]>
Commit | Line | Data |
---|---|---|
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 (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | */ | |
24 | ||
25 | #include <sys/zfs_context.h> | |
26 | #include <sys/txg_impl.h> | |
27 | #include <sys/dmu_impl.h> | |
28 | #include <sys/dmu_tx.h> | |
29 | #include <sys/dsl_pool.h> | |
30 | #include <sys/dsl_scan.h> | |
31 | #include <sys/callb.h> | |
32 | ||
33 | /* | |
34 | * Pool-wide transaction groups. | |
35 | */ | |
36 | ||
37 | static void txg_sync_thread(dsl_pool_t *dp); | |
38 | static void txg_quiesce_thread(dsl_pool_t *dp); | |
39 | ||
40 | int zfs_txg_timeout = 5; /* max seconds worth of delta per txg */ | |
41 | ||
42 | /* | |
43 | * Prepare the txg subsystem. | |
44 | */ | |
45 | void | |
46 | txg_init(dsl_pool_t *dp, uint64_t txg) | |
47 | { | |
48 | tx_state_t *tx = &dp->dp_tx; | |
49 | int c; | |
50 | bzero(tx, sizeof (tx_state_t)); | |
51 | ||
52 | tx->tx_cpu = vmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP); | |
53 | ||
54 | for (c = 0; c < max_ncpus; c++) { | |
55 | int i; | |
56 | ||
57 | mutex_init(&tx->tx_cpu[c].tc_lock, NULL, MUTEX_DEFAULT, NULL); | |
58 | for (i = 0; i < TXG_SIZE; i++) { | |
59 | cv_init(&tx->tx_cpu[c].tc_cv[i], NULL, CV_DEFAULT, | |
60 | NULL); | |
61 | list_create(&tx->tx_cpu[c].tc_callbacks[i], | |
62 | sizeof (dmu_tx_callback_t), | |
63 | offsetof(dmu_tx_callback_t, dcb_node)); | |
64 | } | |
65 | } | |
66 | ||
67 | mutex_init(&tx->tx_sync_lock, NULL, MUTEX_DEFAULT, NULL); | |
68 | ||
69 | cv_init(&tx->tx_sync_more_cv, NULL, CV_DEFAULT, NULL); | |
70 | cv_init(&tx->tx_sync_done_cv, NULL, CV_DEFAULT, NULL); | |
71 | cv_init(&tx->tx_quiesce_more_cv, NULL, CV_DEFAULT, NULL); | |
72 | cv_init(&tx->tx_quiesce_done_cv, NULL, CV_DEFAULT, NULL); | |
73 | cv_init(&tx->tx_exit_cv, NULL, CV_DEFAULT, NULL); | |
74 | ||
75 | tx->tx_open_txg = txg; | |
76 | } | |
77 | ||
78 | /* | |
79 | * Close down the txg subsystem. | |
80 | */ | |
81 | void | |
82 | txg_fini(dsl_pool_t *dp) | |
83 | { | |
84 | tx_state_t *tx = &dp->dp_tx; | |
85 | int c; | |
86 | ||
87 | ASSERT(tx->tx_threads == 0); | |
88 | ||
89 | mutex_destroy(&tx->tx_sync_lock); | |
90 | ||
91 | cv_destroy(&tx->tx_sync_more_cv); | |
92 | cv_destroy(&tx->tx_sync_done_cv); | |
93 | cv_destroy(&tx->tx_quiesce_more_cv); | |
94 | cv_destroy(&tx->tx_quiesce_done_cv); | |
95 | cv_destroy(&tx->tx_exit_cv); | |
96 | ||
97 | for (c = 0; c < max_ncpus; c++) { | |
98 | int i; | |
99 | ||
100 | mutex_destroy(&tx->tx_cpu[c].tc_lock); | |
101 | for (i = 0; i < TXG_SIZE; i++) { | |
102 | cv_destroy(&tx->tx_cpu[c].tc_cv[i]); | |
103 | list_destroy(&tx->tx_cpu[c].tc_callbacks[i]); | |
104 | } | |
105 | } | |
106 | ||
107 | if (tx->tx_commit_cb_taskq != NULL) | |
108 | taskq_destroy(tx->tx_commit_cb_taskq); | |
109 | ||
110 | vmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t)); | |
111 | ||
112 | bzero(tx, sizeof (tx_state_t)); | |
113 | } | |
114 | ||
115 | /* | |
116 | * Start syncing transaction groups. | |
117 | */ | |
118 | void | |
119 | txg_sync_start(dsl_pool_t *dp) | |
120 | { | |
121 | tx_state_t *tx = &dp->dp_tx; | |
122 | ||
123 | mutex_enter(&tx->tx_sync_lock); | |
124 | ||
125 | dprintf("pool %p\n", dp); | |
126 | ||
127 | ASSERT(tx->tx_threads == 0); | |
128 | ||
129 | tx->tx_threads = 2; | |
130 | ||
131 | tx->tx_quiesce_thread = thread_create(NULL, 0, txg_quiesce_thread, | |
132 | dp, 0, &p0, TS_RUN, minclsyspri); | |
133 | ||
134 | /* | |
135 | * The sync thread can need a larger-than-default stack size on | |
136 | * 32-bit x86. This is due in part to nested pools and | |
137 | * scrub_visitbp() recursion. | |
138 | */ | |
139 | tx->tx_sync_thread = thread_create(NULL, 32<<10, txg_sync_thread, | |
140 | dp, 0, &p0, TS_RUN, minclsyspri); | |
141 | ||
142 | mutex_exit(&tx->tx_sync_lock); | |
143 | } | |
144 | ||
145 | static void | |
146 | txg_thread_enter(tx_state_t *tx, callb_cpr_t *cpr) | |
147 | { | |
148 | CALLB_CPR_INIT(cpr, &tx->tx_sync_lock, callb_generic_cpr, FTAG); | |
149 | mutex_enter(&tx->tx_sync_lock); | |
150 | } | |
151 | ||
152 | static void | |
153 | txg_thread_exit(tx_state_t *tx, callb_cpr_t *cpr, kthread_t **tpp) | |
154 | { | |
155 | ASSERT(*tpp != NULL); | |
156 | *tpp = NULL; | |
157 | tx->tx_threads--; | |
158 | cv_broadcast(&tx->tx_exit_cv); | |
159 | CALLB_CPR_EXIT(cpr); /* drops &tx->tx_sync_lock */ | |
160 | thread_exit(); | |
161 | } | |
162 | ||
163 | static void | |
164 | txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time) | |
165 | { | |
166 | CALLB_CPR_SAFE_BEGIN(cpr); | |
167 | ||
168 | if (time) | |
169 | (void) cv_timedwait_interruptible(cv, &tx->tx_sync_lock, | |
170 | ddi_get_lbolt() + time); | |
171 | else | |
172 | cv_wait_interruptible(cv, &tx->tx_sync_lock); | |
173 | ||
174 | CALLB_CPR_SAFE_END(cpr, &tx->tx_sync_lock); | |
175 | } | |
176 | ||
177 | /* | |
178 | * Stop syncing transaction groups. | |
179 | */ | |
180 | void | |
181 | txg_sync_stop(dsl_pool_t *dp) | |
182 | { | |
183 | tx_state_t *tx = &dp->dp_tx; | |
184 | ||
185 | dprintf("pool %p\n", dp); | |
186 | /* | |
187 | * Finish off any work in progress. | |
188 | */ | |
189 | ASSERT(tx->tx_threads == 2); | |
190 | ||
191 | /* | |
192 | * We need to ensure that we've vacated the deferred space_maps. | |
193 | */ | |
194 | txg_wait_synced(dp, tx->tx_open_txg + TXG_DEFER_SIZE); | |
195 | ||
196 | /* | |
197 | * Wake all sync threads and wait for them to die. | |
198 | */ | |
199 | mutex_enter(&tx->tx_sync_lock); | |
200 | ||
201 | ASSERT(tx->tx_threads == 2); | |
202 | ||
203 | tx->tx_exiting = 1; | |
204 | ||
205 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
206 | cv_broadcast(&tx->tx_quiesce_done_cv); | |
207 | cv_broadcast(&tx->tx_sync_more_cv); | |
208 | ||
209 | while (tx->tx_threads != 0) | |
210 | cv_wait(&tx->tx_exit_cv, &tx->tx_sync_lock); | |
211 | ||
212 | tx->tx_exiting = 0; | |
213 | ||
214 | mutex_exit(&tx->tx_sync_lock); | |
215 | } | |
216 | ||
217 | uint64_t | |
218 | txg_hold_open(dsl_pool_t *dp, txg_handle_t *th) | |
219 | { | |
220 | tx_state_t *tx = &dp->dp_tx; | |
221 | tx_cpu_t *tc; | |
222 | uint64_t txg; | |
223 | ||
224 | /* | |
225 | * It appears the processor id is simply used as a "random" | |
226 | * number to index into the array, and there isn't any other | |
227 | * significance to the chosen tx_cpu. Because.. Why not use | |
228 | * the current cpu to index into the array? | |
229 | */ | |
230 | kpreempt_disable(); | |
231 | tc = &tx->tx_cpu[CPU_SEQID]; | |
232 | kpreempt_enable(); | |
233 | ||
234 | mutex_enter(&tc->tc_lock); | |
235 | ||
236 | txg = tx->tx_open_txg; | |
237 | tc->tc_count[txg & TXG_MASK]++; | |
238 | ||
239 | th->th_cpu = tc; | |
240 | th->th_txg = txg; | |
241 | ||
242 | return (txg); | |
243 | } | |
244 | ||
245 | void | |
246 | txg_rele_to_quiesce(txg_handle_t *th) | |
247 | { | |
248 | tx_cpu_t *tc = th->th_cpu; | |
249 | ||
250 | mutex_exit(&tc->tc_lock); | |
251 | } | |
252 | ||
253 | void | |
254 | txg_register_callbacks(txg_handle_t *th, list_t *tx_callbacks) | |
255 | { | |
256 | tx_cpu_t *tc = th->th_cpu; | |
257 | int g = th->th_txg & TXG_MASK; | |
258 | ||
259 | mutex_enter(&tc->tc_lock); | |
260 | list_move_tail(&tc->tc_callbacks[g], tx_callbacks); | |
261 | mutex_exit(&tc->tc_lock); | |
262 | } | |
263 | ||
264 | void | |
265 | txg_rele_to_sync(txg_handle_t *th) | |
266 | { | |
267 | tx_cpu_t *tc = th->th_cpu; | |
268 | int g = th->th_txg & TXG_MASK; | |
269 | ||
270 | mutex_enter(&tc->tc_lock); | |
271 | ASSERT(tc->tc_count[g] != 0); | |
272 | if (--tc->tc_count[g] == 0) | |
273 | cv_broadcast(&tc->tc_cv[g]); | |
274 | mutex_exit(&tc->tc_lock); | |
275 | ||
276 | th->th_cpu = NULL; /* defensive */ | |
277 | } | |
278 | ||
279 | static void | |
280 | txg_quiesce(dsl_pool_t *dp, uint64_t txg) | |
281 | { | |
282 | tx_state_t *tx = &dp->dp_tx; | |
283 | int g = txg & TXG_MASK; | |
284 | int c; | |
285 | ||
286 | /* | |
287 | * Grab all tx_cpu locks so nobody else can get into this txg. | |
288 | */ | |
289 | for (c = 0; c < max_ncpus; c++) | |
290 | mutex_enter(&tx->tx_cpu[c].tc_lock); | |
291 | ||
292 | ASSERT(txg == tx->tx_open_txg); | |
293 | tx->tx_open_txg++; | |
294 | ||
295 | /* | |
296 | * Now that we've incremented tx_open_txg, we can let threads | |
297 | * enter the next transaction group. | |
298 | */ | |
299 | for (c = 0; c < max_ncpus; c++) | |
300 | mutex_exit(&tx->tx_cpu[c].tc_lock); | |
301 | ||
302 | /* | |
303 | * Quiesce the transaction group by waiting for everyone to txg_exit(). | |
304 | */ | |
305 | for (c = 0; c < max_ncpus; c++) { | |
306 | tx_cpu_t *tc = &tx->tx_cpu[c]; | |
307 | mutex_enter(&tc->tc_lock); | |
308 | while (tc->tc_count[g] != 0) | |
309 | cv_wait(&tc->tc_cv[g], &tc->tc_lock); | |
310 | mutex_exit(&tc->tc_lock); | |
311 | } | |
312 | } | |
313 | ||
314 | static void | |
315 | txg_do_callbacks(list_t *cb_list) | |
316 | { | |
317 | dmu_tx_do_callbacks(cb_list, 0); | |
318 | ||
319 | list_destroy(cb_list); | |
320 | ||
321 | kmem_free(cb_list, sizeof (list_t)); | |
322 | } | |
323 | ||
324 | /* | |
325 | * Dispatch the commit callbacks registered on this txg to worker threads. | |
326 | */ | |
327 | static void | |
328 | txg_dispatch_callbacks(dsl_pool_t *dp, uint64_t txg) | |
329 | { | |
330 | int c; | |
331 | tx_state_t *tx = &dp->dp_tx; | |
332 | list_t *cb_list; | |
333 | ||
334 | for (c = 0; c < max_ncpus; c++) { | |
335 | tx_cpu_t *tc = &tx->tx_cpu[c]; | |
336 | /* No need to lock tx_cpu_t at this point */ | |
337 | ||
338 | int g = txg & TXG_MASK; | |
339 | ||
340 | if (list_is_empty(&tc->tc_callbacks[g])) | |
341 | continue; | |
342 | ||
343 | if (tx->tx_commit_cb_taskq == NULL) { | |
344 | /* | |
345 | * Commit callback taskq hasn't been created yet. | |
346 | */ | |
347 | tx->tx_commit_cb_taskq = taskq_create("tx_commit_cb", | |
348 | 100, minclsyspri, max_ncpus, INT_MAX, | |
349 | TASKQ_THREADS_CPU_PCT | TASKQ_PREPOPULATE); | |
350 | } | |
351 | ||
352 | cb_list = kmem_alloc(sizeof (list_t), KM_SLEEP); | |
353 | list_create(cb_list, sizeof (dmu_tx_callback_t), | |
354 | offsetof(dmu_tx_callback_t, dcb_node)); | |
355 | ||
356 | list_move_tail(cb_list, &tc->tc_callbacks[g]); | |
357 | ||
358 | (void) taskq_dispatch(tx->tx_commit_cb_taskq, (task_func_t *) | |
359 | txg_do_callbacks, cb_list, TQ_SLEEP); | |
360 | } | |
361 | } | |
362 | ||
363 | /* | |
364 | * Wait for pending commit callbacks of already-synced transactions to finish | |
365 | * processing. | |
366 | * Calling this function from within a commit callback will deadlock. | |
367 | */ | |
368 | void | |
369 | txg_wait_callbacks(dsl_pool_t *dp) | |
370 | { | |
371 | tx_state_t *tx = &dp->dp_tx; | |
372 | ||
373 | if (tx->tx_commit_cb_taskq != NULL) | |
374 | taskq_wait(tx->tx_commit_cb_taskq); | |
375 | } | |
376 | ||
377 | static void | |
378 | txg_sync_thread(dsl_pool_t *dp) | |
379 | { | |
380 | spa_t *spa = dp->dp_spa; | |
381 | tx_state_t *tx = &dp->dp_tx; | |
382 | callb_cpr_t cpr; | |
383 | uint64_t start, delta; | |
384 | ||
385 | #ifdef _KERNEL | |
386 | /* | |
387 | * Disable the normal reclaim path for the txg_sync thread. This | |
388 | * ensures the thread will never enter dmu_tx_assign() which can | |
389 | * otherwise occur due to direct reclaim. If this is allowed to | |
390 | * happen the system can deadlock. Direct reclaim call path: | |
391 | * | |
392 | * ->shrink_icache_memory->prune_icache->dispose_list-> | |
393 | * clear_inode->zpl_clear_inode->zfs_inactive->dmu_tx_assign | |
394 | */ | |
395 | current->flags |= PF_MEMALLOC; | |
396 | #endif /* _KERNEL */ | |
397 | ||
398 | txg_thread_enter(tx, &cpr); | |
399 | ||
400 | start = delta = 0; | |
401 | for (;;) { | |
402 | uint64_t timer, timeout = zfs_txg_timeout * hz; | |
403 | uint64_t txg; | |
404 | ||
405 | /* | |
406 | * We sync when we're scanning, there's someone waiting | |
407 | * on us, or the quiesce thread has handed off a txg to | |
408 | * us, or we have reached our timeout. | |
409 | */ | |
410 | timer = (delta >= timeout ? 0 : timeout - delta); | |
411 | while (!dsl_scan_active(dp->dp_scan) && | |
412 | !tx->tx_exiting && timer > 0 && | |
413 | tx->tx_synced_txg >= tx->tx_sync_txg_waiting && | |
414 | tx->tx_quiesced_txg == 0) { | |
415 | dprintf("waiting; tx_synced=%llu waiting=%llu dp=%p\n", | |
416 | tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp); | |
417 | txg_thread_wait(tx, &cpr, &tx->tx_sync_more_cv, timer); | |
418 | delta = ddi_get_lbolt() - start; | |
419 | timer = (delta > timeout ? 0 : timeout - delta); | |
420 | } | |
421 | ||
422 | /* | |
423 | * Wait until the quiesce thread hands off a txg to us, | |
424 | * prompting it to do so if necessary. | |
425 | */ | |
426 | while (!tx->tx_exiting && tx->tx_quiesced_txg == 0) { | |
427 | if (tx->tx_quiesce_txg_waiting < tx->tx_open_txg+1) | |
428 | tx->tx_quiesce_txg_waiting = tx->tx_open_txg+1; | |
429 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
430 | txg_thread_wait(tx, &cpr, &tx->tx_quiesce_done_cv, 0); | |
431 | } | |
432 | ||
433 | if (tx->tx_exiting) | |
434 | txg_thread_exit(tx, &cpr, &tx->tx_sync_thread); | |
435 | ||
436 | /* | |
437 | * Consume the quiesced txg which has been handed off to | |
438 | * us. This may cause the quiescing thread to now be | |
439 | * able to quiesce another txg, so we must signal it. | |
440 | */ | |
441 | txg = tx->tx_quiesced_txg; | |
442 | tx->tx_quiesced_txg = 0; | |
443 | tx->tx_syncing_txg = txg; | |
444 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
445 | ||
446 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
447 | txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); | |
448 | mutex_exit(&tx->tx_sync_lock); | |
449 | ||
450 | start = ddi_get_lbolt(); | |
451 | spa_sync(spa, txg); | |
452 | delta = ddi_get_lbolt() - start; | |
453 | ||
454 | mutex_enter(&tx->tx_sync_lock); | |
455 | tx->tx_synced_txg = txg; | |
456 | tx->tx_syncing_txg = 0; | |
457 | cv_broadcast(&tx->tx_sync_done_cv); | |
458 | ||
459 | /* | |
460 | * Dispatch commit callbacks to worker threads. | |
461 | */ | |
462 | txg_dispatch_callbacks(dp, txg); | |
463 | } | |
464 | } | |
465 | ||
466 | static void | |
467 | txg_quiesce_thread(dsl_pool_t *dp) | |
468 | { | |
469 | tx_state_t *tx = &dp->dp_tx; | |
470 | callb_cpr_t cpr; | |
471 | ||
472 | txg_thread_enter(tx, &cpr); | |
473 | ||
474 | for (;;) { | |
475 | uint64_t txg; | |
476 | ||
477 | /* | |
478 | * We quiesce when there's someone waiting on us. | |
479 | * However, we can only have one txg in "quiescing" or | |
480 | * "quiesced, waiting to sync" state. So we wait until | |
481 | * the "quiesced, waiting to sync" txg has been consumed | |
482 | * by the sync thread. | |
483 | */ | |
484 | while (!tx->tx_exiting && | |
485 | (tx->tx_open_txg >= tx->tx_quiesce_txg_waiting || | |
486 | tx->tx_quiesced_txg != 0)) | |
487 | txg_thread_wait(tx, &cpr, &tx->tx_quiesce_more_cv, 0); | |
488 | ||
489 | if (tx->tx_exiting) | |
490 | txg_thread_exit(tx, &cpr, &tx->tx_quiesce_thread); | |
491 | ||
492 | txg = tx->tx_open_txg; | |
493 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
494 | txg, tx->tx_quiesce_txg_waiting, | |
495 | tx->tx_sync_txg_waiting); | |
496 | mutex_exit(&tx->tx_sync_lock); | |
497 | txg_quiesce(dp, txg); | |
498 | mutex_enter(&tx->tx_sync_lock); | |
499 | ||
500 | /* | |
501 | * Hand this txg off to the sync thread. | |
502 | */ | |
503 | dprintf("quiesce done, handing off txg %llu\n", txg); | |
504 | tx->tx_quiesced_txg = txg; | |
505 | cv_broadcast(&tx->tx_sync_more_cv); | |
506 | cv_broadcast(&tx->tx_quiesce_done_cv); | |
507 | } | |
508 | } | |
509 | ||
510 | /* | |
511 | * Delay this thread by 'ticks' if we are still in the open transaction | |
512 | * group and there is already a waiting txg quiesing or quiesced. Abort | |
513 | * the delay if this txg stalls or enters the quiesing state. | |
514 | */ | |
515 | void | |
516 | txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks) | |
517 | { | |
518 | tx_state_t *tx = &dp->dp_tx; | |
519 | clock_t timeout = ddi_get_lbolt() + ticks; | |
520 | ||
521 | /* don't delay if this txg could transition to quiesing immediately */ | |
522 | if (tx->tx_open_txg > txg || | |
523 | tx->tx_syncing_txg == txg-1 || tx->tx_synced_txg == txg-1) | |
524 | return; | |
525 | ||
526 | mutex_enter(&tx->tx_sync_lock); | |
527 | if (tx->tx_open_txg > txg || tx->tx_synced_txg == txg-1) { | |
528 | mutex_exit(&tx->tx_sync_lock); | |
529 | return; | |
530 | } | |
531 | ||
532 | while (ddi_get_lbolt() < timeout && | |
533 | tx->tx_syncing_txg < txg-1 && !txg_stalled(dp)) | |
534 | (void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock, | |
535 | timeout); | |
536 | ||
537 | DMU_TX_STAT_BUMP(dmu_tx_delay); | |
538 | ||
539 | mutex_exit(&tx->tx_sync_lock); | |
540 | } | |
541 | ||
542 | void | |
543 | txg_wait_synced(dsl_pool_t *dp, uint64_t txg) | |
544 | { | |
545 | tx_state_t *tx = &dp->dp_tx; | |
546 | ||
547 | mutex_enter(&tx->tx_sync_lock); | |
548 | ASSERT(tx->tx_threads == 2); | |
549 | if (txg == 0) | |
550 | txg = tx->tx_open_txg + TXG_DEFER_SIZE; | |
551 | if (tx->tx_sync_txg_waiting < txg) | |
552 | tx->tx_sync_txg_waiting = txg; | |
553 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
554 | txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); | |
555 | while (tx->tx_synced_txg < txg) { | |
556 | dprintf("broadcasting sync more " | |
557 | "tx_synced=%llu waiting=%llu dp=%p\n", | |
558 | tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp); | |
559 | cv_broadcast(&tx->tx_sync_more_cv); | |
560 | cv_wait(&tx->tx_sync_done_cv, &tx->tx_sync_lock); | |
561 | } | |
562 | mutex_exit(&tx->tx_sync_lock); | |
563 | } | |
564 | ||
565 | void | |
566 | txg_wait_open(dsl_pool_t *dp, uint64_t txg) | |
567 | { | |
568 | tx_state_t *tx = &dp->dp_tx; | |
569 | ||
570 | mutex_enter(&tx->tx_sync_lock); | |
571 | ASSERT(tx->tx_threads == 2); | |
572 | if (txg == 0) | |
573 | txg = tx->tx_open_txg + 1; | |
574 | if (tx->tx_quiesce_txg_waiting < txg) | |
575 | tx->tx_quiesce_txg_waiting = txg; | |
576 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
577 | txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); | |
578 | while (tx->tx_open_txg < txg) { | |
579 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
580 | cv_wait(&tx->tx_quiesce_done_cv, &tx->tx_sync_lock); | |
581 | } | |
582 | mutex_exit(&tx->tx_sync_lock); | |
583 | } | |
584 | ||
585 | boolean_t | |
586 | txg_stalled(dsl_pool_t *dp) | |
587 | { | |
588 | tx_state_t *tx = &dp->dp_tx; | |
589 | return (tx->tx_quiesce_txg_waiting > tx->tx_open_txg); | |
590 | } | |
591 | ||
592 | boolean_t | |
593 | txg_sync_waiting(dsl_pool_t *dp) | |
594 | { | |
595 | tx_state_t *tx = &dp->dp_tx; | |
596 | ||
597 | return (tx->tx_syncing_txg <= tx->tx_sync_txg_waiting || | |
598 | tx->tx_quiesced_txg != 0); | |
599 | } | |
600 | ||
601 | /* | |
602 | * Per-txg object lists. | |
603 | */ | |
604 | void | |
605 | txg_list_create(txg_list_t *tl, size_t offset) | |
606 | { | |
607 | int t; | |
608 | ||
609 | mutex_init(&tl->tl_lock, NULL, MUTEX_DEFAULT, NULL); | |
610 | ||
611 | tl->tl_offset = offset; | |
612 | ||
613 | for (t = 0; t < TXG_SIZE; t++) | |
614 | tl->tl_head[t] = NULL; | |
615 | } | |
616 | ||
617 | void | |
618 | txg_list_destroy(txg_list_t *tl) | |
619 | { | |
620 | int t; | |
621 | ||
622 | for (t = 0; t < TXG_SIZE; t++) | |
623 | ASSERT(txg_list_empty(tl, t)); | |
624 | ||
625 | mutex_destroy(&tl->tl_lock); | |
626 | } | |
627 | ||
628 | int | |
629 | txg_list_empty(txg_list_t *tl, uint64_t txg) | |
630 | { | |
631 | return (tl->tl_head[txg & TXG_MASK] == NULL); | |
632 | } | |
633 | ||
634 | /* | |
635 | * Add an entry to the list. | |
636 | * Returns 0 if it's a new entry, 1 if it's already there. | |
637 | */ | |
638 | int | |
639 | txg_list_add(txg_list_t *tl, void *p, uint64_t txg) | |
640 | { | |
641 | int t = txg & TXG_MASK; | |
642 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
643 | int already_on_list; | |
644 | ||
645 | mutex_enter(&tl->tl_lock); | |
646 | already_on_list = tn->tn_member[t]; | |
647 | if (!already_on_list) { | |
648 | tn->tn_member[t] = 1; | |
649 | tn->tn_next[t] = tl->tl_head[t]; | |
650 | tl->tl_head[t] = tn; | |
651 | } | |
652 | mutex_exit(&tl->tl_lock); | |
653 | ||
654 | return (already_on_list); | |
655 | } | |
656 | ||
657 | /* | |
658 | * Add an entry to the end of the list (walks list to find end). | |
659 | * Returns 0 if it's a new entry, 1 if it's already there. | |
660 | */ | |
661 | int | |
662 | txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg) | |
663 | { | |
664 | int t = txg & TXG_MASK; | |
665 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
666 | int already_on_list; | |
667 | ||
668 | mutex_enter(&tl->tl_lock); | |
669 | already_on_list = tn->tn_member[t]; | |
670 | if (!already_on_list) { | |
671 | txg_node_t **tp; | |
672 | ||
673 | for (tp = &tl->tl_head[t]; *tp != NULL; tp = &(*tp)->tn_next[t]) | |
674 | continue; | |
675 | ||
676 | tn->tn_member[t] = 1; | |
677 | tn->tn_next[t] = NULL; | |
678 | *tp = tn; | |
679 | } | |
680 | mutex_exit(&tl->tl_lock); | |
681 | ||
682 | return (already_on_list); | |
683 | } | |
684 | ||
685 | /* | |
686 | * Remove the head of the list and return it. | |
687 | */ | |
688 | void * | |
689 | txg_list_remove(txg_list_t *tl, uint64_t txg) | |
690 | { | |
691 | int t = txg & TXG_MASK; | |
692 | txg_node_t *tn; | |
693 | void *p = NULL; | |
694 | ||
695 | mutex_enter(&tl->tl_lock); | |
696 | if ((tn = tl->tl_head[t]) != NULL) { | |
697 | p = (char *)tn - tl->tl_offset; | |
698 | tl->tl_head[t] = tn->tn_next[t]; | |
699 | tn->tn_next[t] = NULL; | |
700 | tn->tn_member[t] = 0; | |
701 | } | |
702 | mutex_exit(&tl->tl_lock); | |
703 | ||
704 | return (p); | |
705 | } | |
706 | ||
707 | /* | |
708 | * Remove a specific item from the list and return it. | |
709 | */ | |
710 | void * | |
711 | txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg) | |
712 | { | |
713 | int t = txg & TXG_MASK; | |
714 | txg_node_t *tn, **tp; | |
715 | ||
716 | mutex_enter(&tl->tl_lock); | |
717 | ||
718 | for (tp = &tl->tl_head[t]; (tn = *tp) != NULL; tp = &tn->tn_next[t]) { | |
719 | if ((char *)tn - tl->tl_offset == p) { | |
720 | *tp = tn->tn_next[t]; | |
721 | tn->tn_next[t] = NULL; | |
722 | tn->tn_member[t] = 0; | |
723 | mutex_exit(&tl->tl_lock); | |
724 | return (p); | |
725 | } | |
726 | } | |
727 | ||
728 | mutex_exit(&tl->tl_lock); | |
729 | ||
730 | return (NULL); | |
731 | } | |
732 | ||
733 | int | |
734 | txg_list_member(txg_list_t *tl, void *p, uint64_t txg) | |
735 | { | |
736 | int t = txg & TXG_MASK; | |
737 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
738 | ||
739 | return (tn->tn_member[t]); | |
740 | } | |
741 | ||
742 | /* | |
743 | * Walk a txg list -- only safe if you know it's not changing. | |
744 | */ | |
745 | void * | |
746 | txg_list_head(txg_list_t *tl, uint64_t txg) | |
747 | { | |
748 | int t = txg & TXG_MASK; | |
749 | txg_node_t *tn = tl->tl_head[t]; | |
750 | ||
751 | return (tn == NULL ? NULL : (char *)tn - tl->tl_offset); | |
752 | } | |
753 | ||
754 | void * | |
755 | txg_list_next(txg_list_t *tl, void *p, uint64_t txg) | |
756 | { | |
757 | int t = txg & TXG_MASK; | |
758 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
759 | ||
760 | tn = tn->tn_next[t]; | |
761 | ||
762 | return (tn == NULL ? NULL : (char *)tn - tl->tl_offset); | |
763 | } | |
764 | ||
765 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
766 | EXPORT_SYMBOL(txg_init); | |
767 | EXPORT_SYMBOL(txg_fini); | |
768 | EXPORT_SYMBOL(txg_sync_start); | |
769 | EXPORT_SYMBOL(txg_sync_stop); | |
770 | EXPORT_SYMBOL(txg_hold_open); | |
771 | EXPORT_SYMBOL(txg_rele_to_quiesce); | |
772 | EXPORT_SYMBOL(txg_rele_to_sync); | |
773 | EXPORT_SYMBOL(txg_register_callbacks); | |
774 | EXPORT_SYMBOL(txg_delay); | |
775 | EXPORT_SYMBOL(txg_wait_synced); | |
776 | EXPORT_SYMBOL(txg_wait_open); | |
777 | EXPORT_SYMBOL(txg_wait_callbacks); | |
778 | EXPORT_SYMBOL(txg_stalled); | |
779 | EXPORT_SYMBOL(txg_sync_waiting); | |
780 | #endif |