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 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 */ 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 cv_init(&tx->tx_sync_more_cv, NULL, CV_DEFAULT, NULL); 67 cv_init(&tx->tx_sync_done_cv, NULL, CV_DEFAULT, NULL); 68 cv_init(&tx->tx_quiesce_more_cv, NULL, CV_DEFAULT, NULL); 69 cv_init(&tx->tx_quiesce_done_cv, NULL, CV_DEFAULT, NULL); 70 cv_init(&tx->tx_exit_cv, NULL, CV_DEFAULT, NULL); 71 72 tx->tx_open_txg = txg; 73 } 74 75 /* 76 * Close down the txg subsystem. 77 */ 78 void 79 txg_fini(dsl_pool_t *dp) 80 { 81 tx_state_t *tx = &dp->dp_tx; 82 int c; 83 84 ASSERT(tx->tx_threads == 0); 85 86 rw_destroy(&tx->tx_suspend); 87 mutex_destroy(&tx->tx_sync_lock); 88 89 cv_destroy(&tx->tx_sync_more_cv); 90 cv_destroy(&tx->tx_sync_done_cv); 91 cv_destroy(&tx->tx_quiesce_more_cv); 92 cv_destroy(&tx->tx_quiesce_done_cv); 93 cv_destroy(&tx->tx_exit_cv); 94 95 for (c = 0; c < max_ncpus; c++) { 96 int i; 97 98 mutex_destroy(&tx->tx_cpu[c].tc_lock); 99 for (i = 0; i < TXG_SIZE; i++) 100 cv_destroy(&tx->tx_cpu[c].tc_cv[i]); 101 } 102 103 kmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t)); 104 105 bzero(tx, sizeof (tx_state_t)); 106 } 107 108 /* 109 * Start syncing transaction groups. 110 */ 111 void 112 txg_sync_start(dsl_pool_t *dp) 113 { 114 tx_state_t *tx = &dp->dp_tx; 115 116 mutex_enter(&tx->tx_sync_lock); 117 118 dprintf("pool %p\n", dp); 119 120 ASSERT(tx->tx_threads == 0); 121 122 tx->tx_threads = 2; 123 124 tx->tx_quiesce_thread = thread_create(NULL, 0, txg_quiesce_thread, 125 dp, 0, &p0, TS_RUN, minclsyspri); 126 127 /* 128 * The sync thread can need a larger-than-default stack size on 129 * 32-bit x86. This is due in part to nested pools and 130 * scrub_visitbp() recursion. 131 */ 132 tx->tx_sync_thread = thread_create(NULL, 12<<10, txg_sync_thread, 133 dp, 0, &p0, TS_RUN, minclsyspri); 134 135 mutex_exit(&tx->tx_sync_lock); 136 } 137 138 static void 139 txg_thread_enter(tx_state_t *tx, callb_cpr_t *cpr) 140 { 141 CALLB_CPR_INIT(cpr, &tx->tx_sync_lock, callb_generic_cpr, FTAG); 142 mutex_enter(&tx->tx_sync_lock); 143 } 144 145 static void 146 txg_thread_exit(tx_state_t *tx, callb_cpr_t *cpr, kthread_t **tpp) 147 { 148 ASSERT(*tpp != NULL); 149 *tpp = NULL; 150 tx->tx_threads--; 151 cv_broadcast(&tx->tx_exit_cv); 152 CALLB_CPR_EXIT(cpr); /* drops &tx->tx_sync_lock */ 153 thread_exit(); 154 } 155 156 static void 157 txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time) 158 { 159 CALLB_CPR_SAFE_BEGIN(cpr); 160 161 if (time) 162 (void) cv_timedwait(cv, &tx->tx_sync_lock, lbolt + time); 163 else 164 cv_wait(cv, &tx->tx_sync_lock); 165 166 CALLB_CPR_SAFE_END(cpr, &tx->tx_sync_lock); 167 } 168 169 /* 170 * Stop syncing transaction groups. 171 */ 172 void 173 txg_sync_stop(dsl_pool_t *dp) 174 { 175 tx_state_t *tx = &dp->dp_tx; 176 177 dprintf("pool %p\n", dp); 178 /* 179 * Finish off any work in progress. 180 */ 181 ASSERT(tx->tx_threads == 2); 182 txg_wait_synced(dp, 0); 183 184 /* 185 * Wake all sync threads and wait for them to die. 186 */ 187 mutex_enter(&tx->tx_sync_lock); 188 189 ASSERT(tx->tx_threads == 2); 190 191 tx->tx_exiting = 1; 192 193 cv_broadcast(&tx->tx_quiesce_more_cv); 194 cv_broadcast(&tx->tx_quiesce_done_cv); 195 cv_broadcast(&tx->tx_sync_more_cv); 196 197 while (tx->tx_threads != 0) 198 cv_wait(&tx->tx_exit_cv, &tx->tx_sync_lock); 199 200 tx->tx_exiting = 0; 201 202 mutex_exit(&tx->tx_sync_lock); 203 } 204 205 uint64_t 206 txg_hold_open(dsl_pool_t *dp, txg_handle_t *th) 207 { 208 tx_state_t *tx = &dp->dp_tx; 209 tx_cpu_t *tc = &tx->tx_cpu[CPU_SEQID]; 210 uint64_t txg; 211 212 mutex_enter(&tc->tc_lock); 213 214 txg = tx->tx_open_txg; 215 tc->tc_count[txg & TXG_MASK]++; 216 217 th->th_cpu = tc; 218 th->th_txg = txg; 219 220 return (txg); 221 } 222 223 void 224 txg_rele_to_quiesce(txg_handle_t *th) 225 { 226 tx_cpu_t *tc = th->th_cpu; 227 228 mutex_exit(&tc->tc_lock); 229 } 230 231 void 232 txg_rele_to_sync(txg_handle_t *th) 233 { 234 tx_cpu_t *tc = th->th_cpu; 235 int g = th->th_txg & TXG_MASK; 236 237 mutex_enter(&tc->tc_lock); 238 ASSERT(tc->tc_count[g] != 0); 239 if (--tc->tc_count[g] == 0) 240 cv_broadcast(&tc->tc_cv[g]); 241 mutex_exit(&tc->tc_lock); 242 243 th->th_cpu = NULL; /* defensive */ 244 } 245 246 static void 247 txg_quiesce(dsl_pool_t *dp, uint64_t txg) 248 { 249 tx_state_t *tx = &dp->dp_tx; 250 int g = txg & TXG_MASK; 251 int c; 252 253 /* 254 * Grab all tx_cpu locks so nobody else can get into this txg. 255 */ 256 for (c = 0; c < max_ncpus; c++) 257 mutex_enter(&tx->tx_cpu[c].tc_lock); 258 259 ASSERT(txg == tx->tx_open_txg); 260 tx->tx_open_txg++; 261 262 /* 263 * Now that we've incremented tx_open_txg, we can let threads 264 * enter the next transaction group. 265 */ 266 for (c = 0; c < max_ncpus; c++) 267 mutex_exit(&tx->tx_cpu[c].tc_lock); 268 269 /* 270 * Quiesce the transaction group by waiting for everyone to txg_exit(). 271 */ 272 for (c = 0; c < max_ncpus; c++) { 273 tx_cpu_t *tc = &tx->tx_cpu[c]; 274 mutex_enter(&tc->tc_lock); 275 while (tc->tc_count[g] != 0) 276 cv_wait(&tc->tc_cv[g], &tc->tc_lock); 277 mutex_exit(&tc->tc_lock); 278 } 279 } 280 281 static void 282 txg_sync_thread(dsl_pool_t *dp) 283 { 284 tx_state_t *tx = &dp->dp_tx; 285 callb_cpr_t cpr; 286 uint64_t start, delta; 287 288 txg_thread_enter(tx, &cpr); 289 290 start = delta = 0; 291 for (;;) { 292 uint64_t timer, timeout = zfs_txg_timeout * hz; 293 uint64_t txg; 294 295 /* 296 * We sync when we're scrubbing, there's someone waiting 297 * on us, or the quiesce thread has handed off a txg to 298 * us, or we have reached our timeout. 299 */ 300 timer = (delta >= timeout ? 0 : timeout - delta); 301 while ((dp->dp_scrub_func == SCRUB_FUNC_NONE || 302 spa_shutting_down(dp->dp_spa)) && 303 !tx->tx_exiting && timer > 0 && 304 tx->tx_synced_txg >= tx->tx_sync_txg_waiting && 305 tx->tx_quiesced_txg == 0) { 306 dprintf("waiting; tx_synced=%llu waiting=%llu dp=%p\n", 307 tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp); 308 txg_thread_wait(tx, &cpr, &tx->tx_sync_more_cv, timer); 309 delta = lbolt - start; 310 timer = (delta > timeout ? 0 : timeout - delta); 311 } 312 313 /* 314 * Wait until the quiesce thread hands off a txg to us, 315 * prompting it to do so if necessary. 316 */ 317 while (!tx->tx_exiting && tx->tx_quiesced_txg == 0) { 318 if (tx->tx_quiesce_txg_waiting < tx->tx_open_txg+1) 319 tx->tx_quiesce_txg_waiting = tx->tx_open_txg+1; 320 cv_broadcast(&tx->tx_quiesce_more_cv); 321 txg_thread_wait(tx, &cpr, &tx->tx_quiesce_done_cv, 0); 322 } 323 324 if (tx->tx_exiting) 325 txg_thread_exit(tx, &cpr, &tx->tx_sync_thread); 326 327 rw_enter(&tx->tx_suspend, RW_WRITER); 328 329 /* 330 * Consume the quiesced txg which has been handed off to 331 * us. This may cause the quiescing thread to now be 332 * able to quiesce another txg, so we must signal it. 333 */ 334 txg = tx->tx_quiesced_txg; 335 tx->tx_quiesced_txg = 0; 336 tx->tx_syncing_txg = txg; 337 cv_broadcast(&tx->tx_quiesce_more_cv); 338 rw_exit(&tx->tx_suspend); 339 340 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", 341 txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); 342 mutex_exit(&tx->tx_sync_lock); 343 344 start = lbolt; 345 spa_sync(dp->dp_spa, txg); 346 delta = lbolt - start; 347 348 mutex_enter(&tx->tx_sync_lock); 349 rw_enter(&tx->tx_suspend, RW_WRITER); 350 tx->tx_synced_txg = txg; 351 tx->tx_syncing_txg = 0; 352 rw_exit(&tx->tx_suspend); 353 cv_broadcast(&tx->tx_sync_done_cv); 354 } 355 } 356 357 static void 358 txg_quiesce_thread(dsl_pool_t *dp) 359 { 360 tx_state_t *tx = &dp->dp_tx; 361 callb_cpr_t cpr; 362 363 txg_thread_enter(tx, &cpr); 364 365 for (;;) { 366 uint64_t txg; 367 368 /* 369 * We quiesce when there's someone waiting on us. 370 * However, we can only have one txg in "quiescing" or 371 * "quiesced, waiting to sync" state. So we wait until 372 * the "quiesced, waiting to sync" txg has been consumed 373 * by the sync thread. 374 */ 375 while (!tx->tx_exiting && 376 (tx->tx_open_txg >= tx->tx_quiesce_txg_waiting || 377 tx->tx_quiesced_txg != 0)) 378 txg_thread_wait(tx, &cpr, &tx->tx_quiesce_more_cv, 0); 379 380 if (tx->tx_exiting) 381 txg_thread_exit(tx, &cpr, &tx->tx_quiesce_thread); 382 383 txg = tx->tx_open_txg; 384 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", 385 txg, tx->tx_quiesce_txg_waiting, 386 tx->tx_sync_txg_waiting); 387 mutex_exit(&tx->tx_sync_lock); 388 txg_quiesce(dp, txg); 389 mutex_enter(&tx->tx_sync_lock); 390 391 /* 392 * Hand this txg off to the sync thread. 393 */ 394 dprintf("quiesce done, handing off txg %llu\n", txg); 395 tx->tx_quiesced_txg = txg; 396 cv_broadcast(&tx->tx_sync_more_cv); 397 cv_broadcast(&tx->tx_quiesce_done_cv); 398 } 399 } 400 401 /* 402 * Delay this thread by 'ticks' if we are still in the open transaction 403 * group and there is already a waiting txg quiesing or quiesced. Abort 404 * the delay if this txg stalls or enters the quiesing state. 405 */ 406 void 407 txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks) 408 { 409 tx_state_t *tx = &dp->dp_tx; 410 int timeout = lbolt + ticks; 411 412 /* don't delay if this txg could transition to quiesing immediately */ 413 if (tx->tx_open_txg > txg || 414 tx->tx_syncing_txg == txg-1 || tx->tx_synced_txg == txg-1) 415 return; 416 417 mutex_enter(&tx->tx_sync_lock); 418 if (tx->tx_open_txg > txg || tx->tx_synced_txg == txg-1) { 419 mutex_exit(&tx->tx_sync_lock); 420 return; 421 } 422 423 while (lbolt < timeout && 424 tx->tx_syncing_txg < txg-1 && !txg_stalled(dp)) 425 (void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock, 426 timeout); 427 428 mutex_exit(&tx->tx_sync_lock); 429 } 430 431 void 432 txg_wait_synced(dsl_pool_t *dp, uint64_t txg) 433 { 434 tx_state_t *tx = &dp->dp_tx; 435 436 mutex_enter(&tx->tx_sync_lock); 437 ASSERT(tx->tx_threads == 2); 438 if (txg == 0) 439 txg = tx->tx_open_txg; 440 if (tx->tx_sync_txg_waiting < txg) 441 tx->tx_sync_txg_waiting = txg; 442 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", 443 txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); 444 while (tx->tx_synced_txg < txg) { 445 dprintf("broadcasting sync more " 446 "tx_synced=%llu waiting=%llu dp=%p\n", 447 tx->tx_synced_txg, tx->tx_sync_txg_waiting, dp); 448 cv_broadcast(&tx->tx_sync_more_cv); 449 cv_wait(&tx->tx_sync_done_cv, &tx->tx_sync_lock); 450 } 451 mutex_exit(&tx->tx_sync_lock); 452 } 453 454 void 455 txg_wait_open(dsl_pool_t *dp, uint64_t txg) 456 { 457 tx_state_t *tx = &dp->dp_tx; 458 459 mutex_enter(&tx->tx_sync_lock); 460 ASSERT(tx->tx_threads == 2); 461 if (txg == 0) 462 txg = tx->tx_open_txg + 1; 463 if (tx->tx_quiesce_txg_waiting < txg) 464 tx->tx_quiesce_txg_waiting = txg; 465 dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n", 466 txg, tx->tx_quiesce_txg_waiting, tx->tx_sync_txg_waiting); 467 while (tx->tx_open_txg < txg) { 468 cv_broadcast(&tx->tx_quiesce_more_cv); 469 cv_wait(&tx->tx_quiesce_done_cv, &tx->tx_sync_lock); 470 } 471 mutex_exit(&tx->tx_sync_lock); 472 } 473 474 boolean_t 475 txg_stalled(dsl_pool_t *dp) 476 { 477 tx_state_t *tx = &dp->dp_tx; 478 return (tx->tx_quiesce_txg_waiting > tx->tx_open_txg); 479 } 480 481 boolean_t 482 txg_sync_waiting(dsl_pool_t *dp) 483 { 484 tx_state_t *tx = &dp->dp_tx; 485 486 return (tx->tx_syncing_txg <= tx->tx_sync_txg_waiting || 487 tx->tx_quiesced_txg != 0); 488 } 489 490 void 491 txg_suspend(dsl_pool_t *dp) 492 { 493 tx_state_t *tx = &dp->dp_tx; 494 /* XXX some code paths suspend when they are already suspended! */ 495 rw_enter(&tx->tx_suspend, RW_READER); 496 } 497 498 void 499 txg_resume(dsl_pool_t *dp) 500 { 501 tx_state_t *tx = &dp->dp_tx; 502 rw_exit(&tx->tx_suspend); 503 } 504 505 /* 506 * Per-txg object lists. 507 */ 508 void 509 txg_list_create(txg_list_t *tl, size_t offset) 510 { 511 int t; 512 513 mutex_init(&tl->tl_lock, NULL, MUTEX_DEFAULT, NULL); 514 515 tl->tl_offset = offset; 516 517 for (t = 0; t < TXG_SIZE; t++) 518 tl->tl_head[t] = NULL; 519 } 520 521 void 522 txg_list_destroy(txg_list_t *tl) 523 { 524 int t; 525 526 for (t = 0; t < TXG_SIZE; t++) 527 ASSERT(txg_list_empty(tl, t)); 528 529 mutex_destroy(&tl->tl_lock); 530 } 531 532 int 533 txg_list_empty(txg_list_t *tl, uint64_t txg) 534 { 535 return (tl->tl_head[txg & TXG_MASK] == NULL); 536 } 537 538 /* 539 * Add an entry to the list. 540 * Returns 0 if it's a new entry, 1 if it's already there. 541 */ 542 int 543 txg_list_add(txg_list_t *tl, void *p, uint64_t txg) 544 { 545 int t = txg & TXG_MASK; 546 txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); 547 int already_on_list; 548 549 mutex_enter(&tl->tl_lock); 550 already_on_list = tn->tn_member[t]; 551 if (!already_on_list) { 552 tn->tn_member[t] = 1; 553 tn->tn_next[t] = tl->tl_head[t]; 554 tl->tl_head[t] = tn; 555 } 556 mutex_exit(&tl->tl_lock); 557 558 return (already_on_list); 559 } 560 561 /* 562 * Remove the head of the list and return it. 563 */ 564 void * 565 txg_list_remove(txg_list_t *tl, uint64_t txg) 566 { 567 int t = txg & TXG_MASK; 568 txg_node_t *tn; 569 void *p = NULL; 570 571 mutex_enter(&tl->tl_lock); 572 if ((tn = tl->tl_head[t]) != NULL) { 573 p = (char *)tn - tl->tl_offset; 574 tl->tl_head[t] = tn->tn_next[t]; 575 tn->tn_next[t] = NULL; 576 tn->tn_member[t] = 0; 577 } 578 mutex_exit(&tl->tl_lock); 579 580 return (p); 581 } 582 583 /* 584 * Remove a specific item from the list and return it. 585 */ 586 void * 587 txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg) 588 { 589 int t = txg & TXG_MASK; 590 txg_node_t *tn, **tp; 591 592 mutex_enter(&tl->tl_lock); 593 594 for (tp = &tl->tl_head[t]; (tn = *tp) != NULL; tp = &tn->tn_next[t]) { 595 if ((char *)tn - tl->tl_offset == p) { 596 *tp = tn->tn_next[t]; 597 tn->tn_next[t] = NULL; 598 tn->tn_member[t] = 0; 599 mutex_exit(&tl->tl_lock); 600 return (p); 601 } 602 } 603 604 mutex_exit(&tl->tl_lock); 605 606 return (NULL); 607 } 608 609 int 610 txg_list_member(txg_list_t *tl, void *p, uint64_t txg) 611 { 612 int t = txg & TXG_MASK; 613 txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); 614 615 return (tn->tn_member[t]); 616 } 617 618 /* 619 * Walk a txg list -- only safe if you know it's not changing. 620 */ 621 void * 622 txg_list_head(txg_list_t *tl, uint64_t txg) 623 { 624 int t = txg & TXG_MASK; 625 txg_node_t *tn = tl->tl_head[t]; 626 627 return (tn == NULL ? NULL : (char *)tn - tl->tl_offset); 628 } 629 630 void * 631 txg_list_next(txg_list_t *tl, void *p, uint64_t txg) 632 { 633 int t = txg & TXG_MASK; 634 txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset); 635 636 tn = tn->tn_next[t]; 637 638 return (tn == NULL ? NULL : (char *)tn - tl->tl_offset); 639 } 640