]>
git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/txg.c
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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
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 #include <sys/spa_impl.h>
35 * Pool-wide transaction groups.
38 static void txg_sync_thread(dsl_pool_t
*dp
);
39 static void txg_quiesce_thread(dsl_pool_t
*dp
);
41 int zfs_txg_timeout
= 5; /* max seconds worth of delta per txg */
44 * Prepare the txg subsystem.
47 txg_init(dsl_pool_t
*dp
, uint64_t txg
)
49 tx_state_t
*tx
= &dp
->dp_tx
;
51 bzero(tx
, sizeof (tx_state_t
));
53 tx
->tx_cpu
= vmem_zalloc(max_ncpus
* sizeof (tx_cpu_t
), KM_SLEEP
);
55 for (c
= 0; c
< max_ncpus
; c
++) {
58 mutex_init(&tx
->tx_cpu
[c
].tc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
59 for (i
= 0; i
< TXG_SIZE
; i
++) {
60 cv_init(&tx
->tx_cpu
[c
].tc_cv
[i
], NULL
, CV_DEFAULT
,
62 list_create(&tx
->tx_cpu
[c
].tc_callbacks
[i
],
63 sizeof (dmu_tx_callback_t
),
64 offsetof(dmu_tx_callback_t
, dcb_node
));
68 mutex_init(&tx
->tx_sync_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
70 cv_init(&tx
->tx_sync_more_cv
, NULL
, CV_DEFAULT
, NULL
);
71 cv_init(&tx
->tx_sync_done_cv
, NULL
, CV_DEFAULT
, NULL
);
72 cv_init(&tx
->tx_quiesce_more_cv
, NULL
, CV_DEFAULT
, NULL
);
73 cv_init(&tx
->tx_quiesce_done_cv
, NULL
, CV_DEFAULT
, NULL
);
74 cv_init(&tx
->tx_exit_cv
, NULL
, CV_DEFAULT
, NULL
);
76 tx
->tx_open_txg
= txg
;
80 * Close down the txg subsystem.
83 txg_fini(dsl_pool_t
*dp
)
85 tx_state_t
*tx
= &dp
->dp_tx
;
88 ASSERT(tx
->tx_threads
== 0);
90 mutex_destroy(&tx
->tx_sync_lock
);
92 cv_destroy(&tx
->tx_sync_more_cv
);
93 cv_destroy(&tx
->tx_sync_done_cv
);
94 cv_destroy(&tx
->tx_quiesce_more_cv
);
95 cv_destroy(&tx
->tx_quiesce_done_cv
);
96 cv_destroy(&tx
->tx_exit_cv
);
98 for (c
= 0; c
< max_ncpus
; c
++) {
101 mutex_destroy(&tx
->tx_cpu
[c
].tc_lock
);
102 for (i
= 0; i
< TXG_SIZE
; i
++) {
103 cv_destroy(&tx
->tx_cpu
[c
].tc_cv
[i
]);
104 list_destroy(&tx
->tx_cpu
[c
].tc_callbacks
[i
]);
108 if (tx
->tx_commit_cb_taskq
!= NULL
)
109 taskq_destroy(tx
->tx_commit_cb_taskq
);
111 vmem_free(tx
->tx_cpu
, max_ncpus
* sizeof (tx_cpu_t
));
113 bzero(tx
, sizeof (tx_state_t
));
117 * Start syncing transaction groups.
120 txg_sync_start(dsl_pool_t
*dp
)
122 tx_state_t
*tx
= &dp
->dp_tx
;
124 mutex_enter(&tx
->tx_sync_lock
);
126 dprintf("pool %p\n", dp
);
128 ASSERT(tx
->tx_threads
== 0);
132 tx
->tx_quiesce_thread
= thread_create(NULL
, 0, txg_quiesce_thread
,
133 dp
, 0, &p0
, TS_RUN
, minclsyspri
);
136 * The sync thread can need a larger-than-default stack size on
137 * 32-bit x86. This is due in part to nested pools and
138 * scrub_visitbp() recursion.
140 tx
->tx_sync_thread
= thread_create(NULL
, 32<<10, txg_sync_thread
,
141 dp
, 0, &p0
, TS_RUN
, minclsyspri
);
143 mutex_exit(&tx
->tx_sync_lock
);
147 txg_thread_enter(tx_state_t
*tx
, callb_cpr_t
*cpr
)
149 CALLB_CPR_INIT(cpr
, &tx
->tx_sync_lock
, callb_generic_cpr
, FTAG
);
150 mutex_enter(&tx
->tx_sync_lock
);
154 txg_thread_exit(tx_state_t
*tx
, callb_cpr_t
*cpr
, kthread_t
**tpp
)
156 ASSERT(*tpp
!= NULL
);
159 cv_broadcast(&tx
->tx_exit_cv
);
160 CALLB_CPR_EXIT(cpr
); /* drops &tx->tx_sync_lock */
165 txg_thread_wait(tx_state_t
*tx
, callb_cpr_t
*cpr
, kcondvar_t
*cv
, uint64_t time
)
167 CALLB_CPR_SAFE_BEGIN(cpr
);
170 (void) cv_timedwait_interruptible(cv
, &tx
->tx_sync_lock
,
171 ddi_get_lbolt() + time
);
173 cv_wait_interruptible(cv
, &tx
->tx_sync_lock
);
175 CALLB_CPR_SAFE_END(cpr
, &tx
->tx_sync_lock
);
179 * Stop syncing transaction groups.
182 txg_sync_stop(dsl_pool_t
*dp
)
184 tx_state_t
*tx
= &dp
->dp_tx
;
186 dprintf("pool %p\n", dp
);
188 * Finish off any work in progress.
190 ASSERT(tx
->tx_threads
== 2);
193 * We need to ensure that we've vacated the deferred space_maps.
195 txg_wait_synced(dp
, tx
->tx_open_txg
+ TXG_DEFER_SIZE
);
198 * Wake all sync threads and wait for them to die.
200 mutex_enter(&tx
->tx_sync_lock
);
202 ASSERT(tx
->tx_threads
== 2);
206 cv_broadcast(&tx
->tx_quiesce_more_cv
);
207 cv_broadcast(&tx
->tx_quiesce_done_cv
);
208 cv_broadcast(&tx
->tx_sync_more_cv
);
210 while (tx
->tx_threads
!= 0)
211 cv_wait(&tx
->tx_exit_cv
, &tx
->tx_sync_lock
);
215 mutex_exit(&tx
->tx_sync_lock
);
219 txg_hold_open(dsl_pool_t
*dp
, txg_handle_t
*th
)
221 tx_state_t
*tx
= &dp
->dp_tx
;
226 * It appears the processor id is simply used as a "random"
227 * number to index into the array, and there isn't any other
228 * significance to the chosen tx_cpu. Because.. Why not use
229 * the current cpu to index into the array?
232 tc
= &tx
->tx_cpu
[CPU_SEQID
];
235 mutex_enter(&tc
->tc_lock
);
237 txg
= tx
->tx_open_txg
;
238 tc
->tc_count
[txg
& TXG_MASK
]++;
247 txg_rele_to_quiesce(txg_handle_t
*th
)
249 tx_cpu_t
*tc
= th
->th_cpu
;
251 mutex_exit(&tc
->tc_lock
);
255 txg_register_callbacks(txg_handle_t
*th
, list_t
*tx_callbacks
)
257 tx_cpu_t
*tc
= th
->th_cpu
;
258 int g
= th
->th_txg
& TXG_MASK
;
260 mutex_enter(&tc
->tc_lock
);
261 list_move_tail(&tc
->tc_callbacks
[g
], tx_callbacks
);
262 mutex_exit(&tc
->tc_lock
);
266 txg_rele_to_sync(txg_handle_t
*th
)
268 tx_cpu_t
*tc
= th
->th_cpu
;
269 int g
= th
->th_txg
& TXG_MASK
;
271 mutex_enter(&tc
->tc_lock
);
272 ASSERT(tc
->tc_count
[g
] != 0);
273 if (--tc
->tc_count
[g
] == 0)
274 cv_broadcast(&tc
->tc_cv
[g
]);
275 mutex_exit(&tc
->tc_lock
);
277 th
->th_cpu
= NULL
; /* defensive */
281 txg_quiesce(dsl_pool_t
*dp
, uint64_t txg
)
285 tx_state_t
*tx
= &dp
->dp_tx
;
286 int g
= txg
& TXG_MASK
;
290 * Grab all tx_cpu locks so nobody else can get into this txg.
292 for (c
= 0; c
< max_ncpus
; c
++)
293 mutex_enter(&tx
->tx_cpu
[c
].tc_lock
);
295 ASSERT(txg
== tx
->tx_open_txg
);
299 * Measure how long the txg was open and replace the kstat.
301 th
= dsl_pool_txg_history_get(dp
, txg
);
302 th
->th_kstat
.open_time
= gethrtime() - th
->th_kstat
.birth
;
303 th
->th_kstat
.state
= TXG_STATE_QUIESCING
;
304 dsl_pool_txg_history_put(th
);
305 dsl_pool_txg_history_add(dp
, tx
->tx_open_txg
);
308 * Now that we've incremented tx_open_txg, we can let threads
309 * enter the next transaction group.
311 for (c
= 0; c
< max_ncpus
; c
++)
312 mutex_exit(&tx
->tx_cpu
[c
].tc_lock
);
315 * Quiesce the transaction group by waiting for everyone to txg_exit().
319 for (c
= 0; c
< max_ncpus
; c
++) {
320 tx_cpu_t
*tc
= &tx
->tx_cpu
[c
];
321 mutex_enter(&tc
->tc_lock
);
322 while (tc
->tc_count
[g
] != 0)
323 cv_wait(&tc
->tc_cv
[g
], &tc
->tc_lock
);
324 mutex_exit(&tc
->tc_lock
);
328 * Measure how long the txg took to quiesce.
330 th
= dsl_pool_txg_history_get(dp
, txg
);
331 th
->th_kstat
.quiesce_time
= gethrtime() - start
;
332 dsl_pool_txg_history_put(th
);
336 txg_do_callbacks(list_t
*cb_list
)
338 dmu_tx_do_callbacks(cb_list
, 0);
340 list_destroy(cb_list
);
342 kmem_free(cb_list
, sizeof (list_t
));
346 * Dispatch the commit callbacks registered on this txg to worker threads.
349 txg_dispatch_callbacks(dsl_pool_t
*dp
, uint64_t txg
)
352 tx_state_t
*tx
= &dp
->dp_tx
;
355 for (c
= 0; c
< max_ncpus
; c
++) {
356 tx_cpu_t
*tc
= &tx
->tx_cpu
[c
];
357 /* No need to lock tx_cpu_t at this point */
359 int g
= txg
& TXG_MASK
;
361 if (list_is_empty(&tc
->tc_callbacks
[g
]))
364 if (tx
->tx_commit_cb_taskq
== NULL
) {
366 * Commit callback taskq hasn't been created yet.
368 tx
->tx_commit_cb_taskq
= taskq_create("tx_commit_cb",
369 100, minclsyspri
, max_ncpus
, INT_MAX
,
370 TASKQ_THREADS_CPU_PCT
| TASKQ_PREPOPULATE
);
373 cb_list
= kmem_alloc(sizeof (list_t
), KM_PUSHPAGE
);
374 list_create(cb_list
, sizeof (dmu_tx_callback_t
),
375 offsetof(dmu_tx_callback_t
, dcb_node
));
377 list_move_tail(cb_list
, &tc
->tc_callbacks
[g
]);
379 (void) taskq_dispatch(tx
->tx_commit_cb_taskq
, (task_func_t
*)
380 txg_do_callbacks
, cb_list
, TQ_SLEEP
);
385 * Wait for pending commit callbacks of already-synced transactions to finish
387 * Calling this function from within a commit callback will deadlock.
390 txg_wait_callbacks(dsl_pool_t
*dp
)
392 tx_state_t
*tx
= &dp
->dp_tx
;
394 if (tx
->tx_commit_cb_taskq
!= NULL
)
395 taskq_wait(tx
->tx_commit_cb_taskq
);
399 txg_sync_thread(dsl_pool_t
*dp
)
401 spa_t
*spa
= dp
->dp_spa
;
402 tx_state_t
*tx
= &dp
->dp_tx
;
404 uint64_t start
, delta
;
408 * Annotate this process with a flag that indicates that it is
409 * unsafe to use KM_SLEEP during memory allocations due to the
410 * potential for a deadlock. KM_PUSHPAGE should be used instead.
412 current
->flags
|= PF_NOFS
;
415 txg_thread_enter(tx
, &cpr
);
421 uint64_t timer
, timeout
;
424 timeout
= zfs_txg_timeout
* hz
;
427 * We sync when we're scanning, there's someone waiting
428 * on us, or the quiesce thread has handed off a txg to
429 * us, or we have reached our timeout.
431 timer
= (delta
>= timeout
? 0 : timeout
- delta
);
432 while (!dsl_scan_active(dp
->dp_scan
) &&
433 !tx
->tx_exiting
&& timer
> 0 &&
434 tx
->tx_synced_txg
>= tx
->tx_sync_txg_waiting
&&
435 tx
->tx_quiesced_txg
== 0) {
436 dprintf("waiting; tx_synced=%llu waiting=%llu dp=%p\n",
437 tx
->tx_synced_txg
, tx
->tx_sync_txg_waiting
, dp
);
438 txg_thread_wait(tx
, &cpr
, &tx
->tx_sync_more_cv
, timer
);
439 delta
= ddi_get_lbolt() - start
;
440 timer
= (delta
> timeout
? 0 : timeout
- delta
);
444 * Wait until the quiesce thread hands off a txg to us,
445 * prompting it to do so if necessary.
447 while (!tx
->tx_exiting
&& tx
->tx_quiesced_txg
== 0) {
448 if (tx
->tx_quiesce_txg_waiting
< tx
->tx_open_txg
+1)
449 tx
->tx_quiesce_txg_waiting
= tx
->tx_open_txg
+1;
450 cv_broadcast(&tx
->tx_quiesce_more_cv
);
451 txg_thread_wait(tx
, &cpr
, &tx
->tx_quiesce_done_cv
, 0);
455 txg_thread_exit(tx
, &cpr
, &tx
->tx_sync_thread
);
458 * Consume the quiesced txg which has been handed off to
459 * us. This may cause the quiescing thread to now be
460 * able to quiesce another txg, so we must signal it.
462 txg
= tx
->tx_quiesced_txg
;
463 tx
->tx_quiesced_txg
= 0;
464 tx
->tx_syncing_txg
= txg
;
465 cv_broadcast(&tx
->tx_quiesce_more_cv
);
467 th
= dsl_pool_txg_history_get(dp
, txg
);
468 th
->th_kstat
.state
= TXG_STATE_SYNCING
;
469 vdev_get_stats(spa
->spa_root_vdev
, &th
->th_vs1
);
470 dsl_pool_txg_history_put(th
);
472 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
473 txg
, tx
->tx_quiesce_txg_waiting
, tx
->tx_sync_txg_waiting
);
474 mutex_exit(&tx
->tx_sync_lock
);
476 start
= ddi_get_lbolt();
477 hrstart
= gethrtime();
479 delta
= ddi_get_lbolt() - start
;
481 mutex_enter(&tx
->tx_sync_lock
);
482 tx
->tx_synced_txg
= txg
;
483 tx
->tx_syncing_txg
= 0;
484 cv_broadcast(&tx
->tx_sync_done_cv
);
487 * Dispatch commit callbacks to worker threads.
489 txg_dispatch_callbacks(dp
, txg
);
492 * Measure the txg sync time determine the amount of I/O done.
494 th
= dsl_pool_txg_history_get(dp
, txg
);
495 vdev_get_stats(spa
->spa_root_vdev
, &th
->th_vs2
);
496 th
->th_kstat
.sync_time
= gethrtime() - hrstart
;
497 th
->th_kstat
.nread
= th
->th_vs2
.vs_bytes
[ZIO_TYPE_READ
] -
498 th
->th_vs1
.vs_bytes
[ZIO_TYPE_READ
];
499 th
->th_kstat
.nwritten
= th
->th_vs2
.vs_bytes
[ZIO_TYPE_WRITE
] -
500 th
->th_vs1
.vs_bytes
[ZIO_TYPE_WRITE
];
501 th
->th_kstat
.reads
= th
->th_vs2
.vs_ops
[ZIO_TYPE_READ
] -
502 th
->th_vs1
.vs_ops
[ZIO_TYPE_READ
];
503 th
->th_kstat
.writes
= th
->th_vs2
.vs_ops
[ZIO_TYPE_WRITE
] -
504 th
->th_vs1
.vs_ops
[ZIO_TYPE_WRITE
];
505 th
->th_kstat
.state
= TXG_STATE_COMMITTED
;
506 dsl_pool_txg_history_put(th
);
511 txg_quiesce_thread(dsl_pool_t
*dp
)
513 tx_state_t
*tx
= &dp
->dp_tx
;
516 txg_thread_enter(tx
, &cpr
);
522 * We quiesce when there's someone waiting on us.
523 * However, we can only have one txg in "quiescing" or
524 * "quiesced, waiting to sync" state. So we wait until
525 * the "quiesced, waiting to sync" txg has been consumed
526 * by the sync thread.
528 while (!tx
->tx_exiting
&&
529 (tx
->tx_open_txg
>= tx
->tx_quiesce_txg_waiting
||
530 tx
->tx_quiesced_txg
!= 0))
531 txg_thread_wait(tx
, &cpr
, &tx
->tx_quiesce_more_cv
, 0);
534 txg_thread_exit(tx
, &cpr
, &tx
->tx_quiesce_thread
);
536 txg
= tx
->tx_open_txg
;
537 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
538 txg
, tx
->tx_quiesce_txg_waiting
,
539 tx
->tx_sync_txg_waiting
);
540 mutex_exit(&tx
->tx_sync_lock
);
541 txg_quiesce(dp
, txg
);
542 mutex_enter(&tx
->tx_sync_lock
);
545 * Hand this txg off to the sync thread.
547 dprintf("quiesce done, handing off txg %llu\n", txg
);
548 tx
->tx_quiesced_txg
= txg
;
549 cv_broadcast(&tx
->tx_sync_more_cv
);
550 cv_broadcast(&tx
->tx_quiesce_done_cv
);
555 * Delay this thread by 'ticks' if we are still in the open transaction
556 * group and there is already a waiting txg quiesing or quiesced. Abort
557 * the delay if this txg stalls or enters the quiesing state.
560 txg_delay(dsl_pool_t
*dp
, uint64_t txg
, int ticks
)
562 tx_state_t
*tx
= &dp
->dp_tx
;
563 clock_t timeout
= ddi_get_lbolt() + ticks
;
565 /* don't delay if this txg could transition to quiesing immediately */
566 if (tx
->tx_open_txg
> txg
||
567 tx
->tx_syncing_txg
== txg
-1 || tx
->tx_synced_txg
== txg
-1)
570 mutex_enter(&tx
->tx_sync_lock
);
571 if (tx
->tx_open_txg
> txg
|| tx
->tx_synced_txg
== txg
-1) {
572 mutex_exit(&tx
->tx_sync_lock
);
576 while (ddi_get_lbolt() < timeout
&&
577 tx
->tx_syncing_txg
< txg
-1 && !txg_stalled(dp
))
578 (void) cv_timedwait(&tx
->tx_quiesce_more_cv
, &tx
->tx_sync_lock
,
581 DMU_TX_STAT_BUMP(dmu_tx_delay
);
583 mutex_exit(&tx
->tx_sync_lock
);
587 txg_wait_synced(dsl_pool_t
*dp
, uint64_t txg
)
589 tx_state_t
*tx
= &dp
->dp_tx
;
591 mutex_enter(&tx
->tx_sync_lock
);
592 ASSERT(tx
->tx_threads
== 2);
594 txg
= tx
->tx_open_txg
+ TXG_DEFER_SIZE
;
595 if (tx
->tx_sync_txg_waiting
< txg
)
596 tx
->tx_sync_txg_waiting
= txg
;
597 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
598 txg
, tx
->tx_quiesce_txg_waiting
, tx
->tx_sync_txg_waiting
);
599 while (tx
->tx_synced_txg
< txg
) {
600 dprintf("broadcasting sync more "
601 "tx_synced=%llu waiting=%llu dp=%p\n",
602 tx
->tx_synced_txg
, tx
->tx_sync_txg_waiting
, dp
);
603 cv_broadcast(&tx
->tx_sync_more_cv
);
604 cv_wait(&tx
->tx_sync_done_cv
, &tx
->tx_sync_lock
);
606 mutex_exit(&tx
->tx_sync_lock
);
610 txg_wait_open(dsl_pool_t
*dp
, uint64_t txg
)
612 tx_state_t
*tx
= &dp
->dp_tx
;
614 mutex_enter(&tx
->tx_sync_lock
);
615 ASSERT(tx
->tx_threads
== 2);
617 txg
= tx
->tx_open_txg
+ 1;
618 if (tx
->tx_quiesce_txg_waiting
< txg
)
619 tx
->tx_quiesce_txg_waiting
= txg
;
620 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
621 txg
, tx
->tx_quiesce_txg_waiting
, tx
->tx_sync_txg_waiting
);
622 while (tx
->tx_open_txg
< txg
) {
623 cv_broadcast(&tx
->tx_quiesce_more_cv
);
624 cv_wait(&tx
->tx_quiesce_done_cv
, &tx
->tx_sync_lock
);
626 mutex_exit(&tx
->tx_sync_lock
);
630 txg_stalled(dsl_pool_t
*dp
)
632 tx_state_t
*tx
= &dp
->dp_tx
;
633 return (tx
->tx_quiesce_txg_waiting
> tx
->tx_open_txg
);
637 txg_sync_waiting(dsl_pool_t
*dp
)
639 tx_state_t
*tx
= &dp
->dp_tx
;
641 return (tx
->tx_syncing_txg
<= tx
->tx_sync_txg_waiting
||
642 tx
->tx_quiesced_txg
!= 0);
646 * Per-txg object lists.
649 txg_list_create(txg_list_t
*tl
, size_t offset
)
653 mutex_init(&tl
->tl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
655 tl
->tl_offset
= offset
;
657 for (t
= 0; t
< TXG_SIZE
; t
++)
658 tl
->tl_head
[t
] = NULL
;
662 txg_list_destroy(txg_list_t
*tl
)
666 for (t
= 0; t
< TXG_SIZE
; t
++)
667 ASSERT(txg_list_empty(tl
, t
));
669 mutex_destroy(&tl
->tl_lock
);
673 txg_list_empty(txg_list_t
*tl
, uint64_t txg
)
675 return (tl
->tl_head
[txg
& TXG_MASK
] == NULL
);
679 * Add an entry to the list.
680 * Returns 0 if it's a new entry, 1 if it's already there.
683 txg_list_add(txg_list_t
*tl
, void *p
, uint64_t txg
)
685 int t
= txg
& TXG_MASK
;
686 txg_node_t
*tn
= (txg_node_t
*)((char *)p
+ tl
->tl_offset
);
689 mutex_enter(&tl
->tl_lock
);
690 already_on_list
= tn
->tn_member
[t
];
691 if (!already_on_list
) {
692 tn
->tn_member
[t
] = 1;
693 tn
->tn_next
[t
] = tl
->tl_head
[t
];
696 mutex_exit(&tl
->tl_lock
);
698 return (already_on_list
);
702 * Add an entry to the end of the list (walks list to find end).
703 * Returns 0 if it's a new entry, 1 if it's already there.
706 txg_list_add_tail(txg_list_t
*tl
, void *p
, uint64_t txg
)
708 int t
= txg
& TXG_MASK
;
709 txg_node_t
*tn
= (txg_node_t
*)((char *)p
+ tl
->tl_offset
);
712 mutex_enter(&tl
->tl_lock
);
713 already_on_list
= tn
->tn_member
[t
];
714 if (!already_on_list
) {
717 for (tp
= &tl
->tl_head
[t
]; *tp
!= NULL
; tp
= &(*tp
)->tn_next
[t
])
720 tn
->tn_member
[t
] = 1;
721 tn
->tn_next
[t
] = NULL
;
724 mutex_exit(&tl
->tl_lock
);
726 return (already_on_list
);
730 * Remove the head of the list and return it.
733 txg_list_remove(txg_list_t
*tl
, uint64_t txg
)
735 int t
= txg
& TXG_MASK
;
739 mutex_enter(&tl
->tl_lock
);
740 if ((tn
= tl
->tl_head
[t
]) != NULL
) {
741 p
= (char *)tn
- tl
->tl_offset
;
742 tl
->tl_head
[t
] = tn
->tn_next
[t
];
743 tn
->tn_next
[t
] = NULL
;
744 tn
->tn_member
[t
] = 0;
746 mutex_exit(&tl
->tl_lock
);
752 * Remove a specific item from the list and return it.
755 txg_list_remove_this(txg_list_t
*tl
, void *p
, uint64_t txg
)
757 int t
= txg
& TXG_MASK
;
758 txg_node_t
*tn
, **tp
;
760 mutex_enter(&tl
->tl_lock
);
762 for (tp
= &tl
->tl_head
[t
]; (tn
= *tp
) != NULL
; tp
= &tn
->tn_next
[t
]) {
763 if ((char *)tn
- tl
->tl_offset
== p
) {
764 *tp
= tn
->tn_next
[t
];
765 tn
->tn_next
[t
] = NULL
;
766 tn
->tn_member
[t
] = 0;
767 mutex_exit(&tl
->tl_lock
);
772 mutex_exit(&tl
->tl_lock
);
778 txg_list_member(txg_list_t
*tl
, void *p
, uint64_t txg
)
780 int t
= txg
& TXG_MASK
;
781 txg_node_t
*tn
= (txg_node_t
*)((char *)p
+ tl
->tl_offset
);
783 return (tn
->tn_member
[t
]);
787 * Walk a txg list -- only safe if you know it's not changing.
790 txg_list_head(txg_list_t
*tl
, uint64_t txg
)
792 int t
= txg
& TXG_MASK
;
793 txg_node_t
*tn
= tl
->tl_head
[t
];
795 return (tn
== NULL
? NULL
: (char *)tn
- tl
->tl_offset
);
799 txg_list_next(txg_list_t
*tl
, void *p
, uint64_t txg
)
801 int t
= txg
& TXG_MASK
;
802 txg_node_t
*tn
= (txg_node_t
*)((char *)p
+ tl
->tl_offset
);
806 return (tn
== NULL
? NULL
: (char *)tn
- tl
->tl_offset
);
809 #if defined(_KERNEL) && defined(HAVE_SPL)
810 EXPORT_SYMBOL(txg_init
);
811 EXPORT_SYMBOL(txg_fini
);
812 EXPORT_SYMBOL(txg_sync_start
);
813 EXPORT_SYMBOL(txg_sync_stop
);
814 EXPORT_SYMBOL(txg_hold_open
);
815 EXPORT_SYMBOL(txg_rele_to_quiesce
);
816 EXPORT_SYMBOL(txg_rele_to_sync
);
817 EXPORT_SYMBOL(txg_register_callbacks
);
818 EXPORT_SYMBOL(txg_delay
);
819 EXPORT_SYMBOL(txg_wait_synced
);
820 EXPORT_SYMBOL(txg_wait_open
);
821 EXPORT_SYMBOL(txg_wait_callbacks
);
822 EXPORT_SYMBOL(txg_stalled
);
823 EXPORT_SYMBOL(txg_sync_waiting
);
825 module_param(zfs_txg_timeout
, int, 0644);
826 MODULE_PARM_DESC(zfs_txg_timeout
, "Max seconds worth of delta per txg");