xref: /illumos-gate/usr/src/uts/common/fs/zfs/txg.c (revision fa9e4066f08beec538e775443c5be79dd423fcab)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/zfs_context.h>
30 #include <sys/txg_impl.h>
31 #include <sys/dmu_impl.h>
32 #include <sys/dsl_pool.h>
33 #include <sys/callb.h>
34 
35 /*
36  * Pool-wide transaction groups.
37  */
38 
39 static void txg_sync_thread(dsl_pool_t *dp);
40 static void txg_quiesce_thread(dsl_pool_t *dp);
41 static void txg_timelimit_thread(dsl_pool_t *dp);
42 
43 int txg_time = 5;	/* max 5 seconds worth of delta per txg */
44 
45 /*
46  * Prepare the txg subsystem.
47  */
48 void
49 txg_init(dsl_pool_t *dp, uint64_t txg)
50 {
51 	tx_state_t *tx = &dp->dp_tx;
52 
53 	bzero(tx, sizeof (tx_state_t));
54 
55 	tx->tx_cpu = kmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP);
56 
57 	rw_init(&tx->tx_suspend, NULL, RW_DEFAULT, NULL);
58 
59 	tx->tx_open_txg = txg;
60 }
61 
62 /*
63  * Close down the txg subsystem.
64  */
65 void
66 txg_fini(dsl_pool_t *dp)
67 {
68 	tx_state_t *tx = &dp->dp_tx;
69 
70 	ASSERT(tx->tx_threads == 0);
71 
72 	rw_destroy(&tx->tx_suspend);
73 
74 	kmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t));
75 
76 	bzero(tx, sizeof (tx_state_t));
77 }
78 
79 /*
80  * Start syncing transaction groups.
81  */
82 void
83 txg_sync_start(dsl_pool_t *dp)
84 {
85 	tx_state_t *tx = &dp->dp_tx;
86 
87 	mutex_enter(&tx->tx_sync_lock);
88 
89 	dprintf("pool %p\n", dp);
90 
91 	ASSERT(tx->tx_threads == 0);
92 
93 	tx->tx_threads = 3;
94 
95 	tx->tx_quiesce_thread = thread_create(NULL, 0, txg_quiesce_thread,
96 	    dp, 0, &p0, TS_RUN, minclsyspri);
97 
98 	tx->tx_sync_thread = thread_create(NULL, 0, txg_sync_thread,
99 	    dp, 0, &p0, TS_RUN, minclsyspri);
100 
101 	tx->tx_timelimit_thread = thread_create(NULL, 0, txg_timelimit_thread,
102 	    dp, 0, &p0, TS_RUN, minclsyspri);
103 
104 	mutex_exit(&tx->tx_sync_lock);
105 }
106 
107 static void
108 txg_thread_enter(tx_state_t *tx, callb_cpr_t *cpr)
109 {
110 	CALLB_CPR_INIT(cpr, &tx->tx_sync_lock, callb_generic_cpr, FTAG);
111 	mutex_enter(&tx->tx_sync_lock);
112 }
113 
114 static void
115 txg_thread_exit(tx_state_t *tx, callb_cpr_t *cpr, kthread_t **tpp)
116 {
117 	ASSERT(*tpp != NULL);
118 	*tpp = NULL;
119 	tx->tx_threads--;
120 	cv_broadcast(&tx->tx_exit_cv);
121 	CALLB_CPR_EXIT(cpr);		/* drops &tx->tx_sync_lock */
122 	thread_exit();
123 }
124 
125 static void
126 txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, int secmax)
127 {
128 	CALLB_CPR_SAFE_BEGIN(cpr);
129 
130 	if (secmax)
131 		(void) cv_timedwait(cv, &tx->tx_sync_lock, lbolt + secmax * hz);
132 	else
133 		cv_wait(cv, &tx->tx_sync_lock);
134 
135 	CALLB_CPR_SAFE_END(cpr, &tx->tx_sync_lock);
136 }
137 
138 /*
139  * Stop syncing transaction groups.
140  */
141 void
142 txg_sync_stop(dsl_pool_t *dp)
143 {
144 	tx_state_t *tx = &dp->dp_tx;
145 
146 	dprintf("pool %p\n", dp);
147 	/*
148 	 * Finish off any work in progress.
149 	 */
150 	ASSERT(tx->tx_threads == 3);
151 	txg_wait_synced(dp, 0);
152 
153 	/*
154 	 * Wake all 3 sync threads (one per state) and wait for them to die.
155 	 */
156 	mutex_enter(&tx->tx_sync_lock);
157 
158 	ASSERT(tx->tx_threads == 3);
159 
160 	tx->tx_exiting = 1;
161 
162 	cv_broadcast(&tx->tx_quiesce_more_cv);
163 	cv_broadcast(&tx->tx_quiesce_done_cv);
164 	cv_broadcast(&tx->tx_sync_more_cv);
165 	cv_broadcast(&tx->tx_timeout_exit_cv);
166 
167 	while (tx->tx_threads != 0)
168 		cv_wait(&tx->tx_exit_cv, &tx->tx_sync_lock);
169 
170 	tx->tx_exiting = 0;
171 
172 	mutex_exit(&tx->tx_sync_lock);
173 }
174 
175 uint64_t
176 txg_hold_open(dsl_pool_t *dp, txg_handle_t *th)
177 {
178 	tx_state_t *tx = &dp->dp_tx;
179 	tx_cpu_t *tc = &tx->tx_cpu[CPU_SEQID];
180 	uint64_t txg;
181 
182 	mutex_enter(&tc->tc_lock);
183 
184 	txg = tx->tx_open_txg;
185 	tc->tc_count[txg & TXG_MASK]++;
186 
187 	th->th_cpu = tc;
188 	th->th_txg = txg;
189 
190 	return (txg);
191 }
192 
193 void
194 txg_rele_to_quiesce(txg_handle_t *th)
195 {
196 	tx_cpu_t *tc = th->th_cpu;
197 
198 	mutex_exit(&tc->tc_lock);
199 }
200 
201 void
202 txg_rele_to_sync(txg_handle_t *th)
203 {
204 	tx_cpu_t *tc = th->th_cpu;
205 	int g = th->th_txg & TXG_MASK;
206 
207 	mutex_enter(&tc->tc_lock);
208 	ASSERT(tc->tc_count[g] != 0);
209 	if (--tc->tc_count[g] == 0)
210 		cv_broadcast(&tc->tc_cv[g]);
211 	mutex_exit(&tc->tc_lock);
212 
213 	th->th_cpu = NULL;	/* defensive */
214 }
215 
216 static void
217 txg_quiesce(dsl_pool_t *dp, uint64_t txg)
218 {
219 	tx_state_t *tx = &dp->dp_tx;
220 	int g = txg & TXG_MASK;
221 	int c;
222 
223 	/*
224 	 * Grab all tx_cpu locks so nobody else can get into this txg.
225 	 */
226 	for (c = 0; c < max_ncpus; c++)
227 		mutex_enter(&tx->tx_cpu[c].tc_lock);
228 
229 	ASSERT(txg == tx->tx_open_txg);
230 	tx->tx_open_txg++;
231 
232 	/*
233 	 * Now that we've incremented tx_open_txg, we can let threads
234 	 * enter the next transaction group.
235 	 */
236 	for (c = 0; c < max_ncpus; c++)
237 		mutex_exit(&tx->tx_cpu[c].tc_lock);
238 
239 	/*
240 	 * Quiesce the transaction group by waiting for everyone to txg_exit().
241 	 */
242 	for (c = 0; c < max_ncpus; c++) {
243 		tx_cpu_t *tc = &tx->tx_cpu[c];
244 		mutex_enter(&tc->tc_lock);
245 		while (tc->tc_count[g] != 0)
246 			cv_wait(&tc->tc_cv[g], &tc->tc_lock);
247 		mutex_exit(&tc->tc_lock);
248 	}
249 }
250 
251 static void
252 txg_sync_thread(dsl_pool_t *dp)
253 {
254 	tx_state_t *tx = &dp->dp_tx;
255 	callb_cpr_t cpr;
256 
257 	txg_thread_enter(tx, &cpr);
258 
259 	for (;;) {
260 		uint64_t txg;
261 
262 		/*
263 		 * We sync when there's someone waiting on us, or the
264 		 * quiesce thread has handed off a txg to us.
265 		 */
266 		while (!tx->tx_exiting &&
267 		    tx->tx_synced_txg >= tx->tx_sync_txg_waiting &&
268 		    tx->tx_quiesced_txg == 0) {
269 			dprintf("waiting; tx_synced=%llu waiting=%llu dp=%p\n",
270 			    tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp);
271 			txg_thread_wait(tx, &cpr, &tx->tx_sync_more_cv, 0);
272 		}
273 
274 		/*
275 		 * Wait until the quiesce thread hands off a txg to us,
276 		 * prompting it to do so if necessary.
277 		 */
278 		while (!tx->tx_exiting && tx->tx_quiesced_txg == 0) {
279 			if (tx->tx_quiesce_txg_waiting < tx->tx_open_txg+1)
280 				tx->tx_quiesce_txg_waiting = tx->tx_open_txg+1;
281 			cv_broadcast(&tx->tx_quiesce_more_cv);
282 			txg_thread_wait(tx, &cpr, &tx->tx_quiesce_done_cv, 0);
283 		}
284 
285 		if (tx->tx_exiting)
286 			txg_thread_exit(tx, &cpr, &tx->tx_sync_thread);
287 
288 		rw_enter(&tx->tx_suspend, RW_WRITER);
289 
290 		/*
291 		 * Consume the quiesced txg which has been handed off to
292 		 * us.  This may cause the quiescing thread to now be
293 		 * able to quiesce another txg, so we must signal it.
294 		 */
295 		txg = tx->tx_quiesced_txg;
296 		tx->tx_quiesced_txg = 0;
297 		tx->tx_syncing_txg = txg;
298 		cv_broadcast(&tx->tx_quiesce_more_cv);
299 		rw_exit(&tx->tx_suspend);
300 
301 		dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
302 			txg, tx->tx_quiesce_txg_waiting,
303 			tx->tx_sync_txg_waiting);
304 		mutex_exit(&tx->tx_sync_lock);
305 		spa_sync(dp->dp_spa, txg);
306 		mutex_enter(&tx->tx_sync_lock);
307 		rw_enter(&tx->tx_suspend, RW_WRITER);
308 		tx->tx_synced_txg = txg;
309 		tx->tx_syncing_txg = 0;
310 		rw_exit(&tx->tx_suspend);
311 		cv_broadcast(&tx->tx_sync_done_cv);
312 	}
313 }
314 
315 static void
316 txg_quiesce_thread(dsl_pool_t *dp)
317 {
318 	tx_state_t *tx = &dp->dp_tx;
319 	callb_cpr_t cpr;
320 
321 	txg_thread_enter(tx, &cpr);
322 
323 	for (;;) {
324 		uint64_t txg;
325 
326 		/*
327 		 * We quiesce when there's someone waiting on us.
328 		 * However, we can only have one txg in "quiescing" or
329 		 * "quiesced, waiting to sync" state.  So we wait until
330 		 * the "quiesced, waiting to sync" txg has been consumed
331 		 * by the sync thread.
332 		 */
333 		while (!tx->tx_exiting &&
334 		    (tx->tx_open_txg >= tx->tx_quiesce_txg_waiting ||
335 		    tx->tx_quiesced_txg != 0))
336 			txg_thread_wait(tx, &cpr, &tx->tx_quiesce_more_cv, 0);
337 
338 		if (tx->tx_exiting)
339 			txg_thread_exit(tx, &cpr, &tx->tx_quiesce_thread);
340 
341 		txg = tx->tx_open_txg;
342 		dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
343 		    txg, tx->tx_quiesce_txg_waiting,
344 		    tx->tx_sync_txg_waiting);
345 		mutex_exit(&tx->tx_sync_lock);
346 		txg_quiesce(dp, txg);
347 		mutex_enter(&tx->tx_sync_lock);
348 
349 		/*
350 		 * Hand this txg off to the sync thread.
351 		 */
352 		dprintf("quiesce done, handing off txg %llu\n", txg);
353 		tx->tx_quiesced_txg = txg;
354 		cv_broadcast(&tx->tx_sync_more_cv);
355 		cv_broadcast(&tx->tx_quiesce_done_cv);
356 	}
357 }
358 
359 void
360 txg_wait_synced(dsl_pool_t *dp, uint64_t txg)
361 {
362 	tx_state_t *tx = &dp->dp_tx;
363 
364 	mutex_enter(&tx->tx_sync_lock);
365 	ASSERT(tx->tx_threads == 3);
366 	if (txg == 0)
367 		txg = tx->tx_open_txg;
368 	if (tx->tx_sync_txg_waiting < txg)
369 		tx->tx_sync_txg_waiting = txg;
370 	dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
371 	    txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting);
372 	while (tx->tx_synced_txg < txg) {
373 		dprintf("broadcasting sync more "
374 		    "tx_synced=%llu waiting=%llu dp=%p\n",
375 		    tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp);
376 		cv_broadcast(&tx->tx_sync_more_cv);
377 		cv_wait(&tx->tx_sync_done_cv, &tx->tx_sync_lock);
378 	}
379 	mutex_exit(&tx->tx_sync_lock);
380 }
381 
382 void
383 txg_wait_open(dsl_pool_t *dp, uint64_t txg)
384 {
385 	tx_state_t *tx = &dp->dp_tx;
386 
387 	mutex_enter(&tx->tx_sync_lock);
388 	ASSERT(tx->tx_threads == 3);
389 	if (txg == 0)
390 		txg = tx->tx_open_txg + 1;
391 	if (tx->tx_quiesce_txg_waiting < txg)
392 		tx->tx_quiesce_txg_waiting = txg;
393 	dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
394 	    txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting);
395 	while (tx->tx_open_txg < txg) {
396 		cv_broadcast(&tx->tx_quiesce_more_cv);
397 		cv_wait(&tx->tx_quiesce_done_cv, &tx->tx_sync_lock);
398 	}
399 	mutex_exit(&tx->tx_sync_lock);
400 }
401 
402 static void
403 txg_timelimit_thread(dsl_pool_t *dp)
404 {
405 	tx_state_t *tx = &dp->dp_tx;
406 	callb_cpr_t cpr;
407 
408 	txg_thread_enter(tx, &cpr);
409 
410 	while (!tx->tx_exiting) {
411 		uint64_t txg = tx->tx_open_txg + 1;
412 
413 		txg_thread_wait(tx, &cpr, &tx->tx_timeout_exit_cv, txg_time);
414 
415 		if (tx->tx_quiesce_txg_waiting < txg)
416 			tx->tx_quiesce_txg_waiting = txg;
417 
418 		while (!tx->tx_exiting && tx->tx_open_txg < txg) {
419 			dprintf("pushing out %llu\n", txg);
420 			cv_broadcast(&tx->tx_quiesce_more_cv);
421 			txg_thread_wait(tx, &cpr, &tx->tx_quiesce_done_cv, 0);
422 		}
423 	}
424 	txg_thread_exit(tx, &cpr, &tx->tx_timelimit_thread);
425 }
426 
427 int
428 txg_stalled(dsl_pool_t *dp)
429 {
430 	tx_state_t *tx = &dp->dp_tx;
431 	return (tx->tx_quiesce_txg_waiting > tx->tx_open_txg);
432 }
433 
434 void
435 txg_suspend(dsl_pool_t *dp)
436 {
437 	tx_state_t *tx = &dp->dp_tx;
438 	/* XXX some code paths suspend when they are already suspended! */
439 	rw_enter(&tx->tx_suspend, RW_READER);
440 }
441 
442 void
443 txg_resume(dsl_pool_t *dp)
444 {
445 	tx_state_t *tx = &dp->dp_tx;
446 	rw_exit(&tx->tx_suspend);
447 }
448 
449 /*
450  * Per-txg object lists.
451  */
452 void
453 txg_list_create(txg_list_t *tl, size_t offset)
454 {
455 	int t;
456 
457 	mutex_init(&tl->tl_lock, NULL, MUTEX_DEFAULT, NULL);
458 
459 	tl->tl_offset = offset;
460 
461 	for (t = 0; t < TXG_SIZE; t++)
462 		tl->tl_head[t] = NULL;
463 }
464 
465 void
466 txg_list_destroy(txg_list_t *tl)
467 {
468 	int t;
469 
470 	for (t = 0; t < TXG_SIZE; t++)
471 		ASSERT(txg_list_empty(tl, t));
472 
473 	mutex_destroy(&tl->tl_lock);
474 }
475 
476 int
477 txg_list_empty(txg_list_t *tl, uint64_t txg)
478 {
479 	return (tl->tl_head[txg & TXG_MASK] == NULL);
480 }
481 
482 /*
483  * Add an entry to the list.
484  * Returns 0 if it's a new entry, 1 if it's already there.
485  */
486 int
487 txg_list_add(txg_list_t *tl, void *p, uint64_t txg)
488 {
489 	int t = txg & TXG_MASK;
490 	txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset);
491 	int already_on_list;
492 
493 	mutex_enter(&tl->tl_lock);
494 	already_on_list = tn->tn_member[t];
495 	if (!already_on_list) {
496 		tn->tn_member[t] = 1;
497 		tn->tn_next[t] = tl->tl_head[t];
498 		tl->tl_head[t] = tn;
499 	}
500 	mutex_exit(&tl->tl_lock);
501 
502 	return (already_on_list);
503 }
504 
505 /*
506  * Remove the head of the list and return it.
507  */
508 void *
509 txg_list_remove(txg_list_t *tl, uint64_t txg)
510 {
511 	int t = txg & TXG_MASK;
512 	txg_node_t *tn;
513 	void *p = NULL;
514 
515 	mutex_enter(&tl->tl_lock);
516 	if ((tn = tl->tl_head[t]) != NULL) {
517 		p = (char *)tn - tl->tl_offset;
518 		tl->tl_head[t] = tn->tn_next[t];
519 		tn->tn_next[t] = NULL;
520 		tn->tn_member[t] = 0;
521 	}
522 	mutex_exit(&tl->tl_lock);
523 
524 	return (p);
525 }
526 
527 /*
528  * Remove a specific item from the list and return it.
529  */
530 void *
531 txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg)
532 {
533 	int t = txg & TXG_MASK;
534 	txg_node_t *tn, **tp;
535 
536 	mutex_enter(&tl->tl_lock);
537 
538 	for (tp = &tl->tl_head[t]; (tn = *tp) != NULL; tp = &tn->tn_next[t]) {
539 		if ((char *)tn - tl->tl_offset == p) {
540 			*tp = tn->tn_next[t];
541 			tn->tn_next[t] = NULL;
542 			tn->tn_member[t] = 0;
543 			mutex_exit(&tl->tl_lock);
544 			return (p);
545 		}
546 	}
547 
548 	mutex_exit(&tl->tl_lock);
549 
550 	return (NULL);
551 }
552 
553 int
554 txg_list_member(txg_list_t *tl, void *p, uint64_t txg)
555 {
556 	int t = txg & TXG_MASK;
557 	txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset);
558 
559 	return (tn->tn_member[t]);
560 }
561 
562 /*
563  * Walk a txg list -- only safe if you know it's not changing.
564  */
565 void *
566 txg_list_head(txg_list_t *tl, uint64_t txg)
567 {
568 	int t = txg & TXG_MASK;
569 	txg_node_t *tn = tl->tl_head[t];
570 
571 	return (tn == NULL ? NULL : (char *)tn - tl->tl_offset);
572 }
573 
574 void *
575 txg_list_next(txg_list_t *tl, void *p, uint64_t txg)
576 {
577 	int t = txg & TXG_MASK;
578 	txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset);
579 
580 	tn = tn->tn_next[t];
581 
582 	return (tn == NULL ? NULL : (char *)tn - tl->tl_offset);
583 }
584