle)
204 {
205 	struct dlm_ctxt *dlm;
206 	dlm = mle->dlm;
207 
208 	assert_spin_locked(&dlm->spinlock);
209 	assert_spin_locked(&dlm->master_lock);
210 	mle->inuse++;
211 	kref_get(&mle->mle_refs);
212 }
213 
dlm_put_mle_inuse(struct dlm_master_list_entry * mle)214 static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle)
215 {
216 	struct dlm_ctxt *dlm;
217 	dlm = mle->dlm;
218 
219 	spin_lock(&dlm->spinlock);
220 	spin_lock(&dlm->master_lock);
221 	mle->inuse--;
222 	__dlm_put_mle(mle);
223 	spin_unlock(&dlm->master_lock);
224 	spin_unlock(&dlm->spinlock);
225 
226 }
227 
228 /* remove from list and free */
__dlm_put_mle(struct dlm_master_list_entry * mle)229 static void __dlm_put_mle(struct dlm_master_list_entry *mle)
230 {
231 	struct dlm_ctxt *dlm;
232 	dlm = mle->dlm;
233 
234 	assert_spin_locked(&dlm->spinlock);
235 	assert_spin_locked(&dlm->master_lock);
236 	if (!atomic_read(&mle->mle_refs.refcount)) {
237 		/* this may or may not crash, but who cares.
238 		 * it's a BUG. */
239 		mlog(ML_ERROR, "bad mle: %p\n", mle);
240 		dlm_print_one_mle(mle);
241 		BUG();
242 	} else
243 		kref_put(&mle->mle_refs, dlm_mle_release);
244 }
245 
246 
247 /* must not have any spinlocks coming in */
dlm_put_mle(struct dlm_master_list_entry * mle)248 static void dlm_put_mle(struct dlm_master_list_entry *mle)
249 {
250 	struct dlm_ctxt *dlm;
251 	dlm = mle->dlm;
252 
253 	spin_lock(&dlm->spinlock);
254 	spin_lock(&dlm->master_lock);
255 	__dlm_put_mle(mle);
256 	spin_unlock(&dlm->master_lock);
257 	spin_unlock(&dlm->spinlock);
258 }
259 
dlm_get_mle(struct dlm_master_list_entry * mle)260 static inline void dlm_get_mle(struct dlm_master_list_entry *mle)
261 {
262 	kref_get(&mle->mle_refs);
263 }
264 
dlm_init_mle(struct dlm_master_list_entry * mle,enum dlm_mle_type type,struct dlm_ctxt * dlm,struct dlm_lock_resource * res,const char * name,unsigned int namelen)265 static void dlm_init_mle(struct dlm_master_list_entry *mle,
266 			enum dlm_mle_type type,
267 			struct dlm_ctxt *dlm,
268 			struct dlm_lock_resource *res,
269 			const char *name,
270 			unsigned int namelen)
271 {
272 	assert_spin_locked(&dlm->spinlock);
273 
274 	mle->dlm = dlm;
275 	mle->type = type;
276 	INIT_HLIST_NODE(&mle->master_hash_node);
277 	INIT_LIST_HEAD(&mle->hb_events);
278 	memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
279 	spin_lock_init(&mle->spinlock);
280 	init_waitqueue_head(&mle->wq);
281 	atomic_set(&mle->woken, 0);
282 	kref_init(&mle->mle_refs);
283 	memset(mle->response_map, 0, sizeof(mle->response_map));
284 	mle->master = O2NM_MAX_NODES;
285 	mle->new_master = O2NM_MAX_NODES;
286 	mle->inuse = 0;
287 
288 	BUG_ON(mle->type != DLM_MLE_BLOCK &&
289 	       mle->type != DLM_MLE_MASTER &&
290 	       mle->type != DLM_MLE_MIGRATION);
291 
292 	if (mle->type == DLM_MLE_MASTER) {
293 		BUG_ON(!res);
294 		mle->mleres = res;
295 		memcpy(mle->mname, res->lockname.name, res->lockname.len);
296 		mle->mnamelen = res->lockname.len;
297 		mle->mnamehash = res->lockname.hash;
298 	} else {
299 		BUG_ON(!name);
300 		mle->mleres = NULL;
301 		memcpy(mle->mname, name, namelen);
302 		mle->mnamelen = namelen;
303 		mle->mnamehash = dlm_lockid_hash(name, namelen);
304 	}
305 
306 	atomic_inc(&dlm->mle_tot_count[mle->type]);
307 	atomic_inc(&dlm->mle_cur_count[mle->type]);
308 
309 	/* copy off the node_map and register hb callbacks on our copy */
310 	memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map));
311 	memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map));
312 	clear_bit(dlm->node_num, mle->vote_map);
313 	clear_bit(dlm->node_num, mle->node_map);
314 
315 	/* attach the mle to the domain node up/down events */
316 	__dlm_mle_attach_hb_events(dlm, mle);
317 }
318 
__dlm_unlink_mle(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)319 void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle)
320 {
321 	assert_spin_locked(&dlm->spinlock);
322 	assert_spin_locked(&dlm->master_lock);
323 
324 	if (!hlist_unhashed(&mle->master_hash_node))
325 		hlist_del_init(&mle->master_hash_node);
326 }
327 
__dlm_insert_mle(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)328 void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle)
329 {
330 	struct hlist_head *bucket;
331 
332 	assert_spin_locked(&dlm->master_lock);
333 
334 	bucket = dlm_master_hash(dlm, mle->mnamehash);
335 	hlist_add_head(&mle->master_hash_node, bucket);
336 }
337 
338 /* returns 1 if found, 0 if not */
dlm_find_mle(struct dlm_ctxt * dlm,struct dlm_master_list_entry ** mle,char * name,unsigned int namelen)339 static int dlm_find_mle(struct dlm_ctxt *dlm,
340 			struct dlm_master_list_entry **mle,
341 			char *name, unsigned int namelen)
342 {
343 	struct dlm_master_list_entry *tmpmle;
344 	struct hlist_head *bucket;
345 	struct hlist_node *list;
346 	unsigned int hash;
347 
348 	assert_spin_locked(&dlm->master_lock);
349 
350 	hash = dlm_lockid_hash(name, namelen);
351 	bucket = dlm_master_hash(dlm, hash);
352 	hlist_for_each(list, bucket) {
353 		tmpmle = hlist_entry(list, struct dlm_master_list_entry,
354 				     master_hash_node);
355 		if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
356 			continue;
357 		dlm_get_mle(tmpmle);
358 		*mle = tmpmle;
359 		return 1;
360 	}
361 	return 0;
362 }
363 
dlm_hb_event_notify_attached(struct dlm_ctxt * dlm,int idx,int node_up)364 void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up)
365 {
366 	struct dlm_master_list_entry *mle;
367 
368 	assert_spin_locked(&dlm->spinlock);
369 
370 	list_for_each_entry(mle, &dlm->mle_hb_events, hb_events) {
371 		if (node_up)
372 			dlm_mle_node_up(dlm, mle, NULL, idx);
373 		else
374 			dlm_mle_node_down(dlm, mle, NULL, idx);
375 	}
376 }
377 
dlm_mle_node_down(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle,struct o2nm_node * node,int idx)378 static void dlm_mle_node_down(struct dlm_ctxt *dlm,
379 			      struct dlm_master_list_entry *mle,
380 			      struct o2nm_node *node, int idx)
381 {
382 	spin_lock(&mle->spinlock);
383 
384 	if (!test_bit(idx, mle->node_map))
385 		mlog(0, "node %u already removed from nodemap!\n", idx);
386 	else
387 		clear_bit(idx, mle->node_map);
388 
389 	spin_unlock(&mle->spinlock);
390 }
391 
dlm_mle_node_up(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle,struct o2nm_node * node,int idx)392 static void dlm_mle_node_up(struct dlm_ctxt *dlm,
393 			    struct dlm_master_list_entry *mle,
394 			    struct o2nm_node *node, int idx)
395 {
396 	spin_lock(&mle->spinlock);
397 
398 	if (test_bit(idx, mle->node_map))
399 		mlog(0, "node %u already in node map!\n", idx);
400 	else
401 		set_bit(idx, mle->node_map);
402 
403 	spin_unlock(&mle->spinlock);
404 }
405 
406 
dlm_init_mle_cache(void)407 int dlm_init_mle_cache(void)
408 {
409 	dlm_mle_cache = kmem_cache_create("o2dlm_mle",
410 					  sizeof(struct dlm_master_list_entry),
411 					  0, SLAB_HWCACHE_ALIGN,
412 					  NULL);
413 	if (dlm_mle_cache == NULL)
414 		return -ENOMEM;
415 	return 0;
416 }
417 
dlm_destroy_mle_cache(void)418 void dlm_destroy_mle_cache(void)
419 {
420 	if (dlm_mle_cache)
421 		kmem_cache_destroy(dlm_mle_cache);
422 }
423 
dlm_mle_release(struct kref * kref)424 static void dlm_mle_release(struct kref *kref)
425 {
426 	struct dlm_master_list_entry *mle;
427 	struct dlm_ctxt *dlm;
428 
429 	mle = container_of(kref, struct dlm_master_list_entry, mle_refs);
430 	dlm = mle->dlm;
431 
432 	assert_spin_locked(&dlm->spinlock);
433 	assert_spin_locked(&dlm->master_lock);
434 
435 	mlog(0, "Releasing mle for %.*s, type %d\n", mle->mnamelen, mle->mname,
436 	     mle->type);
437 
438 	/* remove from list if not already */
439 	__dlm_unlink_mle(dlm, mle);
440 
441 	/* detach the mle from the domain node up/down events */
442 	__dlm_mle_detach_hb_events(dlm, mle);
443 
444 	atomic_dec(&dlm->mle_cur_count[mle->type]);
445 
446 	/* NOTE: kfree under spinlock here.
447 	 * if this is bad, we can move this to a freelist. */
448 	kmem_cache_free(dlm_mle_cache, mle);
449 }
450 
451 
452 /*
453  * LOCK RESOURCE FUNCTIONS
454  */
455 
dlm_init_master_caches(void)456 int dlm_init_master_caches(void)
457 {
458 	dlm_lockres_cache = kmem_cache_create("o2dlm_lockres",
459 					      sizeof(struct dlm_lock_resource),
460 					      0, SLAB_HWCACHE_ALIGN, NULL);
461 	if (!dlm_lockres_cache)
462 		goto bail;
463 
464 	dlm_lockname_cache = kmem_cache_create("o2dlm_lockname",
465 					       DLM_LOCKID_NAME_MAX, 0,
466 					       SLAB_HWCACHE_ALIGN, NULL);
467 	if (!dlm_lockname_cache)
468 		goto bail;
469 
470 	return 0;
471 bail:
472 	dlm_destroy_master_caches();
473 	return -ENOMEM;
474 }
475 
dlm_destroy_master_caches(void)476 void dlm_destroy_master_caches(void)
477 {
478 	if (dlm_lockname_cache)
479 		kmem_cache_destroy(dlm_lockname_cache);
480 
481 	if (dlm_lockres_cache)
482 		kmem_cache_destroy(dlm_lockres_cache);
483 }
484 
dlm_lockres_release(struct kref * kref)485 static void dlm_lockres_release(struct kref *kref)
486 {
487 	struct dlm_lock_resource *res;
488 	struct dlm_ctxt *dlm;
489 
490 	res = container_of(kref, struct dlm_lock_resource, refs);
491 	dlm = res->dlm;
492 
493 	/* This should not happen -- all lockres' have a name
494 	 * associated with them at init time. */
495 	BUG_ON(!res->lockname.name);
496 
497 	mlog(0, "destroying lockres %.*s\n", res->lockname.len,
498 	     res->lockname.name);
499 
500 	spin_lock(&dlm->track_lock);
501 	if (!list_empty(&res->tracking))
502 		list_del_init(&res->tracking);
503 	else {
504 		mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n",
505 		     res->lockname.len, res->lockname.name);
506 		dlm_print_one_lock_resource(res);
507 	}
508 	spin_unlock(&dlm->track_lock);
509 
510 	atomic_dec(&dlm->res_cur_count);
511 
512 	if (!hlist_unhashed(&res->hash_node) ||
513 	    !list_empty(&res->granted) ||
514 	    !list_empty(&res->converting) ||
515 	    !list_empty(&res->blocked) ||
516 	    !list_empty(&res->dirty) ||
517 	    !list_empty(&res->recovering) ||
518 	    !list_empty(&res->purge)) {
519 		mlog(ML_ERROR,
520 		     "Going to BUG for resource %.*s."
521 		     "  We're on a list! [%c%c%c%c%c%c%c]\n",
522 		     res->lockname.len, res->lockname.name,
523 		     !hlist_unhashed(&res->hash_node) ? 'H' : ' ',
524 		     !list_empty(&res->granted) ? 'G' : ' ',
525 		     !list_empty(&res->converting) ? 'C' : ' ',
526 		     !list_empty(&res->blocked) ? 'B' : ' ',
527 		     !list_empty(&res->dirty) ? 'D' : ' ',
528 		     !list_empty(&res->recovering) ? 'R' : ' ',
529 		     !list_empty(&res->purge) ? 'P' : ' ');
530 
531 		dlm_print_one_lock_resource(res);
532 	}
533 
534 	/* By the time we're ready to blow this guy away, we shouldn't
535 	 * be on any lists. */
536 	BUG_ON(!hlist_unhashed(&res->hash_node));
537 	BUG_ON(!list_empty(&res->granted));
538 	BUG_ON(!list_empty(&res->converting));
539 	BUG_ON(!list_empty(&res->blocked));
540 	BUG_ON(!list_empty(&res->dirty));
541 	BUG_ON(!list_empty(&res->recovering));
542 	BUG_ON(!list_empty(&res->purge));
543 
544 	kmem_cache_free(dlm_lockname_cache, (void *)res->lockname.name);
545 
546 	kmem_cache_free(dlm_lockres_cache, res);
547 }
548 
dlm_lockres_put(struct dlm_lock_resource * res)549 void dlm_lockres_put(struct dlm_lock_resource *res)
550 {
551 	kref_put(&res->refs, dlm_lockres_release);
552 }
553 
dlm_init_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,const char * name,unsigned int namelen)554 static void dlm_init_lockres(struct dlm_ctxt *dlm,
555 			     struct dlm_lock_resource *res,
556 			     const char *name, unsigned int namelen)
557 {
558 	char *qname;
559 
560 	/* If we memset here, we lose our reference to the kmalloc'd
561 	 * res->lockname.name, so be sure to init every field
562 	 * correctly! */
563 
564 	qname = (char *) res->lockname.name;
565 	memcpy(qname, name, namelen);
566 
567 	res->lockname.len = namelen;
568 	res->lockname.hash = dlm_lockid_hash(name, namelen);
569 
570 	init_waitqueue_head(&res->wq);
571 	spin_lock_init(&res->spinlock);
572 	INIT_HLIST_NODE(&res->hash_node);
573 	INIT_LIST_HEAD(&res->granted);
574 	INIT_LIST_HEAD(&res->converting);
575 	INIT_LIST_HEAD(&res->blocked);
576 	INIT_LIST_HEAD(&res->dirty);
577 	INIT_LIST_HEAD(&res->recovering);
578 	INIT_LIST_HEAD(&res->purge);
579 	INIT_LIST_HEAD(&res->tracking);
580 	atomic_set(&res->asts_reserved, 0);
581 	res->migration_pending = 0;
582 	res->inflight_locks = 0;
583 
584 	res->dlm = dlm;
585 
586 	kref_init(&res->refs);
587 
588 	atomic_inc(&dlm->res_tot_count);
589 	atomic_inc(&dlm->res_cur_count);
590 
591 	/* just for consistency */
592 	spin_lock(&res->spinlock);
593 	dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
594 	spin_unlock(&res->spinlock);
595 
596 	res->state = DLM_LOCK_RES_IN_PROGRESS;
597 
598 	res->last_used = 0;
599 
600 	spin_lock(&dlm->spinlock);
601 	list_add_tail(&res->tracking, &dlm->tracking_list);
602 	spin_unlock(&dlm->spinlock);
603 
604 	memset(res->lvb, 0, DLM_LVB_LEN);
605 	memset(res->refmap, 0, sizeof(res->refmap));
606 }
607 
dlm_new_lockres(struct dlm_ctxt * dlm,const char * name,unsigned int namelen)608 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
609 				   const char *name,
610 				   unsigned int namelen)
611 {
612 	struct dlm_lock_resource *res = NULL;
613 
614 	res = kmem_cache_zalloc(dlm_lockres_cache, GFP_NOFS);
615 	if (!res)
616 		goto error;
617 
618 	res->lockname.name = kmem_cache_zalloc(dlm_lockname_cache, GFP_NOFS);
619 	if (!res->lockname.name)
620 		goto error;
621 
622 	dlm_init_lockres(dlm, res, name, namelen);
623 	return res;
624 
625 error:
626 	if (res && res->lockname.name)
627 		kmem_cache_free(dlm_lockname_cache, (void *)res->lockname.name);
628 
629 	if (res)
630 		kmem_cache_free(dlm_lockres_cache, res);
631 	return NULL;
632 }
633 
dlm_lockres_set_refmap_bit(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,int bit)634 void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
635 				struct dlm_lock_resource *res, int bit)
636 {
637 	assert_spin_locked(&res->spinlock);
638 
639 	mlog(0, "res %.*s, set node %u, %ps()\n", res->lockname.len,
640 	     res->lockname.name, bit, __builtin_return_address(0));
641 
642 	set_bit(bit, res->refmap);
643 }
644 
dlm_lockres_clear_refmap_bit(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,int bit)645 void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
646 				  struct dlm_lock_resource *res, int bit)
647 {
648 	assert_spin_locked(&res->spinlock);
649 
650 	mlog(0, "res %.*s, clr node %u, %ps()\n", res->lockname.len,
651 	     res->lockname.name, bit, __builtin_return_address(0));
652 
653 	clear_bit(bit, res->refmap);
654 }
655 
656 
dlm_lockres_grab_inflight_ref(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)657 void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
658 				   struct dlm_lock_resource *res)
659 {
660 	assert_spin_locked(&res->spinlock);
661 
662 	res->inflight_locks++;
663 
664 	mlog(0, "%s: res %.*s, inflight++: now %u, %ps()\n", dlm->name,
665 	     res->lockname.len, res->lockname.name, res->inflight_locks,
666 	     __builtin_return_address(0));
667 }
668 
dlm_lockres_drop_inflight_ref(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)669 void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
670 				   struct dlm_lock_resource *res)
671 {
672 	assert_spin_locked(&res->spinlock);
673 
674 	BUG_ON(res->inflight_locks == 0);
675 
676 	res->inflight_locks--;
677 
678 	mlog(0, "%s: res %.*s, inflight--: now %u, %ps()\n", dlm->name,
679 	     res->lockname.len, res->lockname.name, res->inflight_locks,
680 	     __builtin_return_address(0));
681 
682 	wake_up(&res->wq);
683 }
684 
685 /*
686  * lookup a lock resource by name.
687  * may already exist in the hashtable.
688  * lockid is null terminated
689  *
690  * if not, allocate enough for the lockres and for
691  * the temporary structure used in doing the mastering.
692  *
693  * also, do a lookup in the dlm->master_list to see
694  * if another node has begun mastering the same lock.
695  * if so, there should be a block entry in there
696  * for this name, and we should *not* attempt to master
697  * the lock here.   need to wait around for that node
698  * to assert_master (or die).
699  *
700  */
dlm_get_lock_resource(struct dlm_ctxt * dlm,const char * lockid,int namelen,int flags)701 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
702 					  const char *lockid,
703 					  int namelen,
704 					  int flags)
705 {
706 	struct dlm_lock_resource *tmpres=NULL, *res=NULL;
707 	struct dlm_master_list_entry *mle = NULL;
708 	struct dlm_master_list_entry *alloc_mle = NULL;
709 	int blocked = 0;
710 	int ret, nodenum;
711 	struct dlm_node_iter iter;
712 	unsigned int hash;
713 	int tries = 0;
714 	int bit, wait_on_recovery = 0;
715 
716 	BUG_ON(!lockid);
717 
718 	hash = dlm_lockid_hash(lockid, namelen);
719 
720 	mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
721 
722 lookup:
723 	spin_lock(&dlm->spinlock);
724 	tmpres = __dlm_lookup_lockres_full(dlm, lockid, namelen, hash);
725 	if (tmpres) {
726 		spin_unlock(&dlm->spinlock);
727 		spin_lock(&tmpres->spinlock);
728 		/* Wait on the thread that is mastering the resource */
729 		if (tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
730 			__dlm_wait_on_lockres(tmpres);
731 			BUG_ON(tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN);
732 			spin_unlock(&tmpres->spinlock);
733 			dlm_lockres_put(tmpres);
734 			tmpres = NULL;
735 			goto lookup;
736 		}
737 
738 		/* Wait on the resource purge to complete before continuing */
739 		if (tmpres->state & DLM_LOCK_RES_DROPPING_REF) {
740 			BUG_ON(tmpres->owner == dlm->node_num);
741 			__dlm_wait_on_lockres_flags(tmpres,
742 						    DLM_LOCK_RES_DROPPING_REF);
743 			spin_unlock(&tmpres->spinlock);
744 			dlm_lockres_put(tmpres);
745 			tmpres = NULL;
746 			goto lookup;
747 		}
748 
749 		/* Grab inflight ref to pin the resource */
750 		dlm_lockres_grab_inflight_ref(dlm, tmpres);
751 
752 		spin_unlock(&tmpres->spinlock);
753 		if (res)
754 			dlm_lockres_put(res);
755 		res = tmpres;
756 		goto leave;
757 	}
758 
759 	if (!res) {
760 		spin_unlock(&dlm->spinlock);
761 		mlog(0, "allocating a new resource\n");
762 		/* nothing found and we need to allocate one. */
763 		alloc_mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
764 		if (!alloc_mle)
765 			goto leave;
766 		res = dlm_new_lockres(dlm, lockid, namelen);
767 		if (!res)
768 			goto leave;
769 		goto lookup;
770 	}
771 
772 	mlog(0, "no lockres found, allocated our own: %p\n", res);
773 
774 	if (flags & LKM_LOCAL) {
775 		/* caller knows it's safe to assume it's not mastered elsewhere
776 		 * DONE!  return right away */
777 		spin_lock(&res->spinlock);
778 		dlm_change_lockres_owner(dlm, res, dlm->node_num);
779 		__dlm_insert_lockres(dlm, res);
780 		dlm_lockres_grab_inflight_ref(dlm, res);
781 		spin_unlock(&res->spinlock);
782 		spin_unlock(&dlm->spinlock);
783 		/* lockres still marked IN_PROGRESS */
784 		goto wake_waiters;
785 	}
786 
787 	/* check master list to see if another node has started mastering it */
788 	spin_lock(&dlm->master_lock);
789 
790 	/* if we found a block, wait for lock to be mastered by another node */
791 	blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen);
792 	if (blocked) {
793 		int mig;
794 		if (mle->type == DLM_MLE_MASTER) {
795 			mlog(ML_ERROR, "master entry for nonexistent lock!\n");
796 			BUG();
797 		}
798 		mig = (mle->type == DLM_MLE_MIGRATION);
799 		/* if there is a migration in progress, let the migration
800 		 * finish before continuing.  we can wait for the absence
801 		 * of the MIGRATION mle: either the migrate finished or
802 		 * one of the nodes died and the mle was cleaned up.
803 		 * if there is a BLOCK here, but it already has a master
804 		 * set, we are too late.  the master does not have a ref
805 		 * for us in the refmap.  detach the mle and drop it.
806 		 * either way, go back to the top and start over. */
807 		if (mig || mle->master != O2NM_MAX_NODES) {
808 			BUG_ON(mig && mle->master == dlm->node_num);
809 			/* we arrived too late.  the master does not
810 			 * have a ref for us. retry. */
811 			mlog(0, "%s:%.*s: late on %s\n",
812 			     dlm->name, namelen, lockid,
813 			     mig ?  "MIGRATION" : "BLOCK");
814 			spin_unlock(&dlm->master_lock);
815 			spin_unlock(&dlm->spinlock);
816 
817 			/* master is known, detach */
818 			if (!mig)
819 				dlm_mle_detach_hb_events(dlm, mle);
820 			dlm_put_mle(mle);
821 			mle = NULL;
822 			/* this is lame, but we can't wait on either
823 			 * the mle or lockres waitqueue here */
824 			if (mig)
825 				msleep(100);
826 			goto lookup;
827 		}
828 	} else {
829 		/* go ahead and try to master lock on this node */
830 		mle = alloc_mle;
831 		/* make sure this does not get freed below */
832 		alloc_mle = NULL;
833 		dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
834 		set_bit(dlm->node_num, mle->maybe_map);
835 		__dlm_insert_mle(dlm, mle);
836 
837 		/* still holding the dlm spinlock, check the recovery map
838 		 * to see if there are any nodes that still need to be
839 		 * considered.  these will not appear in the mle nodemap
840 		 * but they might own this lockres.  wait on them. */
841 		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
842 		if (bit < O2NM_MAX_NODES) {
843 			mlog(0, "%s: res %.*s, At least one node (%d) "
844 			     "to recover before lock mastery can begin\n",
845 			     dlm->name, namelen, (char *)lockid, bit);
846 			wait_on_recovery = 1;
847 		}
848 	}
849 
850 	/* at this point there is either a DLM_MLE_BLOCK or a
851 	 * DLM_MLE_MASTER on the master list, so it's safe to add the
852 	 * lockres to the hashtable.  anyone who finds the lock will
853 	 * still have to wait on the IN_PROGRESS. */
854 
855 	/* finally add the lockres to its hash bucket */
856 	__dlm_insert_lockres(dlm, res);
857 
858 	/* Grab inflight ref to pin the resource */
859 	spin_lock(&res->spinlock);
860 	dlm_lockres_grab_inflight_ref(dlm, res);
861 	spin_unlock(&res->spinlock);
862 
863 	/* get an extra ref on the mle in case this is a BLOCK
864 	 * if so, the creator of the BLOCK may try to put the last
865 	 * ref at this time in the assert master handler, so we
866 	 * need an extra one to keep from a bad ptr deref. */
867 	dlm_get_mle_inuse(mle);
868 	spin_unlock(&dlm->master_lock);
869 	spin_unlock(&dlm->spinlock);
870 
871 redo_request:
872 	while (wait_on_recovery) {
873 		/* any cluster changes that occurred after dropping the
874 		 * dlm spinlock would be detectable be a change on the mle,
875 		 * so we only need to clear out the recovery map once. */
876 		if (dlm_is_recovery_lock(lockid, namelen)) {
877 			mlog(0, "%s: Recovery map is not empty, but must "
878 			     "master $RECOVERY lock now\n", dlm->name);
879 			if (!dlm_pre_master_reco_lockres(dlm, res))
880 				wait_on_recovery = 0;
881 			else {
882 				mlog(0, "%s: waiting 500ms for heartbeat state "
883 				    "change\n", dlm->name);
884 				msleep(500);
885 			}
886 			continue;
887 		}
888 
889 		dlm_kick_recovery_thread(dlm);
890 		msleep(1000);
891 		dlm_wait_for_recovery(dlm);
892 
893 		spin_lock(&dlm->spinlock);
894 		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
895 		if (bit < O2NM_MAX_NODES) {
896 			mlog(0, "%s: res %.*s, At least one node (%d) "
897 			     "to recover before lock mastery can begin\n",
898 			     dlm->name, namelen, (char *)lockid, bit);
899 			wait_on_recovery = 1;
900 		} else
901 			wait_on_recovery = 0;
902 		spin_unlock(&dlm->spinlock);
903 
904 		if (wait_on_recovery)
905 			dlm_wait_for_node_recovery(dlm, bit, 10000);
906 	}
907 
908 	/* must wait for lock to be mastered elsewhere */
909 	if (blocked)
910 		goto wait;
911 
912 	ret = -EINVAL;
913 	dlm_node_iter_init(mle->vote_map, &iter);
914 	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
915 		ret = dlm_do_master_request(res, mle, nodenum);
916 		if (ret < 0)
917 			mlog_errno(ret);
918 		if (mle->master != O2NM_MAX_NODES) {
919 			/* found a master ! */
920 			if (mle->master <= nodenum)
921 				break;
922 			/* if our master request has not reached the master
923 			 * yet, keep going until it does.  this is how the
924 			 * master will know that asserts are needed back to
925 			 * the lower nodes. */
926 			mlog(0, "%s: res %.*s, Requests only up to %u but "
927 			     "master is %u, keep going\n", dlm->name, namelen,
928 			     lockid, nodenum, mle->master);
929 		}
930 	}
931 
932 wait:
933 	/* keep going until the response map includes all nodes */
934 	ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
935 	if (ret < 0) {
936 		wait_on_recovery = 1;
937 		mlog(0, "%s: res %.*s, Node map changed, redo the master "
938 		     "request now, blocked=%d\n", dlm->name, res->lockname.len,
939 		     res->lockname.name, blocked);
940 		if (++tries > 20) {
941 			mlog(ML_ERROR, "%s: res %.*s, Spinning on "
942 			     "dlm_wait_for_lock_mastery, blocked = %d\n",
943 			     dlm->name, res->lockname.len,
944 			     res->lockname.name, blocked);
945 			dlm_print_one_lock_resource(res);
946 			dlm_print_one_mle(mle);
947 			tries = 0;
948 		}
949 		goto redo_request;
950 	}
951 
952 	mlog(0, "%s: res %.*s, Mastered by %u\n", dlm->name, res->lockname.len,
953 	     res->lockname.name, res->owner);
954 	/* make sure we never continue without this */
955 	BUG_ON(res->owner == O2NM_MAX_NODES);
956 
957 	/* master is known, detach if not already detached */
958 	dlm_mle_detach_hb_events(dlm, mle);
959 	dlm_put_mle(mle);
960 	/* put the extra ref */
961 	dlm_put_mle_inuse(mle);
962 
963 wake_waiters:
964 	spin_lock(&res->spinlock);
965 	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
966 	spin_unlock(&res->spinlock);
967 	wake_up(&res->wq);
968 
969 leave:
970 	/* need to free the unused mle */
971 	if (alloc_mle)
972 		kmem_cache_free(dlm_mle_cache, alloc_mle);
973 
974 	return res;
975 }
976 
977 
978 #define DLM_MASTERY_TIMEOUT_MS   5000
979 
dlm_wait_for_lock_mastery(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_master_list_entry * mle,int * blocked)980 static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
981 				     struct dlm_lock_resource *res,
982 				     struct dlm_master_list_entry *mle,
983 				     int *blocked)
984 {
985 	u8 m;
986 	int ret, bit;
987 	int map_changed, voting_done;
988 	int assert, sleep;
989 
990 recheck:
991 	ret = 0;
992 	assert = 0;
993 
994 	/* check if another node has already become the owner */
995 	spin_lock(&res->spinlock);
996 	if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
997 		mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name,
998 		     res->lockname.len, res->lockname.name, res->owner);
999 		spin_unlock(&res->spinlock);
1000 		/* this will cause the master to re-assert across
1001 		 * the whole cluster, freeing up mles */
1002 		if (res->owner != dlm->node_num) {
1003 			ret = dlm_do_master_request(res, mle, res->owner);
1004 			if (ret < 0) {
1005 				/* give recovery a chance to run */
1006 				mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
1007 				msleep(500);
1008 				goto recheck;
1009 			}
1010 		}
1011 		ret = 0;
1012 		goto leave;
1013 	}
1014 	spin_unlock(&res->spinlock);
1015 
1016 	spin_lock(&mle->spinlock);
1017 	m = mle->master;
1018 	map_changed = (memcmp(mle->vote_map, mle->node_map,
1019 			      sizeof(mle->vote_map)) != 0);
1020 	voting_done = (memcmp(mle->vote_map, mle->response_map,
1021 			     sizeof(mle->vote_map)) == 0);
1022 
1023 	/* restart if we hit any errors */
1024 	if (map_changed) {
1025 		int b;
1026 		mlog(0, "%s: %.*s: node map changed, restarting\n",
1027 		     dlm->name, res->lockname.len, res->lockname.name);
1028 		ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked);
1029 		b = (mle->type == DLM_MLE_BLOCK);
1030 		if ((*blocked && !b) || (!*blocked && b)) {
1031 			mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
1032 			     dlm->name, res->lockname.len, res->lockname.name,
1033 			     *blocked, b);
1034 			*blocked = b;
1035 		}
1036 		spin_unlock(&mle->spinlock);
1037 		if (ret < 0) {
1038 			mlog_errno(ret);
1039 			goto leave;
1040 		}
1041 		mlog(0, "%s:%.*s: restart lock mastery succeeded, "
1042 		     "rechecking now\n", dlm->name, res->lockname.len,
1043 		     res->lockname.name);
1044 		goto recheck;
1045 	} else {
1046 		if (!voting_done) {
1047 			mlog(0, "map not changed and voting not done "
1048 			     "for %s:%.*s\n", dlm->name, res->lockname.len,
1049 			     res->lockname.name);
1050 		}
1051 	}
1052 
1053 	if (m != O2NM_MAX_NODES) {
1054 		/* another node has done an assert!
1055 		 * all done! */
1056 		sleep = 0;
1057 	} else {
1058 		sleep = 1;
1059 		/* have all nodes responded? */
1060 		if (voting_done && !*blocked) {
1061 			bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
1062 			if (dlm->node_num <= bit) {
1063 				/* my node number is lowest.
1064 			 	 * now tell other nodes that I am
1065 				 * mastering this. */
1066 				mle->master = dlm->node_num;
1067 				/* ref was grabbed in get_lock_resource
1068 				 * will be dropped in dlmlock_master */
1069 				assert = 1;
1070 				sleep = 0;
1071 			}
1072 			/* if voting is done, but we have not received
1073 			 * an assert master yet, we must sleep */
1074 		}
1075 	}
1076 
1077 	spin_unlock(&mle->spinlock);
1078 
1079 	/* sleep if we haven't finished voting yet */
1080 	if (sleep) {
1081 		unsigned long timeo = msecs_to_jiffies(DLM_MASTERY_TIMEOUT_MS);
1082 
1083 		/*
1084 		if (atomic_read(&mle->mle_refs.refcount) < 2)
1085 			mlog(ML_ERROR, "mle (%p) refs=%d, name=%.*s\n", mle,
1086 			atomic_read(&mle->mle_refs.refcount),
1087 			res->lockname.len, res->lockname.name);
1088 		*/
1089 		atomic_set(&mle->woken, 0);
1090 		(void)wait_event_timeout(mle->wq,
1091 					 (atomic_read(&mle->woken) == 1),
1092 					 timeo);
1093 		if (res->owner == O2NM_MAX_NODES) {
1094 			mlog(0, "%s:%.*s: waiting again\n", dlm->name,
1095 			     res->lockname.len, res->lockname.name);
1096 			goto recheck;
1097 		}
1098 		mlog(0, "done waiting, master is %u\n", res->owner);
1099 		ret = 0;
1100 		goto leave;
1101 	}
1102 
1103 	ret = 0;   /* done */
1104 	if (assert) {
1105 		m = dlm->node_num;
1106 		mlog(0, "about to master %.*s here, this=%u\n",
1107 		     res->lockname.len, res->lockname.name, m);
1108 		ret = dlm_do_assert_master(dlm, res, mle->vote_map, 0);
1109 		if (ret) {
1110 			/* This is a failure in the network path,
1111 			 * not in the response to the assert_master
1112 			 * (any nonzero response is a BUG on this node).
1113 			 * Most likely a socket just got disconnected
1114 			 * due to node death. */
1115 			mlog_errno(ret);
1116 		}
1117 		/* no longer need to restart lock mastery.
1118 		 * all living nodes have been contacted. */
1119 		ret = 0;
1120 	}
1121 
1122 	/* set the lockres owner */
1123 	spin_lock(&res->spinlock);
1124 	/* mastery reference obtained either during
1125 	 * assert_master_handler or in get_lock_resource */
1126 	dlm_change_lockres_owner(dlm, res, m);
1127 	spin_unlock(&res->spinlock);
1128 
1129 leave:
1130 	return ret;
1131 }
1132 
1133 struct dlm_bitmap_diff_iter
1134 {
1135 	int curnode;
1136 	unsigned long *orig_bm;
1137 	unsigned long *cur_bm;
1138 	unsigned long diff_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
1139 };
1140 
1141 enum dlm_node_state_change
1142 {
1143 	NODE_DOWN = -1,
1144 	NODE_NO_CHANGE = 0,
1145 	NODE_UP
1146 };
1147 
dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter * iter,unsigned long * orig_bm,unsigned long * cur_bm)1148 static void dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter *iter,
1149 				      unsigned long *orig_bm,
1150 				      unsigned long *cur_bm)
1151 {
1152 	unsigned long p1, p2;
1153 	int i;
1154 
1155 	iter->curnode = -1;
1156 	iter->orig_bm = orig_bm;
1157 	iter->cur_bm = cur_bm;
1158 
1159 	for (i = 0; i < BITS_TO_LONGS(O2NM_MAX_NODES); i++) {
1160        		p1 = *(iter->orig_bm + i);
1161 	       	p2 = *(iter->cur_bm + i);
1162 		iter->diff_bm[i] = (p1 & ~p2) | (p2 & ~p1);
1163 	}
1164 }
1165 
dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter * iter,enum dlm_node_state_change * state)1166 static int dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter *iter,
1167 				     enum dlm_node_state_change *state)
1168 {
1169 	int bit;
1170 
1171 	if (iter->curnode >= O2NM_MAX_NODES)
1172 		return -ENOENT;
1173 
1174 	bit = find_next_bit(iter->diff_bm, O2NM_MAX_NODES,
1175 			    iter->curnode+1);
1176 	if (bit >= O2NM_MAX_NODES) {
1177 		iter->curnode = O2NM_MAX_NODES;
1178 		return -ENOENT;
1179 	}
1180 
1181 	/* if it was there in the original then this node died */
1182 	if (test_bit(bit, iter->orig_bm))
1183 		*state = NODE_DOWN;
1184 	else
1185 		*state = NODE_UP;
1186 
1187 	iter->curnode = bit;
1188 	return bit;
1189 }
1190 
1191 
dlm_restart_lock_mastery(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_master_list_entry * mle,int blocked)1192 static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
1193 				    struct dlm_lock_resource *res,
1194 				    struct dlm_master_list_entry *mle,
1195 				    int blocked)
1196 {
1197 	struct dlm_bitmap_diff_iter bdi;
1198 	enum dlm_node_state_change sc;
1199 	int node;
1200 	int ret = 0;
1201 
1202 	mlog(0, "something happened such that the "
1203 	     "master process may need to be restarted!\n");
1204 
1205 	assert_spin_locked(&mle->spinlock);
1206 
1207 	dlm_bitmap_diff_iter_init(&bdi, mle->vote_map, mle->node_map);
1208 	node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1209 	while (node >= 0) {
1210 		if (sc == NODE_UP) {
1211 			/* a node came up.  clear any old vote from
1212 			 * the response map and set it in the vote map
1213 			 * then restart the mastery. */
1214 			mlog(ML_NOTICE, "node %d up while restarting\n", node);
1215 
1216 			/* redo the master request, but only for the new node */
1217 			mlog(0, "sending request to new node\n");
1218 			clear_bit(node, mle->response_map);
1219 			set_bit(node, mle->vote_map);
1220 		} else {
1221 			mlog(ML_ERROR, "node down! %d\n", node);
1222 			if (blocked) {
1223 				int lowest = find_next_bit(mle->maybe_map,
1224 						       O2NM_MAX_NODES, 0);
1225 
1226 				/* act like it was never there */
1227 				clear_bit(node, mle->maybe_map);
1228 
1229 			       	if (node == lowest) {
1230 					mlog(0, "expected master %u died"
1231 					    " while this node was blocked "
1232 					    "waiting on it!\n", node);
1233 					lowest = find_next_bit(mle->maybe_map,
1234 						       	O2NM_MAX_NODES,
1235 						       	lowest+1);
1236 					if (lowest < O2NM_MAX_NODES) {
1237 						mlog(0, "%s:%.*s:still "
1238 						     "blocked. waiting on %u "
1239 						     "now\n", dlm->name,
1240 						     res->lockname.len,
1241 						     res->lockname.name,
1242 						     lowest);
1243 					} else {
1244 						/* mle is an MLE_BLOCK, but
1245 						 * there is now nothing left to
1246 						 * block on.  we need to return
1247 						 * all the way back out and try
1248 						 * again with an MLE_MASTER.
1249 						 * dlm_do_local_recovery_cleanup
1250 						 * has already run, so the mle
1251 						 * refcount is ok */
1252 						mlog(0, "%s:%.*s: no "
1253 						     "longer blocking. try to "
1254 						     "master this here\n",
1255 						     dlm->name,
1256 						     res->lockname.len,
1257 						     res->lockname.name);
1258 						mle->type = DLM_MLE_MASTER;
1259 						mle->mleres = res;
1260 					}
1261 				}
1262 			}
1263 
1264 			/* now blank out everything, as if we had never
1265 			 * contacted anyone */
1266 			memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
1267 			memset(mle->response_map, 0, sizeof(mle->response_map));
1268 			/* reset the vote_map to the current node_map */
1269 			memcpy(mle->vote_map, mle->node_map,
1270 			       sizeof(mle->node_map));
1271 			/* put myself into the maybe map */
1272 			if (mle->type != DLM_MLE_BLOCK)
1273 				set_bit(dlm->node_num, mle->maybe_map);
1274 		}
1275 		ret = -EAGAIN;
1276 		node = dlm_bitmap_diff_iter_next(&bdi, &sc);
1277 	}
1278 	return ret;
1279 }
1280 
1281 
1282 /*
1283  * DLM_MASTER_REQUEST_MSG
1284  *
1285  * returns: 0 on success,
1286  *          -errno on a network error
1287  *
1288  * on error, the caller should assume the target node is "dead"
1289  *
1290  */
1291 
dlm_do_master_request(struct dlm_lock_resource * res,struct dlm_master_list_entry * mle,int to)1292 static int dlm_do_master_request(struct dlm_lock_resource *res,
1293 				 struct dlm_master_list_entry *mle, int to)
1294 {
1295 	struct dlm_ctxt *dlm = mle->dlm;
1296 	struct dlm_master_request request;
1297 	int ret, response=0, resend;
1298 
1299 	memset(&request, 0, sizeof(request));
1300 	request.node_idx = dlm->node_num;
1301 
1302 	BUG_ON(mle->type == DLM_MLE_MIGRATION);
1303 
1304 	request.namelen = (u8)mle->mnamelen;
1305 	memcpy(request.name, mle->mname, request.namelen);
1306 
1307 again:
1308 	ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request,
1309 				 sizeof(request), to, &response);
1310 	if (ret < 0)  {
1311 		if (ret == -ESRCH) {
1312 			/* should never happen */
1313 			mlog(ML_ERROR, "TCP stack not ready!\n");
1314 			BUG();
1315 		} else if (ret == -EINVAL) {
1316 			mlog(ML_ERROR, "bad args passed to o2net!\n");
1317 			BUG();
1318 		} else if (ret == -ENOMEM) {
1319 			mlog(ML_ERROR, "out of memory while trying to send "
1320 			     "network message!  retrying\n");
1321 			/* this is totally crude */
1322 			msleep(50);
1323 			goto again;
1324 		} else if (!dlm_is_host_down(ret)) {
1325 			/* not a network error. bad. */
1326 			mlog_errno(ret);
1327 			mlog(ML_ERROR, "unhandled error!");
1328 			BUG();
1329 		}
1330 		/* all other errors should be network errors,
1331 		 * and likely indicate node death */
1332 		mlog(ML_ERROR, "link to %d went down!\n", to);
1333 		goto out;
1334 	}
1335 
1336 	ret = 0;
1337 	resend = 0;
1338 	spin_lock(&mle->spinlock);
1339 	switch (response) {
1340 		case DLM_MASTER_RESP_YES:
1341 			set_bit(to, mle->response_map);
1342 			mlog(0, "node %u is the master, response=YES\n", to);
1343 			mlog(0, "%s:%.*s: master node %u now knows I have a "
1344 			     "reference\n", dlm->name, res->lockname.len,
1345 			     res->lockname.name, to);
1346 			mle->master = to;
1347 			break;
1348 		case DLM_MASTER_RESP_NO:
1349 			mlog(0, "node %u not master, response=NO\n", to);
1350 			set_bit(to, mle->response_map);
1351 			break;
1352 		case DLM_MASTER_RESP_MAYBE:
1353 			mlog(0, "node %u not master, response=MAYBE\n", to);
1354 			set_bit(to, mle->response_map);
1355 			set_bit(to, mle->maybe_map);
1356 			break;
1357 		case DLM_MASTER_RESP_ERROR:
1358 			mlog(0, "node %u hit an error, resending\n", to);
1359 			resend = 1;
1360 			response = 0;
1361 			break;
1362 		default:
1363 			mlog(ML_ERROR, "bad response! %u\n", response);
1364 			BUG();
1365 	}
1366 	spin_unlock(&mle->spinlock);
1367 	if (resend) {
1368 		/* this is also totally crude */
1369 		msleep(50);
1370 		goto again;
1371 	}
1372 
1373 out:
1374 	return ret;
1375 }
1376 
1377 /*
1378  * locks that can be taken here:
1379  * dlm->spinlock
1380  * res->spinlock
1381  * mle->spinlock
1382  * dlm->master_list
1383  *
1384  * if possible, TRIM THIS DOWN!!!
1385  */
dlm_master_request_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)1386 int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
1387 			       void **ret_data)
1388 {
1389 	u8 response = DLM_MASTER_RESP_MAYBE;
1390 	struct dlm_ctxt *dlm = data;
1391 	struct dlm_lock_resource *res = NULL;
1392 	struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
1393 	struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
1394 	char *name;
1395 	unsigned int namelen, hash;
1396 	int found, ret;
1397 	int set_maybe;
1398 	int dispatch_assert = 0;
1399 
1400 	if (!dlm_grab(dlm))
1401 		return DLM_MASTER_RESP_NO;
1402 
1403 	if (!dlm_domain_fully_joined(dlm)) {
1404 		response = DLM_MASTER_RESP_NO;
1405 		goto send_response;
1406 	}
1407 
1408 	name = request->name;
1409 	namelen = request->namelen;
1410 	hash = dlm_lockid_hash(name, namelen);
1411 
1412 	if (namelen > DLM_LOCKID_NAME_MAX) {
1413 		response = DLM_IVBUFLEN;
1414 		goto send_response;
1415 	}
1416 
1417 way_up_top:
1418 	spin_lock(&dlm->spinlock);
1419 	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
1420 	if (res) {
1421 		spin_unlock(&dlm->spinlock);
1422 
1423 		/* take care of the easy cases up front */
1424 		spin_lock(&res->spinlock);
1425 		if (res->state & (DLM_LOCK_RES_RECOVERING|
1426 				  DLM_LOCK_RES_MIGRATING)) {
1427 			spin_unlock(&res->spinlock);
1428 			mlog(0, "returning DLM_MASTER_RESP_ERROR since res is "
1429 			     "being recovered/migrated\n");
1430 			response = DLM_MASTER_RESP_ERROR;
1431 			if (mle)
1432 				kmem_cache_free(dlm_mle_cache, mle);
1433 			goto send_response;
1434 		}
1435 
1436 		if (res->owner == dlm->node_num) {
1437 			dlm_lockres_set_refmap_bit(dlm, res, request->node_idx);
1438 			spin_unlock(&res->spinlock);
1439 			response = DLM_MASTER_RESP_YES;
1440 			if (mle)
1441 				kmem_cache_free(dlm_mle_cache, mle);
1442 
1443 			/* this node is the owner.
1444 			 * there is some extra work that needs to
1445 			 * happen now.  the requesting node has
1446 			 * caused all nodes up to this one to
1447 			 * create mles.  this node now needs to
1448 			 * go back and clean those up. */
1449 			dispatch_assert = 1;
1450 			goto send_response;
1451 		} else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1452 			spin_unlock(&res->spinlock);
1453 			// mlog(0, "node %u is the master\n", res->owner);
1454 			response = DLM_MASTER_RESP_NO;
1455 			if (mle)
1456 				kmem_cache_free(dlm_mle_cache, mle);
1457 			goto send_response;
1458 		}
1459 
1460 		/* ok, there is no owner.  either this node is
1461 		 * being blocked, or it is actively trying to
1462 		 * master this lock. */
1463 		if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1464 			mlog(ML_ERROR, "lock with no owner should be "
1465 			     "in-progress!\n");
1466 			BUG();
1467 		}
1468 
1469 		// mlog(0, "lockres is in progress...\n");
1470 		spin_lock(&dlm->master_lock);
1471 		found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1472 		if (!found) {
1473 			mlog(ML_ERROR, "no mle found for this lock!\n");
1474 			BUG();
1475 		}
1476 		set_maybe = 1;
1477 		spin_lock(&tmpmle->spinlock);
1478 		if (tmpmle->type == DLM_MLE_BLOCK) {
1479 			// mlog(0, "this node is waiting for "
1480 			// "lockres to be mastered\n");
1481 			response = DLM_MASTER_RESP_NO;
1482 		} else if (tmpmle->type == DLM_MLE_MIGRATION) {
1483 			mlog(0, "node %u is master, but trying to migrate to "
1484 			     "node %u.\n", tmpmle->master, tmpmle->new_master);
1485 			if (tmpmle->master == dlm->node_num) {
1486 				mlog(ML_ERROR, "no owner on lockres, but this "
1487 				     "node is trying to migrate it to %u?!\n",
1488 				     tmpmle->new_master);
1489 				BUG();
1490 			} else {
1491 				/* the real master can respond on its own */
1492 				response = DLM_MASTER_RESP_NO;
1493 			}
1494 		} else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) {
1495 			set_maybe = 0;
1496 			if (tmpmle->master == dlm->node_num) {
1497 				response = DLM_MASTER_RESP_YES;
1498 				/* this node will be the owner.
1499 				 * go back and clean the mles on any
1500 				 * other nodes */
1501 				dispatch_assert = 1;
1502 				dlm_lockres_set_refmap_bit(dlm, res,
1503 							   request->node_idx);
1504 			} else
1505 				response = DLM_MASTER_RESP_NO;
1506 		} else {
1507 			// mlog(0, "this node is attempting to "
1508 			// "master lockres\n");
1509 			response = DLM_MASTER_RESP_MAYBE;
1510 		}
1511 		if (set_maybe)
1512 			set_bit(request->node_idx, tmpmle->maybe_map);
1513 		spin_unlock(&tmpmle->spinlock);
1514 
1515 		spin_unlock(&dlm->master_lock);
1516 		spin_unlock(&res->spinlock);
1517 
1518 		/* keep the mle attached to heartbeat events */
1519 		dlm_put_mle(tmpmle);
1520 		if (mle)
1521 			kmem_cache_free(dlm_mle_cache, mle);
1522 		goto send_response;
1523 	}
1524 
1525 	/*
1526 	 * lockres doesn't exist on this node
1527 	 * if there is an MLE_BLOCK, return NO
1528 	 * if there is an MLE_MASTER, return MAYBE
1529 	 * otherwise, add an MLE_BLOCK, return NO
1530 	 */
1531 	spin_lock(&dlm->master_lock);
1532 	found = dlm_find_mle(dlm, &tmpmle, name, namelen);
1533 	if (!found) {
1534 		/* this lockid has never been seen on this node yet */
1535 		// mlog(0, "no mle found\n");
1536 		if (!mle) {
1537 			spin_unlock(&dlm->master_lock);
1538 			spin_unlock(&dlm->spinlock);
1539 
1540 			mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
1541 			if (!mle) {
1542 				response = DLM_MASTER_RESP_ERROR;
1543 				mlog_errno(-ENOMEM);
1544 				goto send_response;
1545 			}
1546 			goto way_up_top;
1547 		}
1548 
1549 		// mlog(0, "this is second time thru, already allocated, "
1550 		// "add the block.\n");
1551 		dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
1552 		set_bit(request->node_idx, mle->maybe_map);
1553 		__dlm_insert_mle(dlm, mle);
1554 		response = DLM_MASTER_RESP_NO;
1555 	} else {
1556 		// mlog(0, "mle was found\n");
1557 		set_maybe = 1;
1558 		spin_lock(&tmpmle->spinlock);
1559 		if (tmpmle->master == dlm->node_num) {
1560 			mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n");
1561 			BUG();
1562 		}
1563 		if (tmpmle->type == DLM_MLE_BLOCK)
1564 			response = DLM_MASTER_RESP_NO;
1565 		else if (tmpmle->type == DLM_MLE_MIGRATION) {
1566 			mlog(0, "migration mle was found (%u->%u)\n",
1567 			     tmpmle->master, tmpmle->new_master);
1568 			/* real master can respond on its own */
1569 			response = DLM_MASTER_RESP_NO;
1570 		} else
1571 			response = DLM_MASTER_RESP_MAYBE;
1572 		if (set_maybe)
1573 			set_bit(request->node_idx, tmpmle->maybe_map);
1574 		spin_unlock(&tmpmle->spinlock);
1575 	}
1576 	spin_unlock(&dlm->master_lock);
1577 	spin_unlock(&dlm->spinlock);
1578 
1579 	if (found) {
1580 		/* keep the mle attached to heartbeat events */
1581 		dlm_put_mle(tmpmle);
1582 	}
1583 send_response:
1584 	/*
1585 	 * __dlm_lookup_lockres() grabbed a reference to this lockres.
1586 	 * The reference is released by dlm_assert_master_worker() under
1587 	 * the call to dlm_dispatch_assert_master().  If
1588 	 * dlm_assert_master_worker() isn't called, we drop it here.
1589 	 */
1590 	if (dispatch_assert) {
1591 		if (response != DLM_MASTER_RESP_YES)
1592 			mlog(ML_ERROR, "invalid response %d\n", response);
1593 		if (!res) {
1594 			mlog(ML_ERROR, "bad lockres while trying to assert!\n");
1595 			BUG();
1596 		}
1597 		mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
1598 			     dlm->node_num, res->lockname.len, res->lockname.name);
1599 		ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
1600 						 DLM_ASSERT_MASTER_MLE_CLEANUP);
1601 		if (ret < 0) {
1602 			mlog(ML_ERROR, "failed to dispatch assert master work\n");
1603 			response = DLM_MASTER_RESP_ERROR;
1604 			dlm_lockres_put(res);
1605 		}
1606 	} else {
1607 		if (res)
1608 			dlm_lockres_put(res);
1609 	}
1610 
1611 	dlm_put(dlm);
1612 	return response;
1613 }
1614 
1615 /*
1616  * DLM_ASSERT_MASTER_MSG
1617  */
1618 
1619 
1620 /*
1621  * NOTE: this can be used for debugging
1622  * can periodically run all locks owned by this node
1623  * and re-assert across the cluster...
1624  */
dlm_do_assert_master(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,void * nodemap,u32 flags)1625 static int dlm_do_assert_master(struct dlm_ctxt *dlm,
1626 				struct dlm_lock_resource *res,
1627 				void *nodemap, u32 flags)
1628 {
1629 	struct dlm_assert_master assert;
1630 	int to, tmpret;
1631 	struct dlm_node_iter iter;
1632 	int ret = 0;
1633 	int reassert;
1634 	const char *lockname = res->lockname.name;
1635 	unsigned int namelen = res->lockname.len;
1636 
1637 	BUG_ON(namelen > O2NM_MAX_NAME_LEN);
1638 
1639 	spin_lock(&res->spinlock);
1640 	res->state |= DLM_LOCK_RES_SETREF_INPROG;
1641 	spin_unlock(&res->spinlock);
1642 
1643 again:
1644 	reassert = 0;
1645 
1646 	/* note that if this nodemap is empty, it returns 0 */
1647 	dlm_node_iter_init(nodemap, &iter);
1648 	while ((to = dlm_node_iter_next(&iter)) >= 0) {
1649 		int r = 0;
1650 		struct dlm_master_list_entry *mle = NULL;
1651 
1652 		mlog(0, "sending assert master to %d (%.*s)\n", to,
1653 		     namelen, lockname);
1654 		memset(&assert, 0, sizeof(assert));
1655 		assert.node_idx = dlm->node_num;
1656 		assert.namelen = namelen;
1657 		memcpy(assert.name, lockname, namelen);
1658 		assert.flags = cpu_to_be32(flags);
1659 
1660 		tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
1661 					    &assert, sizeof(assert), to, &r);
1662 		if (tmpret < 0) {
1663 			mlog(ML_ERROR, "Error %d when sending message %u (key "
1664 			     "0x%x) to node %u\n", tmpret,
1665 			     DLM_ASSERT_MASTER_MSG, dlm->key, to);
1666 			if (!dlm_is_host_down(tmpret)) {
1667 				mlog(ML_ERROR, "unhandled error=%d!\n", tmpret);
1668 				BUG();
1669 			}
1670 			/* a node died.  finish out the rest of the nodes. */
1671 			mlog(0, "link to %d went down!\n", to);
1672 			/* any nonzero status return will do */
1673 			ret = tmpret;
1674 			r = 0;
1675 		} else if (r < 0) {
1676 			/* ok, something horribly messed.  kill thyself. */
1677 			mlog(ML_ERROR,"during assert master of %.*s to %u, "
1678 			     "got %d.\n", namelen, lockname, to, r);
1679 			spin_lock(&dlm->spinlock);
1680 			spin_lock(&dlm->master_lock);
1681 			if (dlm_find_mle(dlm, &mle, (char *)lockname,
1682 					 namelen)) {
1683 				dlm_print_one_mle(mle);
1684 				__dlm_put_mle(mle);
1685 			}
1686 			spin_unlock(&dlm->master_lock);
1687 			spin_unlock(&dlm->spinlock);
1688 			BUG();
1689 		}
1690 
1691 		if (r & DLM_ASSERT_RESPONSE_REASSERT &&
1692 		    !(r & DLM_ASSERT_RESPONSE_MASTERY_REF)) {
1693 				mlog(ML_ERROR, "%.*s: very strange, "
1694 				     "master MLE but no lockres on %u\n",
1695 				     namelen, lockname, to);
1696 		}
1697 
1698 		if (r & DLM_ASSERT_RESPONSE_REASSERT) {
1699 			mlog(0, "%.*s: node %u create mles on other "
1700 			     "nodes and requests a re-assert\n",
1701 			     namelen, lockname, to);
1702 			reassert = 1;
1703 		}
1704 		if (r & DLM_ASSERT_RESPONSE_MASTERY_REF) {
1705 			mlog(0, "%.*s: node %u has a reference to this "
1706 			     "lockres, set the bit in the refmap\n",
1707 			     namelen, lockname, to);
1708 			spin_lock(&res->spinlock);
1709 			dlm_lockres_set_refmap_bit(dlm, res, to);
1710 			spin_unlock(&res->spinlock);
1711 		}
1712 	}
1713 
1714 	if (reassert)
1715 		goto again;
1716 
1717 	spin_lock(&res->spinlock);
1718 	res->state &= ~DLM_LOCK_RES_SETREF_INPROG;
1719 	spin_unlock(&res->spinlock);
1720 	wake_up(&res->wq);
1721 
1722 	return ret;
1723 }
1724 
1725 /*
1726  * locks that can be taken here:
1727  * dlm->spinlock
1728  * res->spinlock
1729  * mle->spinlock
1730  * dlm->master_list
1731  *
1732  * if possible, TRIM THIS DOWN!!!
1733  */
dlm_assert_master_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)1734 int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
1735 			      void **ret_data)
1736 {
1737 	struct dlm_ctxt *dlm = data;
1738 	struct dlm_master_list_entry *mle = NULL;
1739 	struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
1740 	struct dlm_lock_resource *res = NULL;
1741 	char *name;
1742 	unsigned int namelen, hash;
1743 	u32 flags;
1744 	int master_request = 0, have_lockres_ref = 0;
1745 	int ret = 0;
1746 
1747 	if (!dlm_grab(dlm))
1748 		return 0;
1749 
1750 	name = assert->name;
1751 	namelen = assert->namelen;
1752 	hash = dlm_lockid_hash(name, namelen);
1753 	flags = be32_to_cpu(assert->flags);
1754 
1755 	if (namelen > DLM_LOCKID_NAME_MAX) {
1756 		mlog(ML_ERROR, "Invalid name length!");
1757 		goto done;
1758 	}
1759 
1760 	spin_lock(&dlm->spinlock);
1761 
1762 	if (flags)
1763 		mlog(0, "assert_master with flags: %u\n", flags);
1764 
1765 	/* find the MLE */
1766 	spin_lock(&dlm->master_lock);
1767 	if (!dlm_find_mle(dlm, &mle, name, namelen)) {
1768 		/* not an error, could be master just re-asserting */
1769 		mlog(0, "just got an assert_master from %u, but no "
1770 		     "MLE for it! (%.*s)\n", assert->node_idx,
1771 		     namelen, name);
1772 	} else {
1773 		int bit = find_next_bit (mle->maybe_map, O2NM_MAX_NODES, 0);
1774 		if (bit >= O2NM_MAX_NODES) {
1775 			/* not necessarily an error, though less likely.
1776 			 * could be master just re-asserting. */
1777 			mlog(0, "no bits set in the maybe_map, but %u "
1778 			     "is asserting! (%.*s)\n", assert->node_idx,
1779 			     namelen, name);
1780 		} else if (bit != assert->node_idx) {
1781 			if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1782 				mlog(0, "master %u was found, %u should "
1783 				     "back off\n", assert->node_idx, bit);
1784 			} else {
1785 				/* with the fix for bug 569, a higher node
1786 				 * number winning the mastery will respond
1787 				 * YES to mastery requests, but this node
1788 				 * had no way of knowing.  let it pass. */
1789 				mlog(0, "%u is the lowest node, "
1790 				     "%u is asserting. (%.*s)  %u must "
1791 				     "have begun after %u won.\n", bit,
1792 				     assert->node_idx, namelen, name, bit,
1793 				     assert->node_idx);
1794 			}
1795 		}
1796 		if (mle->type == DLM_MLE_MIGRATION) {
1797 			if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
1798 				mlog(0, "%s:%.*s: got cleanup assert"
1799 				     " from %u for migration\n",
1800 				     dlm->name, namelen, name,
1801 				     assert->node_idx);
1802 			} else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) {
1803 				mlog(0, "%s:%.*s: got unrelated assert"
1804 				     " from %u for migration, ignoring\n",
1805 				     dlm->name, namelen, name,
1806 				     assert->node_idx);
1807 				__dlm_put_mle(mle);
1808 				spin_unlock(&dlm->master_lock);
1809 				spin_unlock(&dlm->spinlock);
1810 				goto done;
1811 			}
1812 		}
1813 	}
1814 	spin_unlock(&dlm->master_lock);
1815 
1816 	/* ok everything checks out with the MLE
1817 	 * now check to see if there is a lockres */
1818 	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
1819 	if (res) {
1820 		spin_lock(&res->spinlock);
1821 		if (res->state & DLM_LOCK_RES_RECOVERING)  {
1822 			mlog(ML_ERROR, "%u asserting but %.*s is "
1823 			     "RECOVERING!\n", assert->node_idx, namelen, name);
1824 			goto kill;
1825 		}
1826 		if (!mle) {
1827 			if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN &&
1828 			    res->owner != assert->node_idx) {
1829 				mlog(ML_ERROR, "DIE! Mastery assert from %u, "
1830 				     "but current owner is %u! (%.*s)\n",
1831 				     assert->node_idx, res->owner, namelen,
1832 				     name);
1833 				__dlm_print_one_lock_resource(res);
1834 				BUG();
1835 			}
1836 		} else if (mle->type != DLM_MLE_MIGRATION) {
1837 			if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
1838 				/* owner is just re-asserting */
1839 				if (res->owner == assert->node_idx) {
1840 					mlog(0, "owner %u re-asserting on "
1841 					     "lock %.*s\n", assert->node_idx,
1842 					     namelen, name);
1843 					goto ok;
1844 				}
1845 				mlog(ML_ERROR, "got assert_master from "
1846 				     "node %u, but %u is the owner! "
1847 				     "(%.*s)\n", assert->node_idx,
1848 				     res->owner, namelen, name);
1849 				goto kill;
1850 			}
1851 			if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
1852 				mlog(ML_ERROR, "got assert from %u, but lock "
1853 				     "with no owner should be "
1854 				     "in-progress! (%.*s)\n",
1855 				     assert->node_idx,
1856 				     namelen, name);
1857 				goto kill;
1858 			}
1859 		} else /* mle->type == DLM_MLE_MIGRATION */ {
1860 			/* should only be getting an assert from new master */
1861 			if (assert->node_idx != mle->new_master) {
1862 				mlog(ML_ERROR, "got assert from %u, but "
1863 				     "new master is %u, and old master "
1864 				     "was %u (%.*s)\n",
1865 				     assert->node_idx, mle->new_master,
1866 				     mle->master, namelen, name);
1867 				goto kill;
1868 			}
1869 
1870 		}
1871 ok:
1872 		spin_unlock(&res->spinlock);
1873 	}
1874 
1875 	// mlog(0, "woo!  got an assert_master from node %u!\n",
1876 	// 	     assert->node_idx);
1877 	if (mle) {
1878 		int extra_ref = 0;
1879 		int nn = -1;
1880 		int rr, err = 0;
1881 
1882 		spin_lock(&mle->spinlock);
1883 		if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
1884 			extra_ref = 1;
1885 		else {
1886 			/* MASTER mle: if any bits set in the response map
1887 			 * then the calling node needs to re-assert to clear
1888 			 * up nodes that this node contacted */
1889 			while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
1890 						    nn+1)) < O2NM_MAX_NODES) {
1891 				if (nn != dlm->node_num && nn != assert->node_idx)
1892 					master_request = 1;
1893 			}
1894 		}
1895 		mle->master = assert->node_idx;
1896 		atomic_set(&mle->woken, 1);
1897 		wake_up(&mle->wq);
1898 		spin_unlock(&mle->spinlock);
1899 
1900 		if (res) {
1901 			int wake = 0;
1902 			spin_lock(&res->spinlock);
1903 			if (mle->type == DLM_MLE_MIGRATION) {
1904 				mlog(0, "finishing off migration of lockres %.*s, "
1905 			     		"from %u to %u\n",
1906 			       		res->lockname.len, res->lockname.name,
1907 			       		dlm->node_num, mle->new_master);
1908 				res->state &= ~DLM_LOCK_RES_MIGRATING;
1909 				wake = 1;
1910 				dlm_change_lockres_owner(dlm, res, mle->new_master);
1911 				BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
1912 			} else {
1913 				dlm_change_lockres_owner(dlm, res, mle->master);
1914 			}
1915 			spin_unlock(&res->spinlock);
1916 			have_lockres_ref = 1;
1917 			if (wake)
1918 				wake_up(&res->wq);
1919 		}
1920 
1921 		/* master is known, detach if not already detached.
1922 		 * ensures that only one assert_master call will happen
1923 		 * on this mle. */
1924 		spin_lock(&dlm->master_lock);
1925 
1926 		rr = atomic_read(&mle->mle_refs.refcount);
1927 		if (mle->inuse > 0) {
1928 			if (extra_ref && rr < 3)
1929 				err = 1;
1930 			else if (!extra_ref && rr < 2)
1931 				err = 1;
1932 		} else {
1933 			if (extra_ref && rr < 2)
1934 				err = 1;
1935 			else if (!extra_ref && rr < 1)
1936 				err = 1;
1937 		}
1938 		if (err) {
1939 			mlog(ML_ERROR, "%s:%.*s: got assert master from %u "
1940 			     "that will mess up this node, refs=%d, extra=%d, "
1941 			     "inuse=%d\n", dlm->name, namelen, name,
1942 			     assert->node_idx, rr, extra_ref, mle->inuse);
1943 			dlm_print_one_mle(mle);
1944 		}
1945 		__dlm_unlink_mle(dlm, mle);
1946 		__dlm_mle_detach_hb_events(dlm, mle);
1947 		__dlm_put_mle(mle);
1948 		if (extra_ref) {
1949 			/* the assert master message now balances the extra
1950 		 	 * ref given by the master / migration request message.
1951 		 	 * if this is the last put, it will be removed
1952 		 	 * from the list. */
1953 			__dlm_put_mle(mle);
1954 		}
1955 		spin_unlock(&dlm->master_lock);
1956 	} else if (res) {
1957 		if (res->owner != assert->node_idx) {
1958 			mlog(0, "assert_master from %u, but current "
1959 			     "owner is %u (%.*s), no mle\n", assert->node_idx,
1960 			     res->owner, namelen, name);
1961 		}
1962 	}
1963 	spin_unlock(&dlm->spinlock);
1964 
1965 done:
1966 	ret = 0;
1967 	if (res) {
1968 		spin_lock(&res->spinlock);
1969 		res->state |= DLM_LOCK_RES_SETREF_INPROG;
1970 		spin_unlock(&res->spinlock);
1971 		*ret_data = (void *)res;
1972 	}
1973 	dlm_put(dlm);
1974 	if (master_request) {
1975 		mlog(0, "need to tell master to reassert\n");
1976 		/* positive. negative would shoot down the node. */
1977 		ret |= DLM_ASSERT_RESPONSE_REASSERT;
1978 		if (!have_lockres_ref) {
1979 			mlog(ML_ERROR, "strange, got assert from %u, MASTER "
1980 			     "mle present here for %s:%.*s, but no lockres!\n",
1981 			     assert->node_idx, dlm->name, namelen, name);
1982 		}
1983 	}
1984 	if (have_lockres_ref) {
1985 		/* let the master know we have a reference to the lockres */
1986 		ret |= DLM_ASSERT_RESPONSE_MASTERY_REF;
1987 		mlog(0, "%s:%.*s: got assert from %u, need a ref\n",
1988 		     dlm->name, namelen, name, assert->node_idx);
1989 	}
1990 	return ret;
1991 
1992 kill:
1993 	/* kill the caller! */
1994 	mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
1995 	     "and killing the other node now!  This node is OK and can continue.\n");
1996 	__dlm_print_one_lock_resource(res);
1997 	spin_unlock(&res->spinlock);
1998 	spin_unlock(&dlm->spinlock);
1999 	*ret_data = (void *)res;
2000 	dlm_put(dlm);
2001 	return -EINVAL;
2002 }
2003 
dlm_assert_master_post_handler(int status,void * data,void * ret_data)2004 void dlm_assert_master_post_handler(int status, void *data, void *ret_data)
2005 {
2006 	struct dlm_lock_resource *res = (struct dlm_lock_resource *)ret_data;
2007 
2008 	if (ret_data) {
2009 		spin_lock(&res->spinlock);
2010 		res->state &= ~DLM_LOCK_RES_SETREF_INPROG;
2011 		spin_unlock(&res->spinlock);
2012 		wake_up(&res->wq);
2013 		dlm_lockres_put(res);
2014 	}
2015 	return;
2016 }
2017 
dlm_dispatch_assert_master(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,int ignore_higher,u8 request_from,u32 flags)2018 int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
2019 			       struct dlm_lock_resource *res,
2020 			       int ignore_higher, u8 request_from, u32 flags)
2021 {
2022 	struct dlm_work_item *item;
2023 	item = kzalloc(sizeof(*item), GFP_NOFS);
2024 	if (!item)
2025 		return -ENOMEM;
2026 
2027 
2028 	/* queue up work for dlm_assert_master_worker */
2029 	dlm_grab(dlm);  /* get an extra ref for the work item */
2030 	dlm_init_work_item(dlm, item, dlm_assert_master_worker, NULL);
2031 	item->u.am.lockres = res; /* already have a ref */
2032 	/* can optionally ignore node numbers higher than this node */
2033 	item->u.am.ignore_higher = ignore_higher;
2034 	item->u.am.request_from = request_from;
2035 	item->u.am.flags = flags;
2036 
2037 	if (ignore_higher)
2038 		mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
2039 		     res->lockname.name);
2040 
2041 	spin_lock(&dlm->work_lock);
2042 	list_add_tail(&item->list, &dlm->work_list);
2043 	spin_unlock(&dlm->work_lock);
2044 
2045 	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
2046 	return 0;
2047 }
2048 
dlm_assert_master_worker(struct dlm_work_item * item,void * data)2049 static void dlm_assert_master_worker(struct dlm_work_item *item, void *data)
2050 {
2051 	struct dlm_ctxt *dlm = data;
2052 	int ret = 0;
2053 	struct dlm_lock_resource *res;
2054 	unsigned long nodemap[BITS_TO_LONGS(O2NM_MAX_NODES)];
2055 	int ignore_higher;
2056 	int bit;
2057 	u8 request_from;
2058 	u32 flags;
2059 
2060 	dlm = item->dlm;
2061 	res = item->u.am.lockres;
2062 	ignore_higher = item->u.am.ignore_higher;
2063 	request_from = item->u.am.request_from;
2064 	flags = item->u.am.flags;
2065 
2066 	spin_lock(&dlm->spinlock);
2067 	memcpy(nodemap, dlm->domain_map, sizeof(nodemap));
2068 	spin_unlock(&dlm->spinlock);
2069 
2070 	clear_bit(dlm->node_num, nodemap);
2071 	if (ignore_higher) {
2072 		/* if is this just to clear up mles for nodes below
2073 		 * this node, do not send the message to the original
2074 		 * caller or any node number higher than this */
2075 		clear_bit(request_from, nodemap);
2076 		bit = dlm->node_num;
2077 		while (1) {
2078 			bit = find_next_bit(nodemap, O2NM_MAX_NODES,
2079 					    bit+1);
2080 		       	if (bit >= O2NM_MAX_NODES)
2081 				break;
2082 			clear_bit(bit, nodemap);
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * If we're migrating this lock to someone else, we are no
2088 	 * longer allowed to assert out own mastery.  OTOH, we need to
2089 	 * prevent migration from starting while we're still asserting
2090 	 * our dominance.  The reserved ast delays migration.
2091 	 */
2092 	spin_lock(&res->spinlock);
2093 	if (res->state & DLM_LOCK_RES_MIGRATING) {
2094 		mlog(0, "Someone asked us to assert mastery, but we're "
2095 		     "in the middle of migration.  Skipping assert, "
2096 		     "the new master will handle that.\n");
2097 		spin_unlock(&res->spinlock);
2098 		goto put;
2099 	} else
2100 		__dlm_lockres_reserve_ast(res);
2101 	spin_unlock(&res->spinlock);
2102 
2103 	/* this call now finishes out the nodemap
2104 	 * even if one or more nodes die */
2105 	mlog(0, "worker about to master %.*s here, this=%u\n",
2106 		     res->lockname.len, res->lockname.name, dlm->node_num);
2107 	ret = dlm_do_assert_master(dlm, res, nodemap, flags);
2108 	if (ret < 0) {
2109 		/* no need to restart, we are done */
2110 		if (!dlm_is_host_down(ret))
2111 			mlog_errno(ret);
2112 	}
2113 
2114 	/* Ok, we've asserted ourselves.  Let's let migration start. */
2115 	dlm_lockres_release_ast(dlm, res);
2116 
2117 put:
2118 	dlm_lockres_put(res);
2119 
2120 	mlog(0, "finished with dlm_assert_master_worker\n");
2121 }
2122 
2123 /* SPECIAL CASE for the $RECOVERY lock used by the recovery thread.
2124  * We cannot wait for node recovery to complete to begin mastering this
2125  * lockres because this lockres is used to kick off recovery! ;-)
2126  * So, do a pre-check on all living nodes to see if any of those nodes
2127  * think that $RECOVERY is currently mastered by a dead node.  If so,
2128  * we wait a short time to allow that node to get notified by its own
2129  * heartbeat stack, then check again.  All $RECOVERY lock resources
2130  * mastered by dead nodes are purged when the hearbeat callback is
2131  * fired, so we can know for sure that it is safe to continue once
2132  * the node returns a live node or no node.  */
dlm_pre_master_reco_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2133 static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
2134 				       struct dlm_lock_resource *res)
2135 {
2136 	struct dlm_node_iter iter;
2137 	int nodenum;
2138 	int ret = 0;
2139 	u8 master = DLM_LOCK_RES_OWNER_UNKNOWN;
2140 
2141 	spin_lock(&dlm->spinlock);
2142 	dlm_node_iter_init(dlm->domain_map, &iter);
2143 	spin_unlock(&dlm->spinlock);
2144 
2145 	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
2146 		/* do not send to self */
2147 		if (nodenum == dlm->node_num)
2148 			continue;
2149 		ret = dlm_do_master_requery(dlm, res, nodenum, &master);
2150 		if (ret < 0) {
2151 			mlog_errno(ret);
2152 			if (!dlm_is_host_down(ret))
2153 				BUG();
2154 			/* host is down, so answer for that node would be
2155 			 * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
2156 			ret = 0;
2157 		}
2158 
2159 		if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
2160 			/* check to see if this master is in the recovery map */
2161 			spin_lock(&dlm->spinlock);
2162 			if (test_bit(master, dlm->recovery_map)) {
2163 				mlog(ML_NOTICE, "%s: node %u has not seen "
2164 				     "node %u go down yet, and thinks the "
2165 				     "dead node is mastering the recovery "
2166 				     "lock.  must wait.\n", dlm->name,
2167 				     nodenum, master);
2168 				ret = -EAGAIN;
2169 			}
2170 			spin_unlock(&dlm->spinlock);
2171 			mlog(0, "%s: reco lock master is %u\n", dlm->name,
2172 			     master);
2173 			break;
2174 		}
2175 	}
2176 	return ret;
2177 }
2178 
2179 /*
2180  * DLM_DEREF_LOCKRES_MSG
2181  */
2182 
dlm_drop_lockres_ref(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2183 int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
2184 {
2185 	struct dlm_deref_lockres deref;
2186 	int ret = 0, r;
2187 	const char *lockname;
2188 	unsigned int namelen;
2189 
2190 	lockname = res->lockname.name;
2191 	namelen = res->lockname.len;
2192 	BUG_ON(namelen > O2NM_MAX_NAME_LEN);
2193 
2194 	memset(&deref, 0, sizeof(deref));
2195 	deref.node_idx = dlm->node_num;
2196 	deref.namelen = namelen;
2197 	memcpy(deref.name, lockname, namelen);
2198 
2199 	ret = o2net_send_message(DLM_DEREF_LOCKRES_MSG, dlm->key,
2200 				 &deref, sizeof(deref), res->owner, &r);
2201 	if (ret < 0)
2202 		mlog(ML_ERROR, "%s: res %.*s, error %d send DEREF to node %u\n",
2203 		     dlm->name, namelen, lockname, ret, res->owner);
2204 	else if (r < 0) {
2205 		/* BAD.  other node says I did not have a ref. */
2206 		mlog(ML_ERROR, "%s: res %.*s, DEREF to node %u got %d\n",
2207 		     dlm->name, namelen, lockname, res->owner, r);
2208 		dlm_print_one_lock_resource(res);
2209 		BUG();
2210 	}
2211 	return ret;
2212 }
2213 
dlm_deref_lockres_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)2214 int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
2215 			      void **ret_data)
2216 {
2217 	struct dlm_ctxt *dlm = data;
2218 	struct dlm_deref_lockres *deref = (struct dlm_deref_lockres *)msg->buf;
2219 	struct dlm_lock_resource *res = NULL;
2220 	char *name;
2221 	unsigned int namelen;
2222 	int ret = -EINVAL;
2223 	u8 node;
2224 	unsigned int hash;
2225 	struct dlm_work_item *item;
2226 	int cleared = 0;
2227 	int dispatch = 0;
2228 
2229 	if (!dlm_grab(dlm))
2230 		return 0;
2231 
2232 	name = deref->name;
2233 	namelen = deref->namelen;
2234 	node = deref->node_idx;
2235 
2236 	if (namelen > DLM_LOCKID_NAME_MAX) {
2237 		mlog(ML_ERROR, "Invalid name length!");
2238 		goto done;
2239 	}
2240 	if (deref->node_idx >= O2NM_MAX_NODES) {
2241 		mlog(ML_ERROR, "Invalid node number: %u\n", node);
2242 		goto done;
2243 	}
2244 
2245 	hash = dlm_lockid_hash(name, namelen);
2246 
2247 	spin_lock(&dlm->spinlock);
2248 	res = __dlm_lookup_lockres_full(dlm, name, namelen, hash);
2249 	if (!res) {
2250 		spin_unlock(&dlm->spinlock);
2251 		mlog(ML_ERROR, "%s:%.*s: bad lockres name\n",
2252 		     dlm->name, namelen, name);
2253 		goto done;
2254 	}
2255 	spin_unlock(&dlm->spinlock);
2256 
2257 	spin_lock(&res->spinlock);
2258 	if (res->state & DLM_LOCK_RES_SETREF_INPROG)
2259 		dispatch = 1;
2260 	else {
2261 		BUG_ON(res->state & DLM_LOCK_RES_DROPPING_REF);
2262 		if (test_bit(node, res->refmap)) {
2263 			dlm_lockres_clear_refmap_bit(dlm, res, node);
2264 			cleared = 1;
2265 		}
2266 	}
2267 	spin_unlock(&res->spinlock);
2268 
2269 	if (!dispatch) {
2270 		if (cleared)
2271 			dlm_lockres_calc_usage(dlm, res);
2272 		else {
2273 			mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
2274 		     	"but it is already dropped!\n", dlm->name,
2275 		     	res->lockname.len, res->lockname.name, node);
2276 			dlm_print_one_lock_resource(res);
2277 		}
2278 		ret = 0;
2279 		goto done;
2280 	}
2281 
2282 	item = kzalloc(sizeof(*item), GFP_NOFS);
2283 	if (!item) {
2284 		ret = -ENOMEM;
2285 		mlog_errno(ret);
2286 		goto done;
2287 	}
2288 
2289 	dlm_init_work_item(dlm, item, dlm_deref_lockres_worker, NULL);
2290 	item->u.dl.deref_res = res;
2291 	item->u.dl.deref_node = node;
2292 
2293 	spin_lock(&dlm->work_lock);
2294 	list_add_tail(&item->list, &dlm->work_list);
2295 	spin_unlock(&dlm->work_lock);
2296 
2297 	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
2298 	return 0;
2299 
2300 done:
2301 	if (res)
2302 		dlm_lockres_put(res);
2303 	dlm_put(dlm);
2304 
2305 	return ret;
2306 }
2307 
dlm_deref_lockres_worker(struct dlm_work_item * item,void * data)2308 static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data)
2309 {
2310 	struct dlm_ctxt *dlm;
2311 	struct dlm_lock_resource *res;
2312 	u8 node;
2313 	u8 cleared = 0;
2314 
2315 	dlm = item->dlm;
2316 	res = item->u.dl.deref_res;
2317 	node = item->u.dl.deref_node;
2318 
2319 	spin_lock(&res->spinlock);
2320 	BUG_ON(res->state & DLM_LOCK_RES_DROPPING_REF);
2321 	if (test_bit(node, res->refmap)) {
2322 		__dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
2323 		dlm_lockres_clear_refmap_bit(dlm, res, node);
2324 		cleared = 1;
2325 	}
2326 	spin_unlock(&res->spinlock);
2327 
2328 	if (cleared) {
2329 		mlog(0, "%s:%.*s node %u ref dropped in dispatch\n",
2330 		     dlm->name, res->lockname.len, res->lockname.name, node);
2331 		dlm_lockres_calc_usage(dlm, res);
2332 	} else {
2333 		mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
2334 		     "but it is already dropped!\n", dlm->name,
2335 		     res->lockname.len, res->lockname.name, node);
2336 		dlm_print_one_lock_resource(res);
2337 	}
2338 
2339 	dlm_lockres_put(res);
2340 }
2341 
2342 /*
2343  * A migrateable resource is one that is :
2344  * 1. locally mastered, and,
2345  * 2. zero local locks, and,
2346  * 3. one or more non-local locks, or, one or more references
2347  * Returns 1 if yes, 0 if not.
2348  */
dlm_is_lockres_migrateable(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2349 static int dlm_is_lockres_migrateable(struct dlm_ctxt *dlm,
2350 				      struct dlm_lock_resource *res)
2351 {
2352 	enum dlm_lockres_list idx;
2353 	int nonlocal = 0, node_ref;
2354 	struct list_head *queue;
2355 	struct dlm_lock *lock;
2356 	u64 cookie;
2357 
2358 	assert_spin_locked(&res->spinlock);
2359 
2360 	if (res->owner != dlm->node_num)
2361 		return 0;
2362 
2363         for (idx = DLM_GRANTED_LIST; idx <= DLM_BLOCKED_LIST; idx++) {
2364 		queue = dlm_list_idx_to_ptr(res, idx);
2365 		list_for_each_entry(lock, queue, list) {
2366 			if (lock->ml.node != dlm->node_num) {
2367 				nonlocal++;
2368 				continue;
2369 			}
2370 			cookie = be64_to_cpu(lock->ml.cookie);
2371 			mlog(0, "%s: Not migrateable res %.*s, lock %u:%llu on "
2372 			     "%s list\n", dlm->name, res->lockname.len,
2373 			     res->lockname.name,
2374 			     dlm_get_lock_cookie_node(cookie),
2375 			     dlm_get_lock_cookie_seq(cookie),
2376 			     dlm_list_in_text(idx));
2377 			return 0;
2378 		}
2379 	}
2380 
2381 	if (!nonlocal) {
2382 		node_ref = find_next_bit(res->refmap, O2NM_MAX_NODES, 0);
2383 		if (node_ref >= O2NM_MAX_NODES)
2384 			return 0;
2385 	}
2386 
2387 	mlog(0, "%s: res %.*s, Migrateable\n", dlm->name, res->lockname.len,
2388 	     res->lockname.name);
2389 
2390 	return 1;
2391 }
2392 
2393 /*
2394  * DLM_MIGRATE_LOCKRES
2395  */
2396 
2397 
dlm_migrate_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 target)2398 static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
2399 			       struct dlm_lock_resource *res, u8 target)
2400 {
2401 	struct dlm_master_list_entry *mle = NULL;
2402 	struct dlm_master_list_entry *oldmle = NULL;
2403  	struct dlm_migratable_lockres *mres = NULL;
2404 	int ret = 0;
2405 	const char *name;
2406 	unsigned int namelen;
2407 	int mle_added = 0;
2408 	int wake = 0;
2409 
2410 	if (!dlm_grab(dlm))
2411 		return -EINVAL;
2412 
2413 	BUG_ON(target == O2NM_MAX_NODES);
2414 
2415 	name = res->lockname.name;
2416 	namelen = res->lockname.len;
2417 
2418 	mlog(0, "%s: Migrating %.*s to node %u\n", dlm->name, namelen, name,
2419 	     target);
2420 
2421 	/* preallocate up front. if this fails, abort */
2422 	ret = -ENOMEM;
2423 	mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS);
2424 	if (!mres) {
2425 		mlog_errno(ret);
2426 		goto leave;
2427 	}
2428 
2429 	mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
2430 	if (!mle) {
2431 		mlog_errno(ret);
2432 		goto leave;
2433 	}
2434 	ret = 0;
2435 
2436 	/*
2437 	 * clear any existing master requests and
2438 	 * add the migration mle to the list
2439 	 */
2440 	spin_lock(&dlm->spinlock);
2441 	spin_lock(&dlm->master_lock);
2442 	ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name,
2443 				    namelen, target, dlm->node_num);
2444 	spin_unlock(&dlm->master_lock);
2445 	spin_unlock(&dlm->spinlock);
2446 
2447 	if (ret == -EEXIST) {
2448 		mlog(0, "another process is already migrating it\n");
2449 		goto fail;
2450 	}
2451 	mle_added = 1;
2452 
2453 	/*
2454 	 * set the MIGRATING flag and flush asts
2455 	 * if we fail after this we need to re-dirty the lockres
2456 	 */
2457 	if (dlm_mark_lockres_migrating(dlm, res, target) < 0) {
2458 		mlog(ML_ERROR, "tried to migrate %.*s to %u, but "
2459 		     "the target went down.\n", res->lockname.len,
2460 		     res->lockname.name, target);
2461 		spin_lock(&res->spinlock);
2462 		res->state &= ~DLM_LOCK_RES_MIGRATING;
2463 		wake = 1;
2464 		spin_unlock(&res->spinlock);
2465 		ret = -EINVAL;
2466 	}
2467 
2468 fail:
2469 	if (oldmle) {
2470 		/* master is known, detach if not already detached */
2471 		dlm_mle_detach_hb_events(dlm, oldmle);
2472 		dlm_put_mle(oldmle);
2473 	}
2474 
2475 	if (ret < 0) {
2476 		if (mle_added) {
2477 			dlm_mle_detach_hb_events(dlm, mle);
2478 			dlm_put_mle(mle);
2479 		} else if (mle) {
2480 			kmem_cache_free(dlm_mle_cache, mle);
2481 			mle = NULL;
2482 		}
2483 		goto leave;
2484 	}
2485 
2486 	/*
2487 	 * at this point, we have a migration target, an mle
2488 	 * in the master list, and the MIGRATING flag set on
2489 	 * the lockres
2490 	 */
2491 
2492 	/* now that remote nodes are spinning on the MIGRATING flag,
2493 	 * ensure that all assert_master work is flushed. */
2494 	flush_workqueue(dlm->dlm_worker);
2495 
2496 	/* get an extra reference on the mle.
2497 	 * otherwise the assert_master from the new
2498 	 * master will destroy this.
2499 	 * also, make sure that all callers of dlm_get_mle
2500 	 * take both dlm->spinlock and dlm->master_lock */
2501 	spin_lock(&dlm->spinlock);
2502 	spin_lock(&dlm->master_lock);
2503 	dlm_get_mle_inuse(mle);
2504 	spin_unlock(&dlm->master_lock);
2505 	spin_unlock(&dlm->spinlock);
2506 
2507 	/* notify new node and send all lock state */
2508 	/* call send_one_lockres with migration flag.
2509 	 * this serves as notice to the target node that a
2510 	 * migration is starting. */
2511 	ret = dlm_send_one_lockres(dlm, res, mres, target,
2512 				   DLM_MRES_MIGRATION);
2513 
2514 	if (ret < 0) {
2515 		mlog(0, "migration to node %u failed with %d\n",
2516 		     target, ret);
2517 		/* migration failed, detach and clean up mle */
2518 		dlm_mle_detach_hb_events(dlm, mle);
2519 		dlm_put_mle(mle);
2520 		dlm_put_mle_inuse(mle);
2521 		spin_lock(&res->spinlock);
2522 		res->state &= ~DLM_LOCK_RES_MIGRATING;
2523 		wake = 1;
2524 		spin_unlock(&res->spinlock);
2525 		if (dlm_is_host_down(ret))
2526 			dlm_wait_for_node_death(dlm, target,
2527 						DLM_NODE_DEATH_WAIT_MAX);
2528 		goto leave;
2529 	}
2530 
2531 	/* at this point, the target sends a message to all nodes,
2532 	 * (using dlm_do_migrate_request).  this node is skipped since
2533 	 * we had to put an mle in the list to begin the process.  this
2534 	 * node now waits for target to do an assert master.  this node
2535 	 * will be the last one notified, ensuring that the migration
2536 	 * is complete everywhere.  if the target dies while this is
2537 	 * going on, some nodes could potentially see the target as the
2538 	 * master, so it is important that my recovery finds the migration
2539 	 * mle and sets the master to UNKNOWN. */
2540 
2541 
2542 	/* wait for new node to assert master */
2543 	while (1) {
2544 		ret = wait_event_interruptible_timeout(mle->wq,
2545 					(atomic_read(&mle->woken) == 1),
2546 					msecs_to_jiffies(5000));
2547 
2548 		if (ret >= 0) {
2549 		       	if (atomic_read(&mle->woken) == 1 ||
2550 			    res->owner == target)
2551 				break;
2552 
2553 			mlog(0, "%s:%.*s: timed out during migration\n",
2554 			     dlm->name, res->lockname.len, res->lockname.name);
2555 			/* avoid hang during shutdown when migrating lockres
2556 			 * to a node which also goes down */
2557 			if (dlm_is_node_dead(dlm, target)) {
2558 				mlog(0, "%s:%.*s: expected migration "
2559 				     "target %u is no longer up, restarting\n",
2560 				     dlm->name, res->lockname.len,
2561 				     res->lockname.name, target);
2562 				ret = -EINVAL;
2563 				/* migration failed, detach and clean up mle */
2564 				dlm_mle_detach_hb_events(dlm, mle);
2565 				dlm_put_mle(mle);
2566 				dlm_put_mle_inuse(mle);
2567 				spin_lock(&res->spinlock);
2568 				res->state &= ~DLM_LOCK_RES_MIGRATING;
2569 				wake = 1;
2570 				spin_unlock(&res->spinlock);
2571 				goto leave;
2572 			}
2573 		} else
2574 			mlog(0, "%s:%.*s: caught signal during migration\n",
2575 			     dlm->name, res->lockname.len, res->lockname.name);
2576 	}
2577 
2578 	/* all done, set the owner, clear the flag */
2579 	spin_lock(&res->spinlock);
2580 	dlm_set_lockres_owner(dlm, res, target);
2581 	res->state &= ~DLM_LOCK_RES_MIGRATING;
2582 	dlm_remove_nonlocal_locks(dlm, res);
2583 	spin_unlock(&res->spinlock);
2584 	wake_up(&res->wq);
2585 
2586 	/* master is known, detach if not already detached */
2587 	dlm_mle_detach_hb_events(dlm, mle);
2588 	dlm_put_mle_inuse(mle);
2589 	ret = 0;
2590 
2591 	dlm_lockres_calc_usage(dlm, res);
2592 
2593 leave:
2594 	/* re-dirty the lockres if we failed */
2595 	if (ret < 0)
2596 		dlm_kick_thread(dlm, res);
2597 
2598 	/* wake up waiters if the MIGRATING flag got set
2599 	 * but migration failed */
2600 	if (wake)
2601 		wake_up(&res->wq);
2602 
2603 	if (mres)
2604 		free_page((unsigned long)mres);
2605 
2606 	dlm_put(dlm);
2607 
2608 	mlog(0, "%s: Migrating %.*s to %u, returns %d\n", dlm->name, namelen,
2609 	     name, target, ret);
2610 	return ret;
2611 }
2612 
2613 #define DLM_MIGRATION_RETRY_MS  100
2614 
2615 /*
2616  * Should be called only after beginning the domain leave process.
2617  * There should not be any remaining locks on nonlocal lock resources,
2618  * and there should be no local locks left on locally mastered resources.
2619  *
2620  * Called with the dlm spinlock held, may drop it to do migration, but
2621  * will re-acquire before exit.
2622  *
2623  * Returns: 1 if dlm->spinlock was dropped/retaken, 0 if never dropped
2624  */
dlm_empty_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2625 int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
2626 {
2627 	int ret;
2628 	int lock_dropped = 0;
2629 	u8 target = O2NM_MAX_NODES;
2630 
2631 	assert_spin_locked(&dlm->spinlock);
2632 
2633 	spin_lock(&res->spinlock);
2634 	if (dlm_is_lockres_migrateable(dlm, res))
2635 		target = dlm_pick_migration_target(dlm, res);
2636 	spin_unlock(&res->spinlock);
2637 
2638 	if (target == O2NM_MAX_NODES)
2639 		goto leave;
2640 
2641 	/* Wheee! Migrate lockres here! Will sleep so drop spinlock. */
2642 	spin_unlock(&dlm->spinlock);
2643 	lock_dropped = 1;
2644 	ret = dlm_migrate_lockres(dlm, res, target);
2645 	if (ret)
2646 		mlog(0, "%s: res %.*s, Migrate to node %u failed with %d\n",
2647 		     dlm->name, res->lockname.len, res->lockname.name,
2648 		     target, ret);
2649 	spin_lock(&dlm->spinlock);
2650 leave:
2651 	return lock_dropped;
2652 }
2653 
dlm_lock_basts_flushed(struct dlm_ctxt * dlm,struct dlm_lock * lock)2654 int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock)
2655 {
2656 	int ret;
2657 	spin_lock(&dlm->ast_lock);
2658 	spin_lock(&lock->spinlock);
2659 	ret = (list_empty(&lock->bast_list) && !lock->bast_pending);
2660 	spin_unlock(&lock->spinlock);
2661 	spin_unlock(&dlm->ast_lock);
2662 	return ret;
2663 }
2664 
dlm_migration_can_proceed(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 mig_target)2665 static int dlm_migration_can_proceed(struct dlm_ctxt *dlm,
2666 				     struct dlm_lock_resource *res,
2667 				     u8 mig_target)
2668 {
2669 	int can_proceed;
2670 	spin_lock(&res->spinlock);
2671 	can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING);
2672 	spin_unlock(&res->spinlock);
2673 
2674 	/* target has died, so make the caller break out of the
2675 	 * wait_event, but caller must recheck the domain_map */
2676 	spin_lock(&dlm->spinlock);
2677 	if (!test_bit(mig_target, dlm->domain_map))
2678 		can_proceed = 1;
2679 	spin_unlock(&dlm->spinlock);
2680 	return can_proceed;
2681 }
2682 
dlm_lockres_is_dirty(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2683 static int dlm_lockres_is_dirty(struct dlm_ctxt *dlm,
2684 				struct dlm_lock_resource *res)
2685 {
2686 	int ret;
2687 	spin_lock(&res->spinlock);
2688 	ret = !!(res->state & DLM_LOCK_RES_DIRTY);
2689 	spin_unlock(&res->spinlock);
2690 	return ret;
2691 }
2692 
2693 
dlm_mark_lockres_migrating(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 target)2694 static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
2695 				       struct dlm_lock_resource *res,
2696 				       u8 target)
2697 {
2698 	int ret = 0;
2699 
2700 	mlog(0, "dlm_mark_lockres_migrating: %.*s, from %u to %u\n",
2701 	       res->lockname.len, res->lockname.name, dlm->node_num,
2702 	       target);
2703 	/* need to set MIGRATING flag on lockres.  this is done by
2704 	 * ensuring that all asts have been flushed for this lockres. */
2705 	spin_lock(&res->spinlock);
2706 	BUG_ON(res->migration_pending);
2707 	res->migration_pending = 1;
2708 	/* strategy is to reserve an extra ast then release
2709 	 * it below, letting the release do all of the work */
2710 	__dlm_lockres_reserve_ast(res);
2711 	spin_unlock(&res->spinlock);
2712 
2713 	/* now flush all the pending asts */
2714 	dlm_kick_thread(dlm, res);
2715 	/* before waiting on DIRTY, block processes which may
2716 	 * try to dirty the lockres before MIGRATING is set */
2717 	spin_lock(&res->spinlock);
2718 	BUG_ON(res->state & DLM_LOCK_RES_BLOCK_DIRTY);
2719 	res->state |= DLM_LOCK_RES_BLOCK_DIRTY;
2720 	spin_unlock(&res->spinlock);
2721 	/* now wait on any pending asts and the DIRTY state */
2722 	wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
2723 	dlm_lockres_release_ast(dlm, res);
2724 
2725 	mlog(0, "about to wait on migration_wq, dirty=%s\n",
2726 	       res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no");
2727 	/* if the extra ref we just put was the final one, this
2728 	 * will pass thru immediately.  otherwise, we need to wait
2729 	 * for the last ast to finish. */
2730 again:
2731 	ret = wait_event_interruptible_timeout(dlm->migration_wq,
2732 		   dlm_migration_can_proceed(dlm, res, target),
2733 		   msecs_to_jiffies(1000));
2734 	if (ret < 0) {
2735 		mlog(0, "woken again: migrating? %s, dead? %s\n",
2736 		       res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2737 		       test_bit(target, dlm->domain_map) ? "no":"yes");
2738 	} else {
2739 		mlog(0, "all is well: migrating? %s, dead? %s\n",
2740 		       res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
2741 		       test_bit(target, dlm->domain_map) ? "no":"yes");
2742 	}
2743 	if (!dlm_migration_can_proceed(dlm, res, target)) {
2744 		mlog(0, "trying again...\n");
2745 		goto again;
2746 	}
2747 
2748 	ret = 0;
2749 	/* did the target go down or die? */
2750 	spin_lock(&dlm->spinlock);
2751 	if (!test_bit(target, dlm->domain_map)) {
2752 		mlog(ML_ERROR, "aha. migration target %u just went down\n",
2753 		     target);
2754 		ret = -EHOSTDOWN;
2755 	}
2756 	spin_unlock(&dlm->spinlock);
2757 
2758 	/*
2759 	 * if target is down, we need to clear DLM_LOCK_RES_BLOCK_DIRTY for
2760 	 * another try; otherwise, we are sure the MIGRATING state is there,
2761 	 * drop the unneded state which blocked threads trying to DIRTY
2762 	 */
2763 	spin_lock(&res->spinlock);
2764 	BUG_ON(!(res->state & DLM_LOCK_RES_BLOCK_DIRTY));
2765 	res->state &= ~DLM_LOCK_RES_BLOCK_DIRTY;
2766 	if (!ret)
2767 		BUG_ON(!(res->state & DLM_LOCK_RES_MIGRATING));
2768 	spin_unlock(&res->spinlock);
2769 
2770 	/*
2771 	 * at this point:
2772 	 *
2773 	 *   o the DLM_LOCK_RES_MIGRATING flag is set if target not down
2774 	 *   o there are no pending asts on this lockres
2775 	 *   o all processes trying to reserve an ast on this
2776 	 *     lockres must wait for the MIGRATING flag to clear
2777 	 */
2778 	return ret;
2779 }
2780 
2781 /* last step in the migration process.
2782  * original master calls this to free all of the dlm_lock
2783  * structures that used to be for other nodes. */
dlm_remove_nonlocal_locks(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2784 static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
2785 				      struct dlm_lock_resource *res)
2786 {
2787 	struct list_head *queue = &res->granted;
2788 	int i, bit;
2789 	struct dlm_lock *lock, *next;
2790 
2791 	assert_spin_locked(&res->spinlock);
2792 
2793 	BUG_ON(res->owner == dlm->node_num);
2794 
2795 	for (i=0; i<3; i++) {
2796 		list_for_each_entry_safe(lock, next, queue, list) {
2797 			if (lock->ml.node != dlm->node_num) {
2798 				mlog(0, "putting lock for node %u\n",
2799 				     lock->ml.node);
2800 				/* be extra careful */
2801 				BUG_ON(!list_empty(&lock->ast_list));
2802 				BUG_ON(!list_empty(&lock->bast_list));
2803 				BUG_ON(lock->ast_pending);
2804 				BUG_ON(lock->bast_pending);
2805 				dlm_lockres_clear_refmap_bit(dlm, res,
2806 							     lock->ml.node);
2807 				list_del_init(&lock->list);
2808 				dlm_lock_put(lock);
2809 				/* In a normal unlock, we would have added a
2810 				 * DLM_UNLOCK_FREE_LOCK action. Force it. */
2811 				dlm_lock_put(lock);
2812 			}
2813 		}
2814 		queue++;
2815 	}
2816 	bit = 0;
2817 	while (1) {
2818 		bit = find_next_bit(res->refmap, O2NM_MAX_NODES, bit);
2819 		if (bit >= O2NM_MAX_NODES)
2820 			break;
2821 		/* do not clear the local node reference, if there is a
2822 		 * process holding this, let it drop the ref itself */
2823 		if (bit != dlm->node_num) {
2824 			mlog(0, "%s:%.*s: node %u had a ref to this "
2825 			     "migrating lockres, clearing\n", dlm->name,
2826 			     res->lockname.len, res->lockname.name, bit);
2827 			dlm_lockres_clear_refmap_bit(dlm, res, bit);
2828 		}
2829 		bit++;
2830 	}
2831 }
2832 
2833 /*
2834  * Pick a node to migrate the lock resource to. This function selects a
2835  * potential target based first on the locks and then on refmap. It skips
2836  * nodes that are in the process of exiting the domain.
2837  */
dlm_pick_migration_target(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)2838 static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
2839 				    struct dlm_lock_resource *res)
2840 {
2841 	enum dlm_lockres_list idx;
2842 	struct list_head *queue = &res->granted;
2843 	struct dlm_lock *lock;
2844 	int noderef;
2845 	u8 nodenum = O2NM_MAX_NODES;
2846 
2847 	assert_spin_locked(&dlm->spinlock);
2848 	assert_spin_locked(&res->spinlock);
2849 
2850 	/* Go through all the locks */
2851 	for (idx = DLM_GRANTED_LIST; idx <= DLM_BLOCKED_LIST; idx++) {
2852 		queue = dlm_list_idx_to_ptr(res, idx);
2853 		list_for_each_entry(lock, queue, list) {
2854 			if (lock->ml.node == dlm->node_num)
2855 				continue;
2856 			if (test_bit(lock->ml.node, dlm->exit_domain_map))
2857 				continue;
2858 			nodenum = lock->ml.node;
2859 			goto bail;
2860 		}
2861 	}
2862 
2863 	/* Go thru the refmap */
2864 	noderef = -1;
2865 	while (1) {
2866 		noderef = find_next_bit(res->refmap, O2NM_MAX_NODES,
2867 					noderef + 1);
2868 		if (noderef >= O2NM_MAX_NODES)
2869 			break;
2870 		if (noderef == dlm->node_num)
2871 			continue;
2872 		if (test_bit(noderef, dlm->exit_domain_map))
2873 			continue;
2874 		nodenum = noderef;
2875 		goto bail;
2876 	}
2877 
2878 bail:
2879 	return nodenum;
2880 }
2881 
2882 /* this is called by the new master once all lockres
2883  * data has been received */
dlm_do_migrate_request(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 master,u8 new_master,struct dlm_node_iter * iter)2884 static int dlm_do_migrate_request(struct dlm_ctxt *dlm,
2885 				  struct dlm_lock_resource *res,
2886 				  u8 master, u8 new_master,
2887 				  struct dlm_node_iter *iter)
2888 {
2889 	struct dlm_migrate_request migrate;
2890 	int ret, skip, status = 0;
2891 	int nodenum;
2892 
2893 	memset(&migrate, 0, sizeof(migrate));
2894 	migrate.namelen = res->lockname.len;
2895 	memcpy(migrate.name, res->lockname.name, migrate.namelen);
2896 	migrate.new_master = new_master;
2897 	migrate.master = master;
2898 
2899 	ret = 0;
2900 
2901 	/* send message to all nodes, except the master and myself */
2902 	while ((nodenum = dlm_node_iter_next(iter)) >= 0) {
2903 		if (nodenum == master ||
2904 		    nodenum == new_master)
2905 			continue;
2906 
2907 		/* We could race exit domain. If exited, skip. */
2908 		spin_lock(&dlm->spinlock);
2909 		skip = (!test_bit(nodenum, dlm->domain_map));
2910 		spin_unlock(&dlm->spinlock);
2911 		if (skip) {
2912 			clear_bit(nodenum, iter->node_map);
2913 			continue;
2914 		}
2915 
2916 		ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
2917 					 &migrate, sizeof(migrate), nodenum,
2918 					 &status);
2919 		if (ret < 0) {
2920 			mlog(ML_ERROR, "%s: res %.*s, Error %d send "
2921 			     "MIGRATE_REQUEST to node %u\n", dlm->name,
2922 			     migrate.namelen, migrate.name, ret, nodenum);
2923 			if (!dlm_is_host_down(ret)) {
2924 				mlog(ML_ERROR, "unhandled error=%d!\n", ret);
2925 				BUG();
2926 			}
2927 			clear_bit(nodenum, iter->node_map);
2928 			ret = 0;
2929 		} else if (status < 0) {
2930 			mlog(0, "migrate request (node %u) returned %d!\n",
2931 			     nodenum, status);
2932 			ret = status;
2933 		} else if (status == DLM_MIGRATE_RESPONSE_MASTERY_REF) {
2934 			/* during the migration request we short-circuited
2935 			 * the mastery of the lockres.  make sure we have
2936 			 * a mastery ref for nodenum */
2937 			mlog(0, "%s:%.*s: need ref for node %u\n",
2938 			     dlm->name, res->lockname.len, res->lockname.name,
2939 			     nodenum);
2940 			spin_lock(&res->spinlock);
2941 			dlm_lockres_set_refmap_bit(dlm, res, nodenum);
2942 			spin_unlock(&res->spinlock);
2943 		}
2944 	}
2945 
2946 	if (ret < 0)
2947 		mlog_errno(ret);
2948 
2949 	mlog(0, "returning ret=%d\n", ret);
2950 	return ret;
2951 }
2952 
2953 
2954 /* if there is an existing mle for this lockres, we now know who the master is.
2955  * (the one who sent us *this* message) we can clear it up right away.
2956  * since the process that put the mle on the list still has a reference to it,
2957  * we can unhash it now, set the master and wake the process.  as a result,
2958  * we will have no mle in the list to start with.  now we can add an mle for
2959  * the migration and this should be the only one found for those scanning the
2960  * list.  */
dlm_migrate_request_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)2961 int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
2962 				void **ret_data)
2963 {
2964 	struct dlm_ctxt *dlm = data;
2965 	struct dlm_lock_resource *res = NULL;
2966 	struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
2967 	struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
2968 	const char *name;
2969 	unsigned int namelen, hash;
2970 	int ret = 0;
2971 
2972 	if (!dlm_grab(dlm))
2973 		return -EINVAL;
2974 
2975 	name = migrate->name;
2976 	namelen = migrate->namelen;
2977 	hash = dlm_lockid_hash(name, namelen);
2978 
2979 	/* preallocate.. if this fails, abort */
2980 	mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
2981 
2982 	if (!mle) {
2983 		ret = -ENOMEM;
2984 		goto leave;
2985 	}
2986 
2987 	/* check for pre-existing lock */
2988 	spin_lock(&dlm->spinlock);
2989 	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
2990 	if (res) {
2991 		spin_lock(&res->spinlock);
2992 		if (res->state & DLM_LOCK_RES_RECOVERING) {
2993 			/* if all is working ok, this can only mean that we got
2994 		 	* a migrate request from a node that we now see as
2995 		 	* dead.  what can we do here?  drop it to the floor? */
2996 			spin_unlock(&res->spinlock);
2997 			mlog(ML_ERROR, "Got a migrate request, but the "
2998 			     "lockres is marked as recovering!");
2999 			kmem_cache_free(dlm_mle_cache, mle);
3000 			ret = -EINVAL; /* need a better solution */
3001 			goto unlock;
3002 		}
3003 		res->state |= DLM_LOCK_RES_MIGRATING;
3004 		spin_unlock(&res->spinlock);
3005 	}
3006 
3007 	spin_lock(&dlm->master_lock);
3008 	/* ignore status.  only nonzero status would BUG. */
3009 	ret = dlm_add_migration_mle(dlm, res, mle, &oldmle,
3010 				    name, namelen,
3011 				    migrate->new_master,
3012 				    migrate->master);
3013 
3014 	spin_unlock(&dlm->master_lock);
3015 unlock:
3016 	spin_unlock(&dlm->spinlock);
3017 
3018 	if (oldmle) {
3019 		/* master is known, detach if not already detached */
3020 		dlm_mle_detach_hb_events(dlm, oldmle);
3021 		dlm_put_mle(oldmle);
3022 	}
3023 
3024 	if (res)
3025 		dlm_lockres_put(res);
3026 leave:
3027 	dlm_put(dlm);
3028 	return ret;
3029 }
3030 
3031 /* must be holding dlm->spinlock and dlm->master_lock
3032  * when adding a migration mle, we can clear any other mles
3033  * in the master list because we know with certainty that
3034  * the master is "master".  so we remove any old mle from
3035  * the list after setting it's master field, and then add
3036  * the new migration mle.  this way we can hold with the rule
3037  * of having only one mle for a given lock name at all times. */
dlm_add_migration_mle(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_master_list_entry * mle,struct dlm_master_list_entry ** oldmle,const char * name,unsigned int namelen,u8 new_master,u8 master)3038 static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
3039 				 struct dlm_lock_resource *res,
3040 				 struct dlm_master_list_entry *mle,
3041 				 struct dlm_master_list_entry **oldmle,
3042 				 const char *name, unsigned int namelen,
3043 				 u8 new_master, u8 master)
3044 {
3045 	int found;
3046 	int ret = 0;
3047 
3048 	*oldmle = NULL;
3049 
3050 	assert_spin_locked(&dlm->spinlock);
3051 	assert_spin_locked(&dlm->master_lock);
3052 
3053 	/* caller is responsible for any ref taken here on oldmle */
3054 	found = dlm_find_mle(dlm, oldmle, (char *)name, namelen);
3055 	if (found) {
3056 		struct dlm_master_list_entry *tmp = *oldmle;
3057 		spin_lock(&tmp->spinlock);
3058 		if (tmp->type == DLM_MLE_MIGRATION) {
3059 			if (master == dlm->node_num) {
3060 				/* ah another process raced me to it */
3061 				mlog(0, "tried to migrate %.*s, but some "
3062 				     "process beat me to it\n",
3063 				     namelen, name);
3064 				ret = -EEXIST;
3065 			} else {
3066 				/* bad.  2 NODES are trying to migrate! */
3067 				mlog(ML_ERROR, "migration error  mle: "
3068 				     "master=%u new_master=%u // request: "
3069 				     "master=%u new_master=%u // "
3070 				     "lockres=%.*s\n",
3071 				     tmp->master, tmp->new_master,
3072 				     master, new_master,
3073 				     namelen, name);
3074 				BUG();
3075 			}
3076 		} else {
3077 			/* this is essentially what assert_master does */
3078 			tmp->master = master;
3079 			atomic_set(&tmp->woken, 1);
3080 			wake_up(&tmp->wq);
3081 			/* remove it so that only one mle will be found */
3082 			__dlm_unlink_mle(dlm, tmp);
3083 			__dlm_mle_detach_hb_events(dlm, tmp);
3084 			ret = DLM_MIGRATE_RESPONSE_MASTERY_REF;
3085 			mlog(0, "%s:%.*s: master=%u, newmaster=%u, "
3086 			    "telling master to get ref for cleared out mle "
3087 			    "during migration\n", dlm->name, namelen, name,
3088 			    master, new_master);
3089 		}
3090 		spin_unlock(&tmp->spinlock);
3091 	}
3092 
3093 	/* now add a migration mle to the tail of the list */
3094 	dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen);
3095 	mle->new_master = new_master;
3096 	/* the new master will be sending an assert master for this.
3097 	 * at that point we will get the refmap reference */
3098 	mle->master = master;
3099 	/* do this for consistency with other mle types */
3100 	set_bit(new_master, mle->maybe_map);
3101 	__dlm_insert_mle(dlm, mle);
3102 
3103 	return ret;
3104 }
3105 
3106 /*
3107  * Sets the owner of the lockres, associated to the mle, to UNKNOWN
3108  */
dlm_reset_mleres_owner(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)3109 static struct dlm_lock_resource *dlm_reset_mleres_owner(struct dlm_ctxt *dlm,
3110 					struct dlm_master_list_entry *mle)
3111 {
3112 	struct dlm_lock_resource *res;
3113 
3114 	/* Find the lockres associated to the mle and set its owner to UNK */
3115 	res = __dlm_lookup_lockres(dlm, mle->mname, mle->mnamelen,
3116 				   mle->mnamehash);
3117 	if (res) {
3118 		spin_unlock(&dlm->master_lock);
3119 
3120 		/* move lockres onto recovery list */
3121 		spin_lock(&res->spinlock);
3122 		dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
3123 		dlm_move_lockres_to_recovery_list(dlm, res);
3124 		spin_unlock(&res->spinlock);
3125 		dlm_lockres_put(res);
3126 
3127 		/* about to get rid of mle, detach from heartbeat */
3128 		__dlm_mle_detach_hb_events(dlm, mle);
3129 
3130 		/* dump the mle */
3131 		spin_lock(&dlm->master_lock);
3132 		__dlm_put_mle(mle);
3133 		spin_unlock(&dlm->master_lock);
3134 	}
3135 
3136 	return res;
3137 }
3138 
dlm_clean_migration_mle(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle)3139 static void dlm_clean_migration_mle(struct dlm_ctxt *dlm,
3140 				    struct dlm_master_list_entry *mle)
3141 {
3142 	__dlm_mle_detach_hb_events(dlm, mle);
3143 
3144 	spin_lock(&mle->spinlock);
3145 	__dlm_unlink_mle(dlm, mle);
3146 	atomic_set(&mle->woken, 1);
3147 	spin_unlock(&mle->spinlock);
3148 
3149 	wake_up(&mle->wq);
3150 }
3151 
dlm_clean_block_mle(struct dlm_ctxt * dlm,struct dlm_master_list_entry * mle,u8 dead_node)3152 static void dlm_clean_block_mle(struct dlm_ctxt *dlm,
3153 				struct dlm_master_list_entry *mle, u8 dead_node)
3154 {
3155 	int bit;
3156 
3157 	BUG_ON(mle->type != DLM_MLE_BLOCK);
3158 
3159 	spin_lock(&mle->spinlock);
3160 	bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
3161 	if (bit != dead_node) {
3162 		mlog(0, "mle found, but dead node %u would not have been "
3163 		     "master\n", dead_node);
3164 		spin_unlock(&mle->spinlock);
3165 	} else {
3166 		/* Must drop the refcount by one since the assert_master will
3167 		 * never arrive. This may result in the mle being unlinked and
3168 		 * freed, but there may still be a process waiting in the
3169 		 * dlmlock path which is fine. */
3170 		mlog(0, "node %u was expected master\n", dead_node);
3171 		atomic_set(&mle->woken, 1);
3172 		spin_unlock(&mle->spinlock);
3173 		wake_up(&mle->wq);
3174 
3175 		/* Do not need events any longer, so detach from heartbeat */
3176 		__dlm_mle_detach_hb_events(dlm, mle);
3177 		__dlm_put_mle(mle);
3178 	}
3179 }
3180 
dlm_clean_master_list(struct dlm_ctxt * dlm,u8 dead_node)3181 void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
3182 {
3183 	struct dlm_master_list_entry *mle;
3184 	struct dlm_lock_resource *res;
3185 	struct hlist_head *bucket;
3186 	struct hlist_node *list;
3187 	unsigned int i;
3188 
3189 	mlog(0, "dlm=%s, dead node=%u\n", dlm->name, dead_node);
3190 top:
3191 	assert_spin_locked(&dlm->spinlock);
3192 
3193 	/* clean the master list */
3194 	spin_lock(&dlm->master_lock);
3195 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
3196 		bucket = dlm_master_hash(dlm, i);
3197 		hlist_for_each(list, bucket) {
3198 			mle = hlist_entry(list, struct dlm_master_list_entry,
3199 					  master_hash_node);
3200 
3201 			BUG_ON(mle->type != DLM_MLE_BLOCK &&
3202 			       mle->type != DLM_MLE_MASTER &&
3203 			       mle->type != DLM_MLE_MIGRATION);
3204 
3205 			/* MASTER mles are initiated locally. The waiting
3206 			 * process will notice the node map change shortly.
3207 			 * Let that happen as normal. */
3208 			if (mle->type == DLM_MLE_MASTER)
3209 				continue;
3210 
3211 			/* BLOCK mles are initiated by other nodes. Need to
3212 			 * clean up if the dead node would have been the
3213 			 * master. */
3214 			if (mle->type == DLM_MLE_BLOCK) {
3215 				dlm_clean_block_mle(dlm, mle, dead_node);
3216 				continue;
3217 			}
3218 
3219 			/* Everything else is a MIGRATION mle */
3220 
3221 			/* The rule for MIGRATION mles is that the master
3222 			 * becomes UNKNOWN if *either* the original or the new
3223 			 * master dies. All UNKNOWN lockres' are sent to
3224 			 * whichever node becomes the recovery master. The new
3225 			 * master is responsible for determining if there is
3226 			 * still a master for this lockres, or if he needs to
3227 			 * take over mastery. Either way, this node should
3228 			 * expect another message to resolve this. */
3229 
3230 			if (mle->master != dead_node &&
3231 			    mle->new_master != dead_node)
3232 				continue;
3233 
3234 			/* If we have reached this point, this mle needs to be
3235 			 * removed from the list and freed. */
3236 			dlm_clean_migration_mle(dlm, mle);
3237 
3238 			mlog(0, "%s: node %u died during migration from "
3239 			     "%u to %u!\n", dlm->name, dead_node, mle->master,
3240 			     mle->new_master);
3241 
3242 			/* If we find a lockres associated with the mle, we've
3243 			 * hit this rare case that messes up our lock ordering.
3244 			 * If so, we need to drop the master lock so that we can
3245 			 * take the lockres lock, meaning that we will have to
3246 			 * restart from the head of list. */
3247 			res = dlm_reset_mleres_owner(dlm, mle);
3248 			if (res)
3249 				/* restart */
3250 				goto top;
3251 
3252 			/* This may be the last reference */
3253 			__dlm_put_mle(mle);
3254 		}
3255 	}
3256 	spin_unlock(&dlm->master_lock);
3257 }
3258 
dlm_finish_migration(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,u8 old_master)3259 int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
3260 			 u8 old_master)
3261 {
3262 	struct dlm_node_iter iter;
3263 	int ret = 0;
3264 
3265 	spin_lock(&dlm->spinlock);
3266 	dlm_node_iter_init(dlm->domain_map, &iter);
3267 	clear_bit(old_master, iter.node_map);
3268 	clear_bit(dlm->node_num, iter.node_map);
3269 	spin_unlock(&dlm->spinlock);
3270 
3271 	/* ownership of the lockres is changing.  account for the
3272 	 * mastery reference here since old_master will briefly have
3273 	 * a reference after the migration completes */
3274 	spin_lock(&res->spinlock);
3275 	dlm_lockres_set_refmap_bit(dlm, res, old_master);
3276 	spin_unlock(&res->spinlock);
3277 
3278 	mlog(0, "now time to do a migrate request to other nodes\n");
3279 	ret = dlm_do_migrate_request(dlm, res, old_master,
3280 				     dlm->node_num, &iter);
3281 	if (ret < 0) {
3282 		mlog_errno(ret);
3283 		goto leave;
3284 	}
3285 
3286 	mlog(0, "doing assert master of %.*s to all except the original node\n",
3287 	     res->lockname.len, res->lockname.name);
3288 	/* this call now finishes out the nodemap
3289 	 * even if one or more nodes die */
3290 	ret = dlm_do_assert_master(dlm, res, iter.node_map,
3291 				   DLM_ASSERT_MASTER_FINISH_MIGRATION);
3292 	if (ret < 0) {
3293 		/* no longer need to retry.  all living nodes contacted. */
3294 		mlog_errno(ret);
3295 		ret = 0;
3296 	}
3297 
3298 	memset(iter.node_map, 0, sizeof(iter.node_map));
3299 	set_bit(old_master, iter.node_map);
3300 	mlog(0, "doing assert master of %.*s back to %u\n",
3301 	     res->lockname.len, res->lockname.name, old_master);
3302 	ret = dlm_do_assert_master(dlm, res, iter.node_map,
3303 				   DLM_ASSERT_MASTER_FINISH_MIGRATION);
3304 	if (ret < 0) {
3305 		mlog(0, "assert master to original master failed "
3306 		     "with %d.\n", ret);
3307 		/* the only nonzero status here would be because of
3308 		 * a dead original node.  we're done. */
3309 		ret = 0;
3310 	}
3311 
3312 	/* all done, set the owner, clear the flag */
3313 	spin_lock(&res->spinlock);
3314 	dlm_set_lockres_owner(dlm, res, dlm->node_num);
3315 	res->state &= ~DLM_LOCK_RES_MIGRATING;
3316 	spin_unlock(&res->spinlock);
3317 	/* re-dirty it on the new master */
3318 	dlm_kick_thread(dlm, res);
3319 	wake_up(&res->wq);
3320 leave:
3321 	return ret;
3322 }
3323 
3324 /*
3325  * LOCKRES AST REFCOUNT
3326  * this is integral to migration
3327  */
3328 
3329 /* for future intent to call an ast, reserve one ahead of time.
3330  * this should be called only after waiting on the lockres
3331  * with dlm_wait_on_lockres, and while still holding the
3332  * spinlock after the call. */
__dlm_lockres_reserve_ast(struct dlm_lock_resource * res)3333 void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res)
3334 {
3335 	assert_spin_locked(&res->spinlock);
3336 	if (res->state & DLM_LOCK_RES_MIGRATING) {
3337 		__dlm_print_one_lock_resource(res);
3338 	}
3339 	BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
3340 
3341 	atomic_inc(&res->asts_reserved);
3342 }
3343 
3344 /*
3345  * used to drop the reserved ast, either because it went unused,
3346  * or because the ast/bast was actually called.
3347  *
3348  * also, if there is a pending migration on this lockres,
3349  * and this was the last pending ast on the lockres,
3350  * atomically set the MIGRATING flag before we drop the lock.
3351  * this is how we ensure that migration can proceed with no
3352  * asts in progress.  note that it is ok if the state of the
3353  * queues is such that a lock should be granted in the future
3354  * or that a bast should be fired, because the new master will
3355  * shuffle the lists on this lockres as soon as it is migrated.
3356  */
dlm_lockres_release_ast(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)3357 void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
3358 			     struct dlm_lock_resource *res)
3359 {
3360 	if (!atomic_dec_and_lock(&res->asts_reserved, &res->spinlock))
3361 		return;
3362 
3363 	if (!res->migration_pending) {
3364 		spin_unlock(&res->spinlock);
3365 		return;
3366 	}
3367 
3368 	BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
3369 	res->migration_pending = 0;
3370 	res->state |= DLM_LOCK_RES_MIGRATING;
3371 	spin_unlock(&res->spinlock);
3372 	wake_up(&res->wq);
3373 	wake_up(&dlm->migration_wq);
3374 }
3375 
dlm_force_free_mles(struct dlm_ctxt * dlm)3376 void dlm_force_free_mles(struct dlm_ctxt *dlm)
3377 {
3378 	int i;
3379 	struct hlist_head *bucket;
3380 	struct dlm_master_list_entry *mle;
3381 	struct hlist_node *tmp, *list;
3382 
3383 	/*
3384 	 * We notified all other nodes that we are exiting the domain and
3385 	 * marked the dlm state to DLM_CTXT_LEAVING. If any mles are still
3386 	 * around we force free them and wake any processes that are waiting
3387 	 * on the mles
3388 	 */
3389 	spin_lock(&dlm->spinlock);
3390 	spin_lock(&dlm->master_lock);
3391 
3392 	BUG_ON(dlm->dlm_state != DLM_CTXT_LEAVING);
3393 	BUG_ON((find_next_bit(dlm->domain_map, O2NM_MAX_NODES, 0) < O2NM_MAX_NODES));
3394 
3395 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
3396 		bucket = dlm_master_hash(dlm, i);
3397 		hlist_for_each_safe(list, tmp, bucket) {
3398 			mle = hlist_entry(list, struct dlm_master_list_entry,
3399 					  master_hash_node);
3400 			if (mle->type != DLM_MLE_BLOCK) {
3401 				mlog(ML_ERROR, "bad mle: %p\n", mle);
3402 				dlm_print_one_mle(mle);
3403 			}
3404 			atomic_set(&mle->woken, 1);
3405 			wake_up(&mle->wq);
3406 
3407 			__dlm_unlink_mle(dlm, mle);
3408 			__dlm_mle_detach_hb_events(dlm, mle);
3409 			__dlm_put_mle(mle);
3410 		}
3411 	}
3412 	spin_unlock(&dlm->master_lock);
3413 	spin_unlock(&dlm->spinlock);
3414 }
3415