1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * dlmthread.c
4  *
5  * standalone DLM module
6  *
7  * Copyright (C) 2004 Oracle.  All rights reserved.
8  */
9 
10 
11 #include <linux/module.h>
12 #include <linux/fs.h>
13 #include <linux/types.h>
14 #include <linux/highmem.h>
15 #include <linux/init.h>
16 #include <linux/sysctl.h>
17 #include <linux/random.h>
18 #include <linux/blkdev.h>
19 #include <linux/socket.h>
20 #include <linux/inet.h>
21 #include <linux/timer.h>
22 #include <linux/kthread.h>
23 #include <linux/delay.h>
24 
25 
26 #include "../cluster/heartbeat.h"
27 #include "../cluster/nodemanager.h"
28 #include "../cluster/tcp.h"
29 
30 #include "dlmapi.h"
31 #include "dlmcommon.h"
32 #include "dlmdomain.h"
33 
34 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
35 #include "../cluster/masklog.h"
36 
37 static int dlm_thread(void *data);
38 static void dlm_flush_asts(struct dlm_ctxt *dlm);
39 
40 /* will exit holding res->spinlock, but may drop in function */
41 /* waits until flags are cleared on res->state */
__dlm_wait_on_lockres_flags(struct dlm_lock_resource * res,int flags)42 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
43 {
44 	DECLARE_WAITQUEUE(wait, current);
45 
46 	assert_spin_locked(&res->spinlock);
47 
48 	add_wait_queue(&res->wq, &wait);
49 repeat:
50 	set_current_state(TASK_UNINTERRUPTIBLE);
51 	if (res->state & flags) {
52 		spin_unlock(&res->spinlock);
53 		schedule();
54 		spin_lock(&res->spinlock);
55 		goto repeat;
56 	}
57 	remove_wait_queue(&res->wq, &wait);
58 	__set_current_state(TASK_RUNNING);
59 }
60 
__dlm_lockres_has_locks(struct dlm_lock_resource * res)61 int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
62 {
63 	if (list_empty(&res->granted) &&
64 	    list_empty(&res->converting) &&
65 	    list_empty(&res->blocked))
66 		return 0;
67 	return 1;
68 }
69 
70 /* "unused": the lockres has no locks, is not on the dirty list,
71  * has no inflight locks (in the gap between mastery and acquiring
72  * the first lock), and has no bits in its refmap.
73  * truly ready to be freed. */
__dlm_lockres_unused(struct dlm_lock_resource * res)74 int __dlm_lockres_unused(struct dlm_lock_resource *res)
75 {
76 	int bit;
77 
78 	assert_spin_locked(&res->spinlock);
79 
80 	if (__dlm_lockres_has_locks(res))
81 		return 0;
82 
83 	/* Locks are in the process of being created */
84 	if (res->inflight_locks)
85 		return 0;
86 
87 	if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY)
88 		return 0;
89 
90 	if (res->state & (DLM_LOCK_RES_RECOVERING|
91 			DLM_LOCK_RES_RECOVERY_WAITING))
92 		return 0;
93 
94 	/* Another node has this resource with this node as the master */
95 	bit = find_first_bit(res->refmap, O2NM_MAX_NODES);
96 	if (bit < O2NM_MAX_NODES)
97 		return 0;
98 
99 	return 1;
100 }
101 
102 
103 /* Call whenever you may have added or deleted something from one of
104  * the lockres queue's. This will figure out whether it belongs on the
105  * unused list or not and does the appropriate thing. */
__dlm_lockres_calc_usage(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)106 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
107 			      struct dlm_lock_resource *res)
108 {
109 	assert_spin_locked(&dlm->spinlock);
110 	assert_spin_locked(&res->spinlock);
111 
112 	if (__dlm_lockres_unused(res)){
113 		if (list_empty(&res->purge)) {
114 			mlog(0, "%s: Adding res %.*s to purge list\n",
115 			     dlm->name, res->lockname.len, res->lockname.name);
116 
117 			res->last_used = jiffies;
118 			dlm_lockres_get(res);
119 			list_add_tail(&res->purge, &dlm->purge_list);
120 			dlm->purge_count++;
121 		}
122 	} else if (!list_empty(&res->purge)) {
123 		mlog(0, "%s: Removing res %.*s from purge list\n",
124 		     dlm->name, res->lockname.len, res->lockname.name);
125 
126 		list_del_init(&res->purge);
127 		dlm_lockres_put(res);
128 		dlm->purge_count--;
129 	}
130 }
131 
dlm_lockres_calc_usage(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)132 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
133 			    struct dlm_lock_resource *res)
134 {
135 	spin_lock(&dlm->spinlock);
136 	spin_lock(&res->spinlock);
137 
138 	__dlm_lockres_calc_usage(dlm, res);
139 
140 	spin_unlock(&res->spinlock);
141 	spin_unlock(&dlm->spinlock);
142 }
143 
144 /*
145  * Do the real purge work:
146  *     unhash the lockres, and
147  *     clear flag DLM_LOCK_RES_DROPPING_REF.
148  * It requires dlm and lockres spinlock to be taken.
149  */
__dlm_do_purge_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)150 void __dlm_do_purge_lockres(struct dlm_ctxt *dlm,
151 		struct dlm_lock_resource *res)
152 {
153 	assert_spin_locked(&dlm->spinlock);
154 	assert_spin_locked(&res->spinlock);
155 
156 	if (!list_empty(&res->purge)) {
157 		mlog(0, "%s: Removing res %.*s from purgelist\n",
158 		     dlm->name, res->lockname.len, res->lockname.name);
159 		list_del_init(&res->purge);
160 		dlm_lockres_put(res);
161 		dlm->purge_count--;
162 	}
163 
164 	if (!__dlm_lockres_unused(res)) {
165 		mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
166 		     dlm->name, res->lockname.len, res->lockname.name);
167 		__dlm_print_one_lock_resource(res);
168 		BUG();
169 	}
170 
171 	__dlm_unhash_lockres(dlm, res);
172 
173 	spin_lock(&dlm->track_lock);
174 	if (!list_empty(&res->tracking))
175 		list_del_init(&res->tracking);
176 	else {
177 		mlog(ML_ERROR, "%s: Resource %.*s not on the Tracking list\n",
178 		     dlm->name, res->lockname.len, res->lockname.name);
179 		__dlm_print_one_lock_resource(res);
180 	}
181 	spin_unlock(&dlm->track_lock);
182 
183 	/*
184 	 * lockres is not in the hash now. drop the flag and wake up
185 	 * any processes waiting in dlm_get_lock_resource.
186 	 */
187 	res->state &= ~DLM_LOCK_RES_DROPPING_REF;
188 }
189 
dlm_purge_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)190 static void dlm_purge_lockres(struct dlm_ctxt *dlm,
191 			     struct dlm_lock_resource *res)
192 {
193 	int master;
194 	int ret = 0;
195 
196 	assert_spin_locked(&dlm->spinlock);
197 	assert_spin_locked(&res->spinlock);
198 
199 	master = (res->owner == dlm->node_num);
200 
201 	mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name,
202 	     res->lockname.len, res->lockname.name, master);
203 
204 	if (!master) {
205 		if (res->state & DLM_LOCK_RES_DROPPING_REF) {
206 			mlog(ML_NOTICE, "%s: res %.*s already in DLM_LOCK_RES_DROPPING_REF state\n",
207 				dlm->name, res->lockname.len, res->lockname.name);
208 			spin_unlock(&res->spinlock);
209 			return;
210 		}
211 
212 		res->state |= DLM_LOCK_RES_DROPPING_REF;
213 		/* drop spinlock...  retake below */
214 		spin_unlock(&res->spinlock);
215 		spin_unlock(&dlm->spinlock);
216 
217 		spin_lock(&res->spinlock);
218 		/* This ensures that clear refmap is sent after the set */
219 		__dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
220 		spin_unlock(&res->spinlock);
221 
222 		/* clear our bit from the master's refmap, ignore errors */
223 		ret = dlm_drop_lockres_ref(dlm, res);
224 		if (ret < 0) {
225 			if (!dlm_is_host_down(ret))
226 				BUG();
227 		}
228 		spin_lock(&dlm->spinlock);
229 		spin_lock(&res->spinlock);
230 	}
231 
232 	if (!list_empty(&res->purge)) {
233 		mlog(0, "%s: Removing res %.*s from purgelist, master %d\n",
234 		     dlm->name, res->lockname.len, res->lockname.name, master);
235 		list_del_init(&res->purge);
236 		dlm_lockres_put(res);
237 		dlm->purge_count--;
238 	}
239 
240 	if (!master && ret == DLM_DEREF_RESPONSE_INPROG) {
241 		mlog(0, "%s: deref %.*s in progress\n",
242 			dlm->name, res->lockname.len, res->lockname.name);
243 		spin_unlock(&res->spinlock);
244 		return;
245 	}
246 
247 	if (!__dlm_lockres_unused(res)) {
248 		mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
249 		     dlm->name, res->lockname.len, res->lockname.name);
250 		__dlm_print_one_lock_resource(res);
251 		BUG();
252 	}
253 
254 	__dlm_unhash_lockres(dlm, res);
255 
256 	spin_lock(&dlm->track_lock);
257 	if (!list_empty(&res->tracking))
258 		list_del_init(&res->tracking);
259 	else {
260 		mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n",
261 				res->lockname.len, res->lockname.name);
262 		__dlm_print_one_lock_resource(res);
263 	}
264 	spin_unlock(&dlm->track_lock);
265 
266 	/* lockres is not in the hash now.  drop the flag and wake up
267 	 * any processes waiting in dlm_get_lock_resource. */
268 	if (!master) {
269 		res->state &= ~DLM_LOCK_RES_DROPPING_REF;
270 		spin_unlock(&res->spinlock);
271 		wake_up(&res->wq);
272 	} else
273 		spin_unlock(&res->spinlock);
274 }
275 
dlm_run_purge_list(struct dlm_ctxt * dlm,int purge_now)276 static void dlm_run_purge_list(struct dlm_ctxt *dlm,
277 			       int purge_now)
278 {
279 	unsigned int run_max, unused;
280 	unsigned long purge_jiffies;
281 	struct dlm_lock_resource *lockres;
282 
283 	spin_lock(&dlm->spinlock);
284 	run_max = dlm->purge_count;
285 
286 	while(run_max && !list_empty(&dlm->purge_list)) {
287 		run_max--;
288 
289 		lockres = list_entry(dlm->purge_list.next,
290 				     struct dlm_lock_resource, purge);
291 
292 		spin_lock(&lockres->spinlock);
293 
294 		purge_jiffies = lockres->last_used +
295 			msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
296 
297 		/* Make sure that we want to be processing this guy at
298 		 * this time. */
299 		if (!purge_now && time_after(purge_jiffies, jiffies)) {
300 			/* Since resources are added to the purge list
301 			 * in tail order, we can stop at the first
302 			 * unpurgable resource -- anyone added after
303 			 * him will have a greater last_used value */
304 			spin_unlock(&lockres->spinlock);
305 			break;
306 		}
307 
308 		/* Status of the lockres *might* change so double
309 		 * check. If the lockres is unused, holding the dlm
310 		 * spinlock will prevent people from getting and more
311 		 * refs on it. */
312 		unused = __dlm_lockres_unused(lockres);
313 		if (!unused ||
314 		    (lockres->state & DLM_LOCK_RES_MIGRATING) ||
315 		    (lockres->inflight_assert_workers != 0)) {
316 			mlog(0, "%s: res %.*s is in use or being remastered, "
317 			     "used %d, state %d, assert master workers %u\n",
318 			     dlm->name, lockres->lockname.len,
319 			     lockres->lockname.name,
320 			     !unused, lockres->state,
321 			     lockres->inflight_assert_workers);
322 			list_move_tail(&lockres->purge, &dlm->purge_list);
323 			spin_unlock(&lockres->spinlock);
324 			continue;
325 		}
326 
327 		dlm_lockres_get(lockres);
328 
329 		dlm_purge_lockres(dlm, lockres);
330 
331 		dlm_lockres_put(lockres);
332 
333 		/* Avoid adding any scheduling latencies */
334 		cond_resched_lock(&dlm->spinlock);
335 	}
336 
337 	spin_unlock(&dlm->spinlock);
338 }
339 
dlm_shuffle_lists(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)340 static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
341 			      struct dlm_lock_resource *res)
342 {
343 	struct dlm_lock *lock, *target;
344 	int can_grant = 1;
345 
346 	/*
347 	 * Because this function is called with the lockres
348 	 * spinlock, and because we know that it is not migrating/
349 	 * recovering/in-progress, it is fine to reserve asts and
350 	 * basts right before queueing them all throughout
351 	 */
352 	assert_spin_locked(&dlm->ast_lock);
353 	assert_spin_locked(&res->spinlock);
354 	BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
355 			      DLM_LOCK_RES_RECOVERING|
356 			      DLM_LOCK_RES_IN_PROGRESS)));
357 
358 converting:
359 	if (list_empty(&res->converting))
360 		goto blocked;
361 	mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name,
362 	     res->lockname.len, res->lockname.name);
363 
364 	target = list_entry(res->converting.next, struct dlm_lock, list);
365 	if (target->ml.convert_type == LKM_IVMODE) {
366 		mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n",
367 		     dlm->name, res->lockname.len, res->lockname.name);
368 		BUG();
369 	}
370 	list_for_each_entry(lock, &res->granted, list) {
371 		if (lock==target)
372 			continue;
373 		if (!dlm_lock_compatible(lock->ml.type,
374 					 target->ml.convert_type)) {
375 			can_grant = 0;
376 			/* queue the BAST if not already */
377 			if (lock->ml.highest_blocked == LKM_IVMODE) {
378 				__dlm_lockres_reserve_ast(res);
379 				__dlm_queue_bast(dlm, lock);
380 			}
381 			/* update the highest_blocked if needed */
382 			if (lock->ml.highest_blocked < target->ml.convert_type)
383 				lock->ml.highest_blocked =
384 					target->ml.convert_type;
385 		}
386 	}
387 
388 	list_for_each_entry(lock, &res->converting, list) {
389 		if (lock==target)
390 			continue;
391 		if (!dlm_lock_compatible(lock->ml.type,
392 					 target->ml.convert_type)) {
393 			can_grant = 0;
394 			if (lock->ml.highest_blocked == LKM_IVMODE) {
395 				__dlm_lockres_reserve_ast(res);
396 				__dlm_queue_bast(dlm, lock);
397 			}
398 			if (lock->ml.highest_blocked < target->ml.convert_type)
399 				lock->ml.highest_blocked =
400 					target->ml.convert_type;
401 		}
402 	}
403 
404 	/* we can convert the lock */
405 	if (can_grant) {
406 		spin_lock(&target->spinlock);
407 		BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
408 
409 		mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type "
410 		     "%d => %d, node %u\n", dlm->name, res->lockname.len,
411 		     res->lockname.name,
412 		     dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
413 		     dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
414 		     target->ml.type,
415 		     target->ml.convert_type, target->ml.node);
416 
417 		target->ml.type = target->ml.convert_type;
418 		target->ml.convert_type = LKM_IVMODE;
419 		list_move_tail(&target->list, &res->granted);
420 
421 		BUG_ON(!target->lksb);
422 		target->lksb->status = DLM_NORMAL;
423 
424 		spin_unlock(&target->spinlock);
425 
426 		__dlm_lockres_reserve_ast(res);
427 		__dlm_queue_ast(dlm, target);
428 		/* go back and check for more */
429 		goto converting;
430 	}
431 
432 blocked:
433 	if (list_empty(&res->blocked))
434 		goto leave;
435 	target = list_entry(res->blocked.next, struct dlm_lock, list);
436 
437 	list_for_each_entry(lock, &res->granted, list) {
438 		if (lock==target)
439 			continue;
440 		if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
441 			can_grant = 0;
442 			if (lock->ml.highest_blocked == LKM_IVMODE) {
443 				__dlm_lockres_reserve_ast(res);
444 				__dlm_queue_bast(dlm, lock);
445 			}
446 			if (lock->ml.highest_blocked < target->ml.type)
447 				lock->ml.highest_blocked = target->ml.type;
448 		}
449 	}
450 
451 	list_for_each_entry(lock, &res->converting, list) {
452 		if (lock==target)
453 			continue;
454 		if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
455 			can_grant = 0;
456 			if (lock->ml.highest_blocked == LKM_IVMODE) {
457 				__dlm_lockres_reserve_ast(res);
458 				__dlm_queue_bast(dlm, lock);
459 			}
460 			if (lock->ml.highest_blocked < target->ml.type)
461 				lock->ml.highest_blocked = target->ml.type;
462 		}
463 	}
464 
465 	/* we can grant the blocked lock (only
466 	 * possible if converting list empty) */
467 	if (can_grant) {
468 		spin_lock(&target->spinlock);
469 		BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
470 
471 		mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, "
472 		     "node %u\n", dlm->name, res->lockname.len,
473 		     res->lockname.name,
474 		     dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
475 		     dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
476 		     target->ml.type, target->ml.node);
477 
478 		/* target->ml.type is already correct */
479 		list_move_tail(&target->list, &res->granted);
480 
481 		BUG_ON(!target->lksb);
482 		target->lksb->status = DLM_NORMAL;
483 
484 		spin_unlock(&target->spinlock);
485 
486 		__dlm_lockres_reserve_ast(res);
487 		__dlm_queue_ast(dlm, target);
488 		/* go back and check for more */
489 		goto converting;
490 	}
491 
492 leave:
493 	return;
494 }
495 
496 /* must have NO locks when calling this with res !=NULL * */
dlm_kick_thread(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)497 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
498 {
499 	if (res) {
500 		spin_lock(&dlm->spinlock);
501 		spin_lock(&res->spinlock);
502 		__dlm_dirty_lockres(dlm, res);
503 		spin_unlock(&res->spinlock);
504 		spin_unlock(&dlm->spinlock);
505 	}
506 	wake_up(&dlm->dlm_thread_wq);
507 }
508 
__dlm_dirty_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)509 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
510 {
511 	assert_spin_locked(&dlm->spinlock);
512 	assert_spin_locked(&res->spinlock);
513 
514 	/* don't shuffle secondary queues */
515 	if (res->owner == dlm->node_num) {
516 		if (res->state & (DLM_LOCK_RES_MIGRATING |
517 				  DLM_LOCK_RES_BLOCK_DIRTY))
518 		    return;
519 
520 		if (list_empty(&res->dirty)) {
521 			/* ref for dirty_list */
522 			dlm_lockres_get(res);
523 			list_add_tail(&res->dirty, &dlm->dirty_list);
524 			res->state |= DLM_LOCK_RES_DIRTY;
525 		}
526 	}
527 
528 	mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
529 	     res->lockname.name);
530 }
531 
532 
533 /* Launch the NM thread for the mounted volume */
dlm_launch_thread(struct dlm_ctxt * dlm)534 int dlm_launch_thread(struct dlm_ctxt *dlm)
535 {
536 	mlog(0, "Starting dlm_thread...\n");
537 
538 	dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm-%s",
539 			dlm->name);
540 	if (IS_ERR(dlm->dlm_thread_task)) {
541 		mlog_errno(PTR_ERR(dlm->dlm_thread_task));
542 		dlm->dlm_thread_task = NULL;
543 		return -EINVAL;
544 	}
545 
546 	return 0;
547 }
548 
dlm_complete_thread(struct dlm_ctxt * dlm)549 void dlm_complete_thread(struct dlm_ctxt *dlm)
550 {
551 	if (dlm->dlm_thread_task) {
552 		mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n");
553 		kthread_stop(dlm->dlm_thread_task);
554 		dlm->dlm_thread_task = NULL;
555 	}
556 }
557 
dlm_dirty_list_empty(struct dlm_ctxt * dlm)558 static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
559 {
560 	int empty;
561 
562 	spin_lock(&dlm->spinlock);
563 	empty = list_empty(&dlm->dirty_list);
564 	spin_unlock(&dlm->spinlock);
565 
566 	return empty;
567 }
568 
dlm_flush_asts(struct dlm_ctxt * dlm)569 static void dlm_flush_asts(struct dlm_ctxt *dlm)
570 {
571 	int ret;
572 	struct dlm_lock *lock;
573 	struct dlm_lock_resource *res;
574 	u8 hi;
575 
576 	spin_lock(&dlm->ast_lock);
577 	while (!list_empty(&dlm->pending_asts)) {
578 		lock = list_entry(dlm->pending_asts.next,
579 				  struct dlm_lock, ast_list);
580 		/* get an extra ref on lock */
581 		dlm_lock_get(lock);
582 		res = lock->lockres;
583 		mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, "
584 		     "node %u\n", dlm->name, res->lockname.len,
585 		     res->lockname.name,
586 		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
587 		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
588 		     lock->ml.type, lock->ml.node);
589 
590 		BUG_ON(!lock->ast_pending);
591 
592 		/* remove from list (including ref) */
593 		list_del_init(&lock->ast_list);
594 		dlm_lock_put(lock);
595 		spin_unlock(&dlm->ast_lock);
596 
597 		if (lock->ml.node != dlm->node_num) {
598 			ret = dlm_do_remote_ast(dlm, res, lock);
599 			if (ret < 0)
600 				mlog_errno(ret);
601 		} else
602 			dlm_do_local_ast(dlm, res, lock);
603 
604 		spin_lock(&dlm->ast_lock);
605 
606 		/* possible that another ast was queued while
607 		 * we were delivering the last one */
608 		if (!list_empty(&lock->ast_list)) {
609 			mlog(0, "%s: res %.*s, AST queued while flushing last "
610 			     "one\n", dlm->name, res->lockname.len,
611 			     res->lockname.name);
612 		} else
613 			lock->ast_pending = 0;
614 
615 		/* drop the extra ref.
616 		 * this may drop it completely. */
617 		dlm_lock_put(lock);
618 		dlm_lockres_release_ast(dlm, res);
619 	}
620 
621 	while (!list_empty(&dlm->pending_basts)) {
622 		lock = list_entry(dlm->pending_basts.next,
623 				  struct dlm_lock, bast_list);
624 		/* get an extra ref on lock */
625 		dlm_lock_get(lock);
626 		res = lock->lockres;
627 
628 		BUG_ON(!lock->bast_pending);
629 
630 		/* get the highest blocked lock, and reset */
631 		spin_lock(&lock->spinlock);
632 		BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
633 		hi = lock->ml.highest_blocked;
634 		lock->ml.highest_blocked = LKM_IVMODE;
635 		spin_unlock(&lock->spinlock);
636 
637 		/* remove from list (including ref) */
638 		list_del_init(&lock->bast_list);
639 		dlm_lock_put(lock);
640 		spin_unlock(&dlm->ast_lock);
641 
642 		mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, "
643 		     "blocked %d, node %u\n",
644 		     dlm->name, res->lockname.len, res->lockname.name,
645 		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
646 		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
647 		     hi, lock->ml.node);
648 
649 		if (lock->ml.node != dlm->node_num) {
650 			ret = dlm_send_proxy_bast(dlm, res, lock, hi);
651 			if (ret < 0)
652 				mlog_errno(ret);
653 		} else
654 			dlm_do_local_bast(dlm, res, lock, hi);
655 
656 		spin_lock(&dlm->ast_lock);
657 
658 		/* possible that another bast was queued while
659 		 * we were delivering the last one */
660 		if (!list_empty(&lock->bast_list)) {
661 			mlog(0, "%s: res %.*s, BAST queued while flushing last "
662 			     "one\n", dlm->name, res->lockname.len,
663 			     res->lockname.name);
664 		} else
665 			lock->bast_pending = 0;
666 
667 		/* drop the extra ref.
668 		 * this may drop it completely. */
669 		dlm_lock_put(lock);
670 		dlm_lockres_release_ast(dlm, res);
671 	}
672 	wake_up(&dlm->ast_wq);
673 	spin_unlock(&dlm->ast_lock);
674 }
675 
676 
677 #define DLM_THREAD_TIMEOUT_MS (4 * 1000)
678 #define DLM_THREAD_MAX_DIRTY  100
679 
dlm_thread(void * data)680 static int dlm_thread(void *data)
681 {
682 	struct dlm_lock_resource *res;
683 	struct dlm_ctxt *dlm = data;
684 	unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
685 
686 	mlog(0, "dlm thread running for %s...\n", dlm->name);
687 
688 	while (!kthread_should_stop()) {
689 		int n = DLM_THREAD_MAX_DIRTY;
690 
691 		/* dlm_shutting_down is very point-in-time, but that
692 		 * doesn't matter as we'll just loop back around if we
693 		 * get false on the leading edge of a state
694 		 * transition. */
695 		dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
696 
697 		/* We really don't want to hold dlm->spinlock while
698 		 * calling dlm_shuffle_lists on each lockres that
699 		 * needs to have its queues adjusted and AST/BASTs
700 		 * run.  So let's pull each entry off the dirty_list
701 		 * and drop dlm->spinlock ASAP.  Once off the list,
702 		 * res->spinlock needs to be taken again to protect
703 		 * the queues while calling dlm_shuffle_lists.  */
704 		spin_lock(&dlm->spinlock);
705 		while (!list_empty(&dlm->dirty_list)) {
706 			int delay = 0;
707 			res = list_entry(dlm->dirty_list.next,
708 					 struct dlm_lock_resource, dirty);
709 
710 			/* peel a lockres off, remove it from the list,
711 			 * unset the dirty flag and drop the dlm lock */
712 			BUG_ON(!res);
713 			dlm_lockres_get(res);
714 
715 			spin_lock(&res->spinlock);
716 			/* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */
717 			list_del_init(&res->dirty);
718 			spin_unlock(&res->spinlock);
719 			spin_unlock(&dlm->spinlock);
720 			/* Drop dirty_list ref */
721 			dlm_lockres_put(res);
722 
723 		 	/* lockres can be re-dirtied/re-added to the
724 			 * dirty_list in this gap, but that is ok */
725 
726 			spin_lock(&dlm->ast_lock);
727 			spin_lock(&res->spinlock);
728 			if (res->owner != dlm->node_num) {
729 				__dlm_print_one_lock_resource(res);
730 				mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d,"
731 				     " dirty %d\n", dlm->name,
732 				     !!(res->state & DLM_LOCK_RES_IN_PROGRESS),
733 				     !!(res->state & DLM_LOCK_RES_MIGRATING),
734 				     !!(res->state & DLM_LOCK_RES_RECOVERING),
735 				     !!(res->state & DLM_LOCK_RES_DIRTY));
736 			}
737 			BUG_ON(res->owner != dlm->node_num);
738 
739 			/* it is now ok to move lockreses in these states
740 			 * to the dirty list, assuming that they will only be
741 			 * dirty for a short while. */
742 			BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
743 			if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
744 					  DLM_LOCK_RES_RECOVERING |
745 					  DLM_LOCK_RES_RECOVERY_WAITING)) {
746 				/* move it to the tail and keep going */
747 				res->state &= ~DLM_LOCK_RES_DIRTY;
748 				spin_unlock(&res->spinlock);
749 				spin_unlock(&dlm->ast_lock);
750 				mlog(0, "%s: res %.*s, inprogress, delay list "
751 				     "shuffle, state %d\n", dlm->name,
752 				     res->lockname.len, res->lockname.name,
753 				     res->state);
754 				delay = 1;
755 				goto in_progress;
756 			}
757 
758 			/* at this point the lockres is not migrating/
759 			 * recovering/in-progress.  we have the lockres
760 			 * spinlock and do NOT have the dlm lock.
761 			 * safe to reserve/queue asts and run the lists. */
762 
763 			/* called while holding lockres lock */
764 			dlm_shuffle_lists(dlm, res);
765 			res->state &= ~DLM_LOCK_RES_DIRTY;
766 			spin_unlock(&res->spinlock);
767 			spin_unlock(&dlm->ast_lock);
768 
769 			dlm_lockres_calc_usage(dlm, res);
770 
771 in_progress:
772 
773 			spin_lock(&dlm->spinlock);
774 			/* if the lock was in-progress, stick
775 			 * it on the back of the list */
776 			if (delay) {
777 				spin_lock(&res->spinlock);
778 				__dlm_dirty_lockres(dlm, res);
779 				spin_unlock(&res->spinlock);
780 			}
781 			dlm_lockres_put(res);
782 
783 			/* unlikely, but we may need to give time to
784 			 * other tasks */
785 			if (!--n) {
786 				mlog(0, "%s: Throttling dlm thread\n",
787 				     dlm->name);
788 				break;
789 			}
790 		}
791 
792 		spin_unlock(&dlm->spinlock);
793 		dlm_flush_asts(dlm);
794 
795 		/* yield and continue right away if there is more work to do */
796 		if (!n) {
797 			cond_resched();
798 			continue;
799 		}
800 
801 		wait_event_interruptible_timeout(dlm->dlm_thread_wq,
802 						 !dlm_dirty_list_empty(dlm) ||
803 						 kthread_should_stop(),
804 						 timeout);
805 	}
806 
807 	mlog(0, "quitting DLM thread\n");
808 	return 0;
809 }
810