]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 2008 Sun Microsystems, Inc. All rights reserved. | |
23 | * Use is subject to license terms. | |
24 | */ | |
25 | ||
34dc7c2f BB |
26 | #include <sys/zfs_context.h> |
27 | #include <sys/txg_impl.h> | |
28 | #include <sys/dmu_impl.h> | |
29 | #include <sys/dsl_pool.h> | |
30 | #include <sys/callb.h> | |
31 | ||
32 | /* | |
33 | * Pool-wide transaction groups. | |
34 | */ | |
35 | ||
36 | static void txg_sync_thread(dsl_pool_t *dp); | |
37 | static void txg_quiesce_thread(dsl_pool_t *dp); | |
38 | ||
39 | int zfs_txg_timeout = 30; /* max seconds worth of delta per txg */ | |
34dc7c2f BB |
40 | |
41 | /* | |
42 | * Prepare the txg subsystem. | |
43 | */ | |
44 | void | |
45 | txg_init(dsl_pool_t *dp, uint64_t txg) | |
46 | { | |
47 | tx_state_t *tx = &dp->dp_tx; | |
48 | int c; | |
49 | bzero(tx, sizeof (tx_state_t)); | |
50 | ||
51 | tx->tx_cpu = kmem_zalloc(max_ncpus * sizeof (tx_cpu_t), KM_SLEEP); | |
52 | ||
53 | for (c = 0; c < max_ncpus; c++) { | |
54 | int i; | |
55 | ||
56 | mutex_init(&tx->tx_cpu[c].tc_lock, NULL, MUTEX_DEFAULT, NULL); | |
57 | for (i = 0; i < TXG_SIZE; i++) { | |
58 | cv_init(&tx->tx_cpu[c].tc_cv[i], NULL, CV_DEFAULT, | |
59 | NULL); | |
60 | } | |
61 | } | |
62 | ||
63 | rw_init(&tx->tx_suspend, NULL, RW_DEFAULT, NULL); | |
64 | mutex_init(&tx->tx_sync_lock, NULL, MUTEX_DEFAULT, NULL); | |
65 | ||
66 | tx->tx_open_txg = txg; | |
67 | } | |
68 | ||
69 | /* | |
70 | * Close down the txg subsystem. | |
71 | */ | |
72 | void | |
73 | txg_fini(dsl_pool_t *dp) | |
74 | { | |
75 | tx_state_t *tx = &dp->dp_tx; | |
76 | int c; | |
77 | ||
78 | ASSERT(tx->tx_threads == 0); | |
79 | ||
80 | rw_destroy(&tx->tx_suspend); | |
81 | mutex_destroy(&tx->tx_sync_lock); | |
82 | ||
83 | for (c = 0; c < max_ncpus; c++) { | |
84 | int i; | |
85 | ||
86 | mutex_destroy(&tx->tx_cpu[c].tc_lock); | |
87 | for (i = 0; i < TXG_SIZE; i++) | |
88 | cv_destroy(&tx->tx_cpu[c].tc_cv[i]); | |
89 | } | |
90 | ||
91 | kmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t)); | |
92 | ||
93 | bzero(tx, sizeof (tx_state_t)); | |
94 | } | |
95 | ||
96 | /* | |
97 | * Start syncing transaction groups. | |
98 | */ | |
99 | void | |
100 | txg_sync_start(dsl_pool_t *dp) | |
101 | { | |
102 | tx_state_t *tx = &dp->dp_tx; | |
103 | ||
104 | mutex_enter(&tx->tx_sync_lock); | |
105 | ||
106 | dprintf("pool %p\n", dp); | |
107 | ||
108 | ASSERT(tx->tx_threads == 0); | |
109 | ||
110 | tx->tx_threads = 2; | |
111 | ||
112 | tx->tx_quiesce_thread = thread_create(NULL, 0, txg_quiesce_thread, | |
113 | dp, 0, &p0, TS_RUN, minclsyspri); | |
114 | ||
b128c09f BB |
115 | /* |
116 | * The sync thread can need a larger-than-default stack size on | |
117 | * 32-bit x86. This is due in part to nested pools and | |
118 | * scrub_visitbp() recursion. | |
119 | */ | |
120 | tx->tx_sync_thread = thread_create(NULL, 12<<10, txg_sync_thread, | |
34dc7c2f BB |
121 | dp, 0, &p0, TS_RUN, minclsyspri); |
122 | ||
123 | mutex_exit(&tx->tx_sync_lock); | |
124 | } | |
125 | ||
126 | static void | |
127 | txg_thread_enter(tx_state_t *tx, callb_cpr_t *cpr) | |
128 | { | |
129 | CALLB_CPR_INIT(cpr, &tx->tx_sync_lock, callb_generic_cpr, FTAG); | |
130 | mutex_enter(&tx->tx_sync_lock); | |
131 | } | |
132 | ||
133 | static void | |
134 | txg_thread_exit(tx_state_t *tx, callb_cpr_t *cpr, kthread_t **tpp) | |
135 | { | |
136 | ASSERT(*tpp != NULL); | |
137 | *tpp = NULL; | |
138 | tx->tx_threads--; | |
139 | cv_broadcast(&tx->tx_exit_cv); | |
140 | CALLB_CPR_EXIT(cpr); /* drops &tx->tx_sync_lock */ | |
141 | thread_exit(); | |
142 | } | |
143 | ||
144 | static void | |
145 | txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time) | |
146 | { | |
147 | CALLB_CPR_SAFE_BEGIN(cpr); | |
148 | ||
149 | if (time) | |
150 | (void) cv_timedwait(cv, &tx->tx_sync_lock, lbolt + time); | |
151 | else | |
152 | cv_wait(cv, &tx->tx_sync_lock); | |
153 | ||
154 | CALLB_CPR_SAFE_END(cpr, &tx->tx_sync_lock); | |
155 | } | |
156 | ||
157 | /* | |
158 | * Stop syncing transaction groups. | |
159 | */ | |
160 | void | |
161 | txg_sync_stop(dsl_pool_t *dp) | |
162 | { | |
163 | tx_state_t *tx = &dp->dp_tx; | |
164 | ||
165 | dprintf("pool %p\n", dp); | |
166 | /* | |
167 | * Finish off any work in progress. | |
168 | */ | |
169 | ASSERT(tx->tx_threads == 2); | |
170 | txg_wait_synced(dp, 0); | |
171 | ||
172 | /* | |
173 | * Wake all sync threads and wait for them to die. | |
174 | */ | |
175 | mutex_enter(&tx->tx_sync_lock); | |
176 | ||
177 | ASSERT(tx->tx_threads == 2); | |
178 | ||
179 | tx->tx_exiting = 1; | |
180 | ||
181 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
182 | cv_broadcast(&tx->tx_quiesce_done_cv); | |
183 | cv_broadcast(&tx->tx_sync_more_cv); | |
184 | ||
185 | while (tx->tx_threads != 0) | |
186 | cv_wait(&tx->tx_exit_cv, &tx->tx_sync_lock); | |
187 | ||
188 | tx->tx_exiting = 0; | |
189 | ||
190 | mutex_exit(&tx->tx_sync_lock); | |
191 | } | |
192 | ||
193 | uint64_t | |
194 | txg_hold_open(dsl_pool_t *dp, txg_handle_t *th) | |
195 | { | |
196 | tx_state_t *tx = &dp->dp_tx; | |
197 | tx_cpu_t *tc = &tx->tx_cpu[CPU_SEQID]; | |
198 | uint64_t txg; | |
199 | ||
200 | mutex_enter(&tc->tc_lock); | |
201 | ||
202 | txg = tx->tx_open_txg; | |
203 | tc->tc_count[txg & TXG_MASK]++; | |
204 | ||
205 | th->th_cpu = tc; | |
206 | th->th_txg = txg; | |
207 | ||
208 | return (txg); | |
209 | } | |
210 | ||
211 | void | |
212 | txg_rele_to_quiesce(txg_handle_t *th) | |
213 | { | |
214 | tx_cpu_t *tc = th->th_cpu; | |
215 | ||
216 | mutex_exit(&tc->tc_lock); | |
217 | } | |
218 | ||
219 | void | |
220 | txg_rele_to_sync(txg_handle_t *th) | |
221 | { | |
222 | tx_cpu_t *tc = th->th_cpu; | |
223 | int g = th->th_txg & TXG_MASK; | |
224 | ||
225 | mutex_enter(&tc->tc_lock); | |
226 | ASSERT(tc->tc_count[g] != 0); | |
227 | if (--tc->tc_count[g] == 0) | |
228 | cv_broadcast(&tc->tc_cv[g]); | |
229 | mutex_exit(&tc->tc_lock); | |
230 | ||
231 | th->th_cpu = NULL; /* defensive */ | |
232 | } | |
233 | ||
234 | static void | |
235 | txg_quiesce(dsl_pool_t *dp, uint64_t txg) | |
236 | { | |
237 | tx_state_t *tx = &dp->dp_tx; | |
238 | int g = txg & TXG_MASK; | |
239 | int c; | |
240 | ||
241 | /* | |
242 | * Grab all tx_cpu locks so nobody else can get into this txg. | |
243 | */ | |
244 | for (c = 0; c < max_ncpus; c++) | |
245 | mutex_enter(&tx->tx_cpu[c].tc_lock); | |
246 | ||
247 | ASSERT(txg == tx->tx_open_txg); | |
248 | tx->tx_open_txg++; | |
249 | ||
250 | /* | |
251 | * Now that we've incremented tx_open_txg, we can let threads | |
252 | * enter the next transaction group. | |
253 | */ | |
254 | for (c = 0; c < max_ncpus; c++) | |
255 | mutex_exit(&tx->tx_cpu[c].tc_lock); | |
256 | ||
257 | /* | |
258 | * Quiesce the transaction group by waiting for everyone to txg_exit(). | |
259 | */ | |
260 | for (c = 0; c < max_ncpus; c++) { | |
261 | tx_cpu_t *tc = &tx->tx_cpu[c]; | |
262 | mutex_enter(&tc->tc_lock); | |
263 | while (tc->tc_count[g] != 0) | |
264 | cv_wait(&tc->tc_cv[g], &tc->tc_lock); | |
265 | mutex_exit(&tc->tc_lock); | |
266 | } | |
267 | } | |
268 | ||
269 | static void | |
270 | txg_sync_thread(dsl_pool_t *dp) | |
271 | { | |
272 | tx_state_t *tx = &dp->dp_tx; | |
273 | callb_cpr_t cpr; | |
b128c09f | 274 | uint64_t start, delta; |
34dc7c2f BB |
275 | |
276 | txg_thread_enter(tx, &cpr); | |
277 | ||
278 | start = delta = 0; | |
34dc7c2f | 279 | for (;;) { |
b128c09f BB |
280 | uint64_t timer, timeout = zfs_txg_timeout * hz; |
281 | uint64_t txg; | |
34dc7c2f BB |
282 | |
283 | /* | |
b128c09f BB |
284 | * We sync when we're scrubbing, there's someone waiting |
285 | * on us, or the quiesce thread has handed off a txg to | |
286 | * us, or we have reached our timeout. | |
34dc7c2f BB |
287 | */ |
288 | timer = (delta >= timeout ? 0 : timeout - delta); | |
b128c09f BB |
289 | while ((dp->dp_scrub_func == SCRUB_FUNC_NONE || |
290 | spa_shutting_down(dp->dp_spa)) && | |
291 | !tx->tx_exiting && timer > 0 && | |
34dc7c2f BB |
292 | tx->tx_synced_txg >= tx->tx_sync_txg_waiting && |
293 | tx->tx_quiesced_txg == 0) { | |
294 | dprintf("waiting; tx_synced=%llu waiting=%llu dp=%p\n", | |
295 | tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp); | |
296 | txg_thread_wait(tx, &cpr, &tx->tx_sync_more_cv, timer); | |
297 | delta = lbolt - start; | |
298 | timer = (delta > timeout ? 0 : timeout - delta); | |
299 | } | |
300 | ||
301 | /* | |
302 | * Wait until the quiesce thread hands off a txg to us, | |
303 | * prompting it to do so if necessary. | |
304 | */ | |
305 | while (!tx->tx_exiting && tx->tx_quiesced_txg == 0) { | |
306 | if (tx->tx_quiesce_txg_waiting < tx->tx_open_txg+1) | |
307 | tx->tx_quiesce_txg_waiting = tx->tx_open_txg+1; | |
308 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
309 | txg_thread_wait(tx, &cpr, &tx->tx_quiesce_done_cv, 0); | |
310 | } | |
311 | ||
312 | if (tx->tx_exiting) | |
313 | txg_thread_exit(tx, &cpr, &tx->tx_sync_thread); | |
314 | ||
315 | rw_enter(&tx->tx_suspend, RW_WRITER); | |
316 | ||
317 | /* | |
318 | * Consume the quiesced txg which has been handed off to | |
319 | * us. This may cause the quiescing thread to now be | |
320 | * able to quiesce another txg, so we must signal it. | |
321 | */ | |
322 | txg = tx->tx_quiesced_txg; | |
323 | tx->tx_quiesced_txg = 0; | |
324 | tx->tx_syncing_txg = txg; | |
325 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
326 | rw_exit(&tx->tx_suspend); | |
327 | ||
328 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
329 | txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); | |
330 | mutex_exit(&tx->tx_sync_lock); | |
b128c09f | 331 | |
34dc7c2f BB |
332 | start = lbolt; |
333 | spa_sync(dp->dp_spa, txg); | |
334 | delta = lbolt - start; | |
335 | ||
34dc7c2f BB |
336 | mutex_enter(&tx->tx_sync_lock); |
337 | rw_enter(&tx->tx_suspend, RW_WRITER); | |
338 | tx->tx_synced_txg = txg; | |
339 | tx->tx_syncing_txg = 0; | |
340 | rw_exit(&tx->tx_suspend); | |
341 | cv_broadcast(&tx->tx_sync_done_cv); | |
342 | } | |
343 | } | |
344 | ||
345 | static void | |
346 | txg_quiesce_thread(dsl_pool_t *dp) | |
347 | { | |
348 | tx_state_t *tx = &dp->dp_tx; | |
349 | callb_cpr_t cpr; | |
350 | ||
351 | txg_thread_enter(tx, &cpr); | |
352 | ||
353 | for (;;) { | |
354 | uint64_t txg; | |
355 | ||
356 | /* | |
357 | * We quiesce when there's someone waiting on us. | |
358 | * However, we can only have one txg in "quiescing" or | |
359 | * "quiesced, waiting to sync" state. So we wait until | |
360 | * the "quiesced, waiting to sync" txg has been consumed | |
361 | * by the sync thread. | |
362 | */ | |
363 | while (!tx->tx_exiting && | |
364 | (tx->tx_open_txg >= tx->tx_quiesce_txg_waiting || | |
365 | tx->tx_quiesced_txg != 0)) | |
366 | txg_thread_wait(tx, &cpr, &tx->tx_quiesce_more_cv, 0); | |
367 | ||
368 | if (tx->tx_exiting) | |
369 | txg_thread_exit(tx, &cpr, &tx->tx_quiesce_thread); | |
370 | ||
371 | txg = tx->tx_open_txg; | |
372 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
373 | txg, tx->tx_quiesce_txg_waiting, | |
374 | tx->tx_sync_txg_waiting); | |
375 | mutex_exit(&tx->tx_sync_lock); | |
376 | txg_quiesce(dp, txg); | |
377 | mutex_enter(&tx->tx_sync_lock); | |
378 | ||
379 | /* | |
380 | * Hand this txg off to the sync thread. | |
381 | */ | |
382 | dprintf("quiesce done, handing off txg %llu\n", txg); | |
383 | tx->tx_quiesced_txg = txg; | |
384 | cv_broadcast(&tx->tx_sync_more_cv); | |
385 | cv_broadcast(&tx->tx_quiesce_done_cv); | |
386 | } | |
387 | } | |
388 | ||
389 | /* | |
390 | * Delay this thread by 'ticks' if we are still in the open transaction | |
391 | * group and there is already a waiting txg quiesing or quiesced. Abort | |
392 | * the delay if this txg stalls or enters the quiesing state. | |
393 | */ | |
394 | void | |
395 | txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks) | |
396 | { | |
397 | tx_state_t *tx = &dp->dp_tx; | |
398 | int timeout = lbolt + ticks; | |
399 | ||
400 | /* don't delay if this txg could transition to quiesing immediately */ | |
401 | if (tx->tx_open_txg > txg || | |
402 | tx->tx_syncing_txg == txg-1 || tx->tx_synced_txg == txg-1) | |
403 | return; | |
404 | ||
405 | mutex_enter(&tx->tx_sync_lock); | |
406 | if (tx->tx_open_txg > txg || tx->tx_synced_txg == txg-1) { | |
407 | mutex_exit(&tx->tx_sync_lock); | |
408 | return; | |
409 | } | |
410 | ||
411 | while (lbolt < timeout && | |
412 | tx->tx_syncing_txg < txg-1 && !txg_stalled(dp)) | |
413 | (void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock, | |
414 | timeout); | |
415 | ||
416 | mutex_exit(&tx->tx_sync_lock); | |
417 | } | |
418 | ||
419 | void | |
420 | txg_wait_synced(dsl_pool_t *dp, uint64_t txg) | |
421 | { | |
422 | tx_state_t *tx = &dp->dp_tx; | |
423 | ||
424 | mutex_enter(&tx->tx_sync_lock); | |
425 | ASSERT(tx->tx_threads == 2); | |
426 | if (txg == 0) | |
427 | txg = tx->tx_open_txg; | |
428 | if (tx->tx_sync_txg_waiting < txg) | |
429 | tx->tx_sync_txg_waiting = txg; | |
430 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
431 | txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); | |
432 | while (tx->tx_synced_txg < txg) { | |
433 | dprintf("broadcasting sync more " | |
434 | "tx_synced=%llu waiting=%llu dp=%p\n", | |
435 | tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp); | |
436 | cv_broadcast(&tx->tx_sync_more_cv); | |
437 | cv_wait(&tx->tx_sync_done_cv, &tx->tx_sync_lock); | |
438 | } | |
439 | mutex_exit(&tx->tx_sync_lock); | |
440 | } | |
441 | ||
442 | void | |
443 | txg_wait_open(dsl_pool_t *dp, uint64_t txg) | |
444 | { | |
445 | tx_state_t *tx = &dp->dp_tx; | |
446 | ||
447 | mutex_enter(&tx->tx_sync_lock); | |
448 | ASSERT(tx->tx_threads == 2); | |
449 | if (txg == 0) | |
450 | txg = tx->tx_open_txg + 1; | |
451 | if (tx->tx_quiesce_txg_waiting < txg) | |
452 | tx->tx_quiesce_txg_waiting = txg; | |
453 | dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", | |
454 | txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); | |
455 | while (tx->tx_open_txg < txg) { | |
456 | cv_broadcast(&tx->tx_quiesce_more_cv); | |
457 | cv_wait(&tx->tx_quiesce_done_cv, &tx->tx_sync_lock); | |
458 | } | |
459 | mutex_exit(&tx->tx_sync_lock); | |
460 | } | |
461 | ||
b128c09f | 462 | boolean_t |
34dc7c2f BB |
463 | txg_stalled(dsl_pool_t *dp) |
464 | { | |
465 | tx_state_t *tx = &dp->dp_tx; | |
466 | return (tx->tx_quiesce_txg_waiting > tx->tx_open_txg); | |
467 | } | |
468 | ||
b128c09f BB |
469 | boolean_t |
470 | txg_sync_waiting(dsl_pool_t *dp) | |
471 | { | |
472 | tx_state_t *tx = &dp->dp_tx; | |
473 | ||
474 | return (tx->tx_syncing_txg <= tx->tx_sync_txg_waiting || | |
475 | tx->tx_quiesced_txg != 0); | |
476 | } | |
477 | ||
34dc7c2f BB |
478 | void |
479 | txg_suspend(dsl_pool_t *dp) | |
480 | { | |
481 | tx_state_t *tx = &dp->dp_tx; | |
482 | /* XXX some code paths suspend when they are already suspended! */ | |
483 | rw_enter(&tx->tx_suspend, RW_READER); | |
484 | } | |
485 | ||
486 | void | |
487 | txg_resume(dsl_pool_t *dp) | |
488 | { | |
489 | tx_state_t *tx = &dp->dp_tx; | |
490 | rw_exit(&tx->tx_suspend); | |
491 | } | |
492 | ||
493 | /* | |
494 | * Per-txg object lists. | |
495 | */ | |
496 | void | |
497 | txg_list_create(txg_list_t *tl, size_t offset) | |
498 | { | |
499 | int t; | |
500 | ||
501 | mutex_init(&tl->tl_lock, NULL, MUTEX_DEFAULT, NULL); | |
502 | ||
503 | tl->tl_offset = offset; | |
504 | ||
505 | for (t = 0; t < TXG_SIZE; t++) | |
506 | tl->tl_head[t] = NULL; | |
507 | } | |
508 | ||
509 | void | |
510 | txg_list_destroy(txg_list_t *tl) | |
511 | { | |
512 | int t; | |
513 | ||
514 | for (t = 0; t < TXG_SIZE; t++) | |
515 | ASSERT(txg_list_empty(tl, t)); | |
516 | ||
517 | mutex_destroy(&tl->tl_lock); | |
518 | } | |
519 | ||
520 | int | |
521 | txg_list_empty(txg_list_t *tl, uint64_t txg) | |
522 | { | |
523 | return (tl->tl_head[txg & TXG_MASK] == NULL); | |
524 | } | |
525 | ||
526 | /* | |
527 | * Add an entry to the list. | |
528 | * Returns 0 if it's a new entry, 1 if it's already there. | |
529 | */ | |
530 | int | |
531 | txg_list_add(txg_list_t *tl, void *p, uint64_t txg) | |
532 | { | |
533 | int t = txg & TXG_MASK; | |
534 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
535 | int already_on_list; | |
536 | ||
537 | mutex_enter(&tl->tl_lock); | |
538 | already_on_list = tn->tn_member[t]; | |
539 | if (!already_on_list) { | |
540 | tn->tn_member[t] = 1; | |
541 | tn->tn_next[t] = tl->tl_head[t]; | |
542 | tl->tl_head[t] = tn; | |
543 | } | |
544 | mutex_exit(&tl->tl_lock); | |
545 | ||
546 | return (already_on_list); | |
547 | } | |
548 | ||
549 | /* | |
550 | * Remove the head of the list and return it. | |
551 | */ | |
552 | void * | |
553 | txg_list_remove(txg_list_t *tl, uint64_t txg) | |
554 | { | |
555 | int t = txg & TXG_MASK; | |
556 | txg_node_t *tn; | |
557 | void *p = NULL; | |
558 | ||
559 | mutex_enter(&tl->tl_lock); | |
560 | if ((tn = tl->tl_head[t]) != NULL) { | |
561 | p = (char *)tn - tl->tl_offset; | |
562 | tl->tl_head[t] = tn->tn_next[t]; | |
563 | tn->tn_next[t] = NULL; | |
564 | tn->tn_member[t] = 0; | |
565 | } | |
566 | mutex_exit(&tl->tl_lock); | |
567 | ||
568 | return (p); | |
569 | } | |
570 | ||
571 | /* | |
572 | * Remove a specific item from the list and return it. | |
573 | */ | |
574 | void * | |
575 | txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg) | |
576 | { | |
577 | int t = txg & TXG_MASK; | |
578 | txg_node_t *tn, **tp; | |
579 | ||
580 | mutex_enter(&tl->tl_lock); | |
581 | ||
582 | for (tp = &tl->tl_head[t]; (tn = *tp) != NULL; tp = &tn->tn_next[t]) { | |
583 | if ((char *)tn - tl->tl_offset == p) { | |
584 | *tp = tn->tn_next[t]; | |
585 | tn->tn_next[t] = NULL; | |
586 | tn->tn_member[t] = 0; | |
587 | mutex_exit(&tl->tl_lock); | |
588 | return (p); | |
589 | } | |
590 | } | |
591 | ||
592 | mutex_exit(&tl->tl_lock); | |
593 | ||
594 | return (NULL); | |
595 | } | |
596 | ||
597 | int | |
598 | txg_list_member(txg_list_t *tl, void *p, uint64_t txg) | |
599 | { | |
600 | int t = txg & TXG_MASK; | |
601 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
602 | ||
603 | return (tn->tn_member[t]); | |
604 | } | |
605 | ||
606 | /* | |
607 | * Walk a txg list -- only safe if you know it's not changing. | |
608 | */ | |
609 | void * | |
610 | txg_list_head(txg_list_t *tl, uint64_t txg) | |
611 | { | |
612 | int t = txg & TXG_MASK; | |
613 | txg_node_t *tn = tl->tl_head[t]; | |
614 | ||
615 | return (tn == NULL ? NULL : (char *)tn - tl->tl_offset); | |
616 | } | |
617 | ||
618 | void * | |
619 | txg_list_next(txg_list_t *tl, void *p, uint64_t txg) | |
620 | { | |
621 | int t = txg & TXG_MASK; | |
622 | txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); | |
623 | ||
624 | tn = tn->tn_next[t]; | |
625 | ||
626 | return (tn == NULL ? NULL : (char *)tn - tl->tl_offset); | |
627 | } |