1 /* 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2022 Tomohiro Kusumi <tkusumi@netbsd.org> 5 * Copyright (c) 2011-2022 The DragonFly Project. All rights reserved. 6 * 7 * This code is derived from software contributed to The DragonFly Project 8 * by Matthew Dillon <dillon@dragonflybsd.org> 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in 18 * the documentation and/or other materials provided with the 19 * distribution. 20 * 3. Neither the name of The DragonFly Project nor the names of its 21 * contributors may be used to endorse or promote products derived 22 * from this software without specific, prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 25 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 26 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 27 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 28 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 29 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 30 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 31 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 32 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 33 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 34 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 /* 38 * This module implements the hammer2 helper thread API, including 39 * the frontend/backend XOP API. 40 */ 41 #include "hammer2.h" 42 43 #define H2XOPDESCRIPTOR(label) \ 44 hammer2_xop_desc_t hammer2_##label##_desc = { \ 45 .storage_func = hammer2_xop_##label, \ 46 .id = #label \ 47 } 48 49 H2XOPDESCRIPTOR(ipcluster); 50 H2XOPDESCRIPTOR(readdir); 51 H2XOPDESCRIPTOR(nresolve); 52 H2XOPDESCRIPTOR(unlink); 53 H2XOPDESCRIPTOR(nrename); 54 H2XOPDESCRIPTOR(scanlhc); 55 H2XOPDESCRIPTOR(scanall); 56 H2XOPDESCRIPTOR(lookup); 57 H2XOPDESCRIPTOR(delete); 58 H2XOPDESCRIPTOR(inode_mkdirent); 59 H2XOPDESCRIPTOR(inode_create); 60 H2XOPDESCRIPTOR(inode_create_det); 61 H2XOPDESCRIPTOR(inode_create_ins); 62 H2XOPDESCRIPTOR(inode_destroy); 63 H2XOPDESCRIPTOR(inode_chain_sync); 64 H2XOPDESCRIPTOR(inode_unlinkall); 65 H2XOPDESCRIPTOR(inode_connect); 66 H2XOPDESCRIPTOR(inode_flush); 67 H2XOPDESCRIPTOR(strategy_read); 68 H2XOPDESCRIPTOR(strategy_write); 69 70 //struct objcache *cache_xops; 71 static struct thread dummy_td; 72 struct thread *curthread = &dummy_td; 73 74 /* 75 * Set flags and wakeup any waiters. 76 * 77 * WARNING! During teardown (thr) can disappear the instant our cmpset 78 * succeeds. 79 */ 80 void 81 hammer2_thr_signal(hammer2_thread_t *thr, uint32_t flags) 82 { 83 uint32_t oflags; 84 uint32_t nflags; 85 86 for (;;) { 87 oflags = thr->flags; 88 cpu_ccfence(); 89 nflags = (oflags | flags) & ~HAMMER2_THREAD_WAITING; 90 91 if (oflags & HAMMER2_THREAD_WAITING) { 92 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) { 93 wakeup(&thr->flags); 94 break; 95 } 96 } else { 97 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) 98 break; 99 } 100 } 101 } 102 103 /* 104 * Set and clear flags and wakeup any waiters. 105 * 106 * WARNING! During teardown (thr) can disappear the instant our cmpset 107 * succeeds. 108 */ 109 void 110 hammer2_thr_signal2(hammer2_thread_t *thr, uint32_t posflags, uint32_t negflags) 111 { 112 uint32_t oflags; 113 uint32_t nflags; 114 115 for (;;) { 116 oflags = thr->flags; 117 cpu_ccfence(); 118 nflags = (oflags | posflags) & 119 ~(negflags | HAMMER2_THREAD_WAITING); 120 if (oflags & HAMMER2_THREAD_WAITING) { 121 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) { 122 wakeup(&thr->flags); 123 break; 124 } 125 } else { 126 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) 127 break; 128 } 129 } 130 } 131 132 /* 133 * Wait until all the bits in flags are set. 134 * 135 * WARNING! During teardown (thr) can disappear the instant our cmpset 136 * succeeds. 137 */ 138 void 139 hammer2_thr_wait(hammer2_thread_t *thr, uint32_t flags) 140 { 141 uint32_t oflags; 142 uint32_t nflags; 143 144 for (;;) { 145 oflags = thr->flags; 146 cpu_ccfence(); 147 if ((oflags & flags) == flags) 148 break; 149 nflags = oflags | HAMMER2_THREAD_WAITING; 150 tsleep_interlock(&thr->flags, 0); 151 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) { 152 tsleep(&thr->flags, PINTERLOCKED, "h2twait", hz*60); 153 } 154 } 155 } 156 157 /* 158 * Wait until any of the bits in flags are set, with timeout. 159 * 160 * WARNING! During teardown (thr) can disappear the instant our cmpset 161 * succeeds. 162 */ 163 int 164 hammer2_thr_wait_any(hammer2_thread_t *thr, uint32_t flags, int timo) 165 { 166 uint32_t oflags; 167 uint32_t nflags; 168 int error; 169 170 error = 0; 171 for (;;) { 172 oflags = thr->flags; 173 cpu_ccfence(); 174 if (oflags & flags) 175 break; 176 nflags = oflags | HAMMER2_THREAD_WAITING; 177 tsleep_interlock(&thr->flags, 0); 178 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) { 179 error = tsleep(&thr->flags, PINTERLOCKED, 180 "h2twait", timo); 181 } 182 if (error == ETIMEDOUT) { 183 error = HAMMER2_ERROR_ETIMEDOUT; 184 break; 185 } 186 } 187 return error; 188 } 189 190 /* 191 * Wait until the bits in flags are clear. 192 * 193 * WARNING! During teardown (thr) can disappear the instant our cmpset 194 * succeeds. 195 */ 196 void 197 hammer2_thr_wait_neg(hammer2_thread_t *thr, uint32_t flags) 198 { 199 uint32_t oflags; 200 uint32_t nflags; 201 202 for (;;) { 203 oflags = thr->flags; 204 cpu_ccfence(); 205 if ((oflags & flags) == 0) 206 break; 207 nflags = oflags | HAMMER2_THREAD_WAITING; 208 tsleep_interlock(&thr->flags, 0); 209 if (atomic_cmpset_int(&thr->flags, oflags, nflags)) { 210 tsleep(&thr->flags, PINTERLOCKED, "h2twait", hz*60); 211 } 212 } 213 } 214 215 /* 216 * Initialize the supplied thread structure, starting the specified 217 * thread. 218 * 219 * NOTE: thr structure can be retained across mounts and unmounts for this 220 * pmp, so make sure the flags are in a sane state. 221 */ 222 void 223 hammer2_thr_create(hammer2_thread_t *thr, hammer2_pfs_t *pmp, 224 hammer2_dev_t *hmp, 225 const char *id, int clindex, int repidx, 226 void (*func)(void *arg)) 227 { 228 thr->pmp = pmp; /* xop helpers */ 229 thr->hmp = hmp; /* bulkfree */ 230 thr->clindex = clindex; 231 thr->repidx = repidx; 232 TAILQ_INIT(&thr->xopq); 233 atomic_clear_int(&thr->flags, HAMMER2_THREAD_STOP | 234 HAMMER2_THREAD_STOPPED | 235 HAMMER2_THREAD_FREEZE | 236 HAMMER2_THREAD_FROZEN); 237 if (thr->scratch == NULL) 238 thr->scratch = kmalloc(MAXPHYS, M_HAMMER2, M_WAITOK | M_ZERO); 239 #if 0 240 if (repidx >= 0) { 241 lwkt_create(func, thr, &thr->td, NULL, 0, repidx % ncpus, 242 "%s-%s.%02d", id, pmp->pfs_names[clindex], repidx); 243 } else if (pmp) { 244 lwkt_create(func, thr, &thr->td, NULL, 0, -1, 245 "%s-%s", id, pmp->pfs_names[clindex]); 246 } else { 247 lwkt_create(func, thr, &thr->td, NULL, 0, -1, "%s", id); 248 } 249 #else 250 thr->td = &dummy_td; 251 #endif 252 } 253 254 /* 255 * Terminate a thread. This function will silently return if the thread 256 * was never initialized or has already been deleted. 257 * 258 * This is accomplished by setting the STOP flag and waiting for the td 259 * structure to become NULL. 260 */ 261 void 262 hammer2_thr_delete(hammer2_thread_t *thr) 263 { 264 if (thr->td == NULL) 265 return; 266 hammer2_thr_signal(thr, HAMMER2_THREAD_STOP); 267 /* Don't wait, there's no such thread in makefs */ 268 //hammer2_thr_wait(thr, HAMMER2_THREAD_STOPPED); 269 thr->pmp = NULL; 270 if (thr->scratch) { 271 kfree(thr->scratch, M_HAMMER2); 272 thr->scratch = NULL; 273 } 274 KKASSERT(TAILQ_EMPTY(&thr->xopq)); 275 } 276 277 /* 278 * Asynchronous remaster request. Ask the synchronization thread to 279 * start over soon (as if it were frozen and unfrozen, but without waiting). 280 * The thread always recalculates mastership relationships when restarting. 281 */ 282 void 283 hammer2_thr_remaster(hammer2_thread_t *thr) 284 { 285 if (thr->td == NULL) 286 return; 287 hammer2_thr_signal(thr, HAMMER2_THREAD_REMASTER); 288 } 289 290 void 291 hammer2_thr_freeze_async(hammer2_thread_t *thr) 292 { 293 hammer2_thr_signal(thr, HAMMER2_THREAD_FREEZE); 294 } 295 296 void 297 hammer2_thr_freeze(hammer2_thread_t *thr) 298 { 299 if (thr->td == NULL) 300 return; 301 hammer2_thr_signal(thr, HAMMER2_THREAD_FREEZE); 302 hammer2_thr_wait(thr, HAMMER2_THREAD_FROZEN); 303 } 304 305 void 306 hammer2_thr_unfreeze(hammer2_thread_t *thr) 307 { 308 if (thr->td == NULL) 309 return; 310 hammer2_thr_signal(thr, HAMMER2_THREAD_UNFREEZE); 311 hammer2_thr_wait_neg(thr, HAMMER2_THREAD_FROZEN); 312 } 313 314 int 315 hammer2_thr_break(hammer2_thread_t *thr) 316 { 317 if (thr->flags & (HAMMER2_THREAD_STOP | 318 HAMMER2_THREAD_REMASTER | 319 HAMMER2_THREAD_FREEZE)) { 320 return 1; 321 } 322 return 0; 323 } 324 325 /**************************************************************************** 326 * HAMMER2 XOPS API * 327 ****************************************************************************/ 328 329 /* 330 * Allocate a XOP request. 331 * 332 * Once allocated a XOP request can be started, collected, and retired, 333 * and can be retired early if desired. 334 * 335 * NOTE: Fifo indices might not be zero but ri == wi on objcache_get(). 336 */ 337 void * 338 hammer2_xop_alloc(hammer2_inode_t *ip, int flags) 339 { 340 hammer2_xop_t *xop; 341 342 xop = ecalloc(1, sizeof(*xop)); 343 KKASSERT(xop->head.cluster.array[0].chain == NULL); 344 345 xop->head.ip1 = ip; 346 xop->head.desc = NULL; 347 xop->head.flags = flags; 348 xop->head.state = 0; 349 xop->head.error = 0; 350 xop->head.collect_key = 0; 351 xop->head.focus_dio = NULL; 352 353 if (flags & HAMMER2_XOP_MODIFYING) 354 xop->head.mtid = hammer2_trans_sub(ip->pmp); 355 else 356 xop->head.mtid = 0; 357 358 xop->head.cluster.nchains = ip->cluster.nchains; 359 xop->head.cluster.pmp = ip->pmp; 360 xop->head.cluster.flags = HAMMER2_CLUSTER_LOCKED; 361 362 /* 363 * run_mask - Active thread (or frontend) associated with XOP 364 */ 365 xop->head.run_mask = HAMMER2_XOPMASK_VOP; 366 367 hammer2_inode_ref(ip); 368 369 return xop; 370 } 371 372 void 373 hammer2_xop_setname(hammer2_xop_head_t *xop, const char *name, size_t name_len) 374 { 375 xop->name1 = kmalloc(name_len + 1, M_HAMMER2, M_WAITOK | M_ZERO); 376 xop->name1_len = name_len; 377 bcopy(name, xop->name1, name_len); 378 } 379 380 void 381 hammer2_xop_setname2(hammer2_xop_head_t *xop, const char *name, size_t name_len) 382 { 383 xop->name2 = kmalloc(name_len + 1, M_HAMMER2, M_WAITOK | M_ZERO); 384 xop->name2_len = name_len; 385 bcopy(name, xop->name2, name_len); 386 } 387 388 size_t 389 hammer2_xop_setname_inum(hammer2_xop_head_t *xop, hammer2_key_t inum) 390 { 391 const size_t name_len = 18; 392 393 xop->name1 = kmalloc(name_len + 1, M_HAMMER2, M_WAITOK | M_ZERO); 394 xop->name1_len = name_len; 395 ksnprintf(xop->name1, name_len + 1, "0x%016jx", (intmax_t)inum); 396 397 return name_len; 398 } 399 400 401 void 402 hammer2_xop_setip2(hammer2_xop_head_t *xop, hammer2_inode_t *ip2) 403 { 404 xop->ip2 = ip2; 405 hammer2_inode_ref(ip2); 406 } 407 408 void 409 hammer2_xop_setip3(hammer2_xop_head_t *xop, hammer2_inode_t *ip3) 410 { 411 xop->ip3 = ip3; 412 hammer2_inode_ref(ip3); 413 } 414 415 void 416 hammer2_xop_setip4(hammer2_xop_head_t *xop, hammer2_inode_t *ip4) 417 { 418 xop->ip4 = ip4; 419 hammer2_inode_ref(ip4); 420 } 421 422 void 423 hammer2_xop_reinit(hammer2_xop_head_t *xop) 424 { 425 xop->state = 0; 426 xop->error = 0; 427 xop->collect_key = 0; 428 xop->run_mask = HAMMER2_XOPMASK_VOP; 429 } 430 431 /* 432 * A mounted PFS needs Xops threads to support frontend operations. 433 */ 434 void 435 hammer2_xop_helper_create(hammer2_pfs_t *pmp) 436 { 437 int i; 438 int j; 439 440 lockmgr(&pmp->lock, LK_EXCLUSIVE); 441 pmp->has_xop_threads = 1; 442 443 pmp->xop_groups = kmalloc(hammer2_xop_nthreads * 444 sizeof(hammer2_xop_group_t), 445 M_HAMMER2, M_WAITOK | M_ZERO); 446 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) { 447 for (j = 0; j < hammer2_xop_nthreads; ++j) { 448 if (pmp->xop_groups[j].thrs[i].td) 449 continue; 450 hammer2_thr_create(&pmp->xop_groups[j].thrs[i], 451 pmp, NULL, 452 "h2xop", i, j, 453 hammer2_primary_xops_thread); 454 } 455 } 456 lockmgr(&pmp->lock, LK_RELEASE); 457 } 458 459 void 460 hammer2_xop_helper_cleanup(hammer2_pfs_t *pmp) 461 { 462 int i; 463 int j; 464 465 if (pmp->xop_groups == NULL) { 466 KKASSERT(pmp->has_xop_threads == 0); 467 return; 468 } 469 470 for (i = 0; i < pmp->pfs_nmasters; ++i) { 471 for (j = 0; j < hammer2_xop_nthreads; ++j) { 472 if (pmp->xop_groups[j].thrs[i].td) 473 hammer2_thr_delete(&pmp->xop_groups[j].thrs[i]); 474 } 475 } 476 pmp->has_xop_threads = 0; 477 kfree(pmp->xop_groups, M_HAMMER2); 478 pmp->xop_groups = NULL; 479 } 480 481 /* 482 * Start a XOP request, queueing it to all nodes in the cluster to 483 * execute the cluster op. 484 * 485 * XXX optimize single-target case. 486 */ 487 void 488 hammer2_xop_start_except(hammer2_xop_head_t *xop, hammer2_xop_desc_t *desc, 489 int notidx) 490 { 491 hammer2_inode_t *ip1; 492 hammer2_pfs_t *pmp; 493 hammer2_thread_t *thr; 494 int i; 495 int ng; 496 int nchains; 497 498 ip1 = xop->ip1; 499 pmp = ip1->pmp; 500 if (pmp->has_xop_threads == 0) 501 hammer2_xop_helper_create(pmp); 502 503 /* 504 * The sequencer assigns a worker thread to the XOP. 505 * 506 * (1) The worker threads are partitioned into two sets, one for 507 * NON-STRATEGY XOPs, and the other for STRATEGY XOPs. This 508 * guarantees that strategy calls will always be able to make 509 * progress and will not deadlock against non-strategy calls. 510 * 511 * (2) If clustered, non-strategy operations to the same inode must 512 * be serialized. This is to avoid confusion when issuing 513 * modifying operations because a XOP completes the instant a 514 * quorum is reached. 515 * 516 * TODO - RENAME fails here because it is potentially modifying 517 * three different inodes, but we triple-lock the inodes 518 * involved so it shouldn't create a sequencing schism. 519 */ 520 if (xop->flags & HAMMER2_XOP_STRATEGY) { 521 /* 522 * Use worker space 0 associated with the current cpu 523 * for strategy ops. 524 */ 525 /* 526 hammer2_xop_strategy_t *xopst; 527 u_int which; 528 529 xopst = &((hammer2_xop_t *)xop)->xop_strategy; 530 which = ((unsigned int)ip1->ihash + 531 ((unsigned int)xopst->lbase >> HAMMER2_PBUFRADIX)) % 532 hammer2_xop_sgroups; 533 ng = mycpu->gd_cpuid % hammer2_xop_mod + 534 hammer2_xop_mod * which; 535 */ 536 ng = 0; 537 } else if (hammer2_spread_workers == 0 && ip1->cluster.nchains == 1) { 538 /* 539 * For now try to keep the work on the same cpu to reduce 540 * IPI overhead. Several threads are assigned to each cpu, 541 * don't be very smart and select the one to use based on 542 * the inode hash. 543 */ 544 /* 545 u_int which; 546 547 which = (unsigned int)ip1->ihash % hammer2_xop_xgroups; 548 ng = mycpu->gd_cpuid % hammer2_xop_mod + 549 (which * hammer2_xop_mod) + 550 hammer2_xop_xbase; 551 */ 552 ng = 0; 553 } else { 554 /* 555 * Hash based on inode only, must serialize inode to same 556 * thread regardless of current cpu. 557 */ 558 /* 559 ng = (unsigned int)ip1->ihash % 560 (hammer2_xop_mod * hammer2_xop_xgroups) + 561 hammer2_xop_xbase; 562 */ 563 ng = 0; 564 } 565 xop->desc = desc; 566 567 /* 568 * The instant xop is queued another thread can pick it off. In the 569 * case of asynchronous ops, another thread might even finish and 570 * deallocate it. 571 */ 572 hammer2_spin_ex(&pmp->xop_spin); 573 nchains = ip1->cluster.nchains; 574 for (i = 0; i < nchains; ++i) { 575 /* 576 * XXX ip1->cluster.array* not stable here. This temporary 577 * hack fixes basic issues in target XOPs which need to 578 * obtain a starting chain from the inode but does not 579 * address possible races against inode updates which 580 * might NULL-out a chain. 581 */ 582 if (i != notidx && ip1->cluster.array[i].chain) { 583 thr = &pmp->xop_groups[ng].thrs[i]; 584 atomic_set_64(&xop->run_mask, 1LLU << i); 585 atomic_set_64(&xop->chk_mask, 1LLU << i); 586 xop->collect[i].thr = thr; 587 TAILQ_INSERT_TAIL(&thr->xopq, xop, collect[i].entry); 588 } 589 } 590 hammer2_spin_unex(&pmp->xop_spin); 591 /* xop can become invalid at this point */ 592 593 /* 594 * Each thread has its own xopq 595 */ 596 for (i = 0; i < nchains; ++i) { 597 if (i != notidx) { 598 thr = &pmp->xop_groups[ng].thrs[i]; 599 hammer2_thr_signal(thr, HAMMER2_THREAD_XOPQ); 600 hammer2_primary_xops_thread(thr); 601 } 602 } 603 } 604 605 void 606 hammer2_xop_start(hammer2_xop_head_t *xop, hammer2_xop_desc_t *desc) 607 { 608 hammer2_xop_start_except(xop, desc, -1); 609 } 610 611 /* 612 * Retire a XOP. Used by both the VOP frontend and by the XOP backend. 613 */ 614 void 615 hammer2_xop_retire(hammer2_xop_head_t *xop, uint64_t mask) 616 { 617 hammer2_chain_t *chain; 618 uint64_t nmask; 619 int i; 620 621 /* 622 * Remove the frontend collector or remove a backend feeder. 623 * 624 * When removing the frontend we must wakeup any backend feeders 625 * who are waiting for FIFO space. 626 * 627 * When removing the last backend feeder we must wakeup any waiting 628 * frontend. 629 */ 630 KKASSERT(xop->run_mask & mask); 631 nmask = atomic_fetchadd_64(&xop->run_mask, 632 -mask + HAMMER2_XOPMASK_FEED); 633 634 /* 635 * More than one entity left 636 */ 637 if ((nmask & HAMMER2_XOPMASK_ALLDONE) != mask) { 638 /* 639 * Frontend terminating, wakeup any backends waiting on 640 * fifo full. 641 * 642 * NOTE!!! The xop can get ripped out from under us at 643 * this point, so do not reference it again. 644 * The wakeup(xop) doesn't touch the xop and 645 * is ok. 646 */ 647 if (mask == HAMMER2_XOPMASK_VOP) { 648 if (nmask & HAMMER2_XOPMASK_FIFOW) 649 wakeup(xop); 650 } 651 652 /* 653 * Wakeup frontend if the last backend is terminating. 654 */ 655 nmask -= mask; 656 if ((nmask & HAMMER2_XOPMASK_ALLDONE) == HAMMER2_XOPMASK_VOP) { 657 if (nmask & HAMMER2_XOPMASK_WAIT) 658 wakeup(xop); 659 } 660 661 return; 662 } 663 /* else nobody else left, we can ignore FIFOW */ 664 665 /* 666 * All collectors are gone, we can cleanup and dispose of the XOP. 667 * Note that this can wind up being a frontend OR a backend. 668 * Pending chains are locked shared and not owned by any thread. 669 * 670 * Cleanup the collection cluster. 671 */ 672 for (i = 0; i < xop->cluster.nchains; ++i) { 673 xop->cluster.array[i].flags = 0; 674 chain = xop->cluster.array[i].chain; 675 if (chain) { 676 xop->cluster.array[i].chain = NULL; 677 hammer2_chain_drop_unhold(chain); 678 } 679 } 680 681 /* 682 * Cleanup the fifos. Since we are the only entity left on this 683 * xop we don't have to worry about fifo flow control, and one 684 * lfence() will do the job. 685 */ 686 cpu_lfence(); 687 mask = xop->chk_mask; 688 for (i = 0; mask && i < HAMMER2_MAXCLUSTER; ++i) { 689 hammer2_xop_fifo_t *fifo = &xop->collect[i]; 690 while (fifo->ri != fifo->wi) { 691 chain = fifo->array[fifo->ri & HAMMER2_XOPFIFO_MASK]; 692 if (chain) 693 hammer2_chain_drop_unhold(chain); 694 ++fifo->ri; 695 } 696 mask &= ~(1U << i); 697 } 698 699 /* 700 * The inode is only held at this point, simply drop it. 701 */ 702 if (xop->ip1) { 703 hammer2_inode_drop(xop->ip1); 704 xop->ip1 = NULL; 705 } 706 if (xop->ip2) { 707 hammer2_inode_drop(xop->ip2); 708 xop->ip2 = NULL; 709 } 710 if (xop->ip3) { 711 hammer2_inode_drop(xop->ip3); 712 xop->ip3 = NULL; 713 } 714 if (xop->ip4) { 715 hammer2_inode_drop(xop->ip4); 716 xop->ip4 = NULL; 717 } 718 if (xop->name1) { 719 kfree(xop->name1, M_HAMMER2); 720 xop->name1 = NULL; 721 xop->name1_len = 0; 722 } 723 if (xop->name2) { 724 kfree(xop->name2, M_HAMMER2); 725 xop->name2 = NULL; 726 xop->name2_len = 0; 727 } 728 729 free(xop); 730 } 731 732 /* 733 * (Backend) Returns non-zero if the frontend is still attached. 734 */ 735 int 736 hammer2_xop_active(hammer2_xop_head_t *xop) 737 { 738 if (xop->run_mask & HAMMER2_XOPMASK_VOP) 739 return 1; 740 else 741 return 0; 742 } 743 744 /* 745 * (Backend) Feed chain data through the cluster validator and back to 746 * the frontend. Chains are fed from multiple nodes concurrently 747 * and pipelined via per-node FIFOs in the XOP. 748 * 749 * The chain must be locked (either shared or exclusive). The caller may 750 * unlock and drop the chain on return. This function will add an extra 751 * ref and hold the chain's data for the pass-back. 752 * 753 * No xop lock is needed because we are only manipulating fields under 754 * our direct control. 755 * 756 * Returns 0 on success and a hammer2 error code if sync is permanently 757 * lost. The caller retains a ref on the chain but by convention 758 * the lock is typically inherited by the xop (caller loses lock). 759 * 760 * Returns non-zero on error. In this situation the caller retains a 761 * ref on the chain but loses the lock (we unlock here). 762 */ 763 int 764 hammer2_xop_feed(hammer2_xop_head_t *xop, hammer2_chain_t *chain, 765 int clindex, int error) 766 { 767 hammer2_xop_fifo_t *fifo; 768 uint64_t mask; 769 770 /* 771 * Early termination (typicaly of xop_readir) 772 */ 773 if (hammer2_xop_active(xop) == 0) { 774 error = HAMMER2_ERROR_ABORTED; 775 goto done; 776 } 777 778 /* 779 * Multi-threaded entry into the XOP collector. We own the 780 * fifo->wi for our clindex. 781 */ 782 fifo = &xop->collect[clindex]; 783 784 if (fifo->ri == fifo->wi - HAMMER2_XOPFIFO) 785 lwkt_yield(); 786 while (fifo->ri == fifo->wi - HAMMER2_XOPFIFO) { 787 atomic_set_int(&fifo->flags, HAMMER2_XOP_FIFO_STALL); 788 mask = xop->run_mask; 789 if ((mask & HAMMER2_XOPMASK_VOP) == 0) { 790 error = HAMMER2_ERROR_ABORTED; 791 goto done; 792 } 793 tsleep_interlock(xop, 0); 794 if (atomic_cmpset_64(&xop->run_mask, mask, 795 mask | HAMMER2_XOPMASK_FIFOW)) { 796 if (fifo->ri == fifo->wi - HAMMER2_XOPFIFO) { 797 tsleep(xop, PINTERLOCKED, "h2feed", hz*60); 798 } 799 } 800 /* retry */ 801 } 802 atomic_clear_int(&fifo->flags, HAMMER2_XOP_FIFO_STALL); 803 if (chain) 804 hammer2_chain_ref_hold(chain); 805 if (error == 0 && chain) 806 error = chain->error; 807 fifo->errors[fifo->wi & HAMMER2_XOPFIFO_MASK] = error; 808 fifo->array[fifo->wi & HAMMER2_XOPFIFO_MASK] = chain; 809 cpu_sfence(); 810 ++fifo->wi; 811 812 mask = atomic_fetchadd_64(&xop->run_mask, HAMMER2_XOPMASK_FEED); 813 if (mask & HAMMER2_XOPMASK_WAIT) { 814 atomic_clear_64(&xop->run_mask, HAMMER2_XOPMASK_WAIT); 815 wakeup(xop); 816 } 817 error = 0; 818 819 /* 820 * Cleanup. If an error occurred we eat the lock. If no error 821 * occurred the fifo inherits the lock and gains an additional ref. 822 * 823 * The caller's ref remains in both cases. 824 */ 825 done: 826 return error; 827 } 828 829 /* 830 * (Frontend) collect a response from a running cluster op. 831 * 832 * Responses are fed from all appropriate nodes concurrently 833 * and collected into a cohesive response >= collect_key. 834 * 835 * The collector will return the instant quorum or other requirements 836 * are met, even if some nodes get behind or become non-responsive. 837 * 838 * HAMMER2_XOP_COLLECT_NOWAIT - Used to 'poll' a completed collection, 839 * usually called synchronously from the 840 * node XOPs for the strategy code to 841 * fake the frontend collection and complete 842 * the BIO as soon as possible. 843 * 844 * HAMMER2_XOP_SYNCHRONIZER - Reqeuest synchronization with a particular 845 * cluster index, prevents looping when that 846 * index is out of sync so caller can act on 847 * the out of sync element. ESRCH and EDEADLK 848 * can be returned if this flag is specified. 849 * 850 * Returns 0 on success plus a filled out xop->cluster structure. 851 * Return ENOENT on normal termination. 852 * Otherwise return an error. 853 * 854 * WARNING! If the xop returns a cluster with a non-NULL focus, note that 855 * none of the chains in the cluster (or the focus) are either 856 * locked or I/O synchronized with the cpu. hammer2_xop_gdata() 857 * and hammer2_xop_pdata() must be used to safely access the focus 858 * chain's content. 859 * 860 * The frontend can make certain assumptions based on higher-level 861 * locking done by the frontend, but data integrity absolutely 862 * requires using the gdata/pdata API. 863 */ 864 int 865 hammer2_xop_collect(hammer2_xop_head_t *xop, int flags) 866 { 867 hammer2_xop_fifo_t *fifo; 868 hammer2_chain_t *chain; 869 hammer2_key_t lokey; 870 uint64_t mask; 871 int error; 872 int keynull; 873 int adv; /* advance the element */ 874 int i; 875 876 loop: 877 /* 878 * First loop tries to advance pieces of the cluster which 879 * are out of sync. 880 */ 881 lokey = HAMMER2_KEY_MAX; 882 keynull = HAMMER2_CHECK_NULL; 883 mask = xop->run_mask; 884 cpu_lfence(); 885 886 for (i = 0; i < xop->cluster.nchains; ++i) { 887 chain = xop->cluster.array[i].chain; 888 if (chain == NULL) { 889 adv = 1; 890 } else if (chain->bref.key < xop->collect_key) { 891 adv = 1; 892 } else { 893 keynull &= ~HAMMER2_CHECK_NULL; 894 if (lokey > chain->bref.key) 895 lokey = chain->bref.key; 896 adv = 0; 897 } 898 if (adv == 0) 899 continue; 900 901 /* 902 * Advance element if possible, advanced element may be NULL. 903 */ 904 if (chain) 905 hammer2_chain_drop_unhold(chain); 906 907 fifo = &xop->collect[i]; 908 if (fifo->ri != fifo->wi) { 909 cpu_lfence(); 910 chain = fifo->array[fifo->ri & HAMMER2_XOPFIFO_MASK]; 911 error = fifo->errors[fifo->ri & HAMMER2_XOPFIFO_MASK]; 912 ++fifo->ri; 913 xop->cluster.array[i].chain = chain; 914 xop->cluster.array[i].error = error; 915 if (chain == NULL) { 916 /* XXX */ 917 xop->cluster.array[i].flags |= 918 HAMMER2_CITEM_NULL; 919 } 920 if (fifo->wi - fifo->ri <= HAMMER2_XOPFIFO / 2) { 921 if (fifo->flags & HAMMER2_XOP_FIFO_STALL) { 922 atomic_clear_int(&fifo->flags, 923 HAMMER2_XOP_FIFO_STALL); 924 wakeup(xop); 925 lwkt_yield(); 926 } 927 } 928 --i; /* loop on same index */ 929 } else { 930 /* 931 * Retain CITEM_NULL flag. If set just repeat EOF. 932 * If not, the NULL,0 combination indicates an 933 * operation in-progress. 934 */ 935 xop->cluster.array[i].chain = NULL; 936 /* retain any CITEM_NULL setting */ 937 } 938 } 939 940 /* 941 * Determine whether the lowest collected key meets clustering 942 * requirements. Returns HAMMER2_ERROR_*: 943 * 944 * 0 - key valid, cluster can be returned. 945 * 946 * ENOENT - normal end of scan, return ENOENT. 947 * 948 * ESRCH - sufficient elements collected, quorum agreement 949 * that lokey is not a valid element and should be 950 * skipped. 951 * 952 * EDEADLK - sufficient elements collected, no quorum agreement 953 * (and no agreement possible). In this situation a 954 * repair is needed, for now we loop. 955 * 956 * EINPROGRESS - insufficient elements collected to resolve, wait 957 * for event and loop. 958 * 959 * EIO/ECHECK - IO error or CRC check error from hammer2_cluster_check() 960 */ 961 if ((flags & HAMMER2_XOP_COLLECT_WAITALL) && 962 (mask & HAMMER2_XOPMASK_ALLDONE) != HAMMER2_XOPMASK_VOP) { 963 error = HAMMER2_ERROR_EINPROGRESS; 964 } else { 965 error = hammer2_cluster_check(&xop->cluster, lokey, keynull); 966 } 967 if (error == HAMMER2_ERROR_EINPROGRESS) { 968 if (flags & HAMMER2_XOP_COLLECT_NOWAIT) 969 goto done; 970 tsleep_interlock(xop, 0); 971 if (atomic_cmpset_64(&xop->run_mask, 972 mask, mask | HAMMER2_XOPMASK_WAIT)) { 973 tsleep(xop, PINTERLOCKED, "h2coll", hz*60); 974 } 975 goto loop; 976 } 977 if (error == HAMMER2_ERROR_ESRCH) { 978 if (lokey != HAMMER2_KEY_MAX) { 979 xop->collect_key = lokey + 1; 980 goto loop; 981 } 982 error = HAMMER2_ERROR_ENOENT; 983 } 984 if (error == HAMMER2_ERROR_EDEADLK) { 985 kprintf("hammer2: no quorum possible lokey %016jx\n", 986 lokey); 987 if (lokey != HAMMER2_KEY_MAX) { 988 xop->collect_key = lokey + 1; 989 goto loop; 990 } 991 error = HAMMER2_ERROR_ENOENT; 992 } 993 if (lokey == HAMMER2_KEY_MAX) 994 xop->collect_key = lokey; 995 else 996 xop->collect_key = lokey + 1; 997 done: 998 return error; 999 } 1000 1001 /* 1002 * N x M processing threads are available to handle XOPs, N per cluster 1003 * index x M cluster nodes. 1004 * 1005 * Locate and return the next runnable xop, or NULL if no xops are 1006 * present or none of the xops are currently runnable (for various reasons). 1007 * The xop is left on the queue and serves to block other dependent xops 1008 * from being run. 1009 * 1010 * Dependent xops will not be returned. 1011 * 1012 * Sets HAMMER2_XOP_FIFO_RUN on the returned xop or returns NULL. 1013 * 1014 * NOTE! Xops run concurrently for each cluster index. 1015 */ 1016 #define XOP_HASH_SIZE 16 1017 #define XOP_HASH_MASK (XOP_HASH_SIZE - 1) 1018 1019 static __inline 1020 int 1021 xop_testhash(hammer2_thread_t *thr, hammer2_inode_t *ip, uint32_t *hash) 1022 { 1023 uint32_t mask; 1024 int hv; 1025 1026 hv = (int)((uintptr_t)ip + (uintptr_t)thr) / sizeof(hammer2_inode_t); 1027 mask = 1U << (hv & 31); 1028 hv >>= 5; 1029 1030 return ((int)(hash[hv & XOP_HASH_MASK] & mask)); 1031 } 1032 1033 static __inline 1034 void 1035 xop_sethash(hammer2_thread_t *thr, hammer2_inode_t *ip, uint32_t *hash) 1036 { 1037 uint32_t mask; 1038 int hv; 1039 1040 hv = (int)((uintptr_t)ip + (uintptr_t)thr) / sizeof(hammer2_inode_t); 1041 mask = 1U << (hv & 31); 1042 hv >>= 5; 1043 1044 hash[hv & XOP_HASH_MASK] |= mask; 1045 } 1046 1047 static 1048 hammer2_xop_head_t * 1049 hammer2_xop_next(hammer2_thread_t *thr) 1050 { 1051 hammer2_pfs_t *pmp = thr->pmp; 1052 int clindex = thr->clindex; 1053 uint32_t hash[XOP_HASH_SIZE] = { 0 }; 1054 hammer2_xop_head_t *xop; 1055 1056 hammer2_spin_ex(&pmp->xop_spin); 1057 TAILQ_FOREACH(xop, &thr->xopq, collect[clindex].entry) { 1058 /* 1059 * Check dependency 1060 */ 1061 if (xop_testhash(thr, xop->ip1, hash) || 1062 (xop->ip2 && xop_testhash(thr, xop->ip2, hash)) || 1063 (xop->ip3 && xop_testhash(thr, xop->ip3, hash)) || 1064 (xop->ip4 && xop_testhash(thr, xop->ip4, hash))) 1065 { 1066 continue; 1067 } 1068 xop_sethash(thr, xop->ip1, hash); 1069 if (xop->ip2) 1070 xop_sethash(thr, xop->ip2, hash); 1071 if (xop->ip3) 1072 xop_sethash(thr, xop->ip3, hash); 1073 if (xop->ip4) 1074 xop_sethash(thr, xop->ip4, hash); 1075 1076 /* 1077 * Check already running 1078 */ 1079 if (xop->collect[clindex].flags & HAMMER2_XOP_FIFO_RUN) 1080 continue; 1081 1082 /* 1083 * Found a good one, return it. 1084 */ 1085 atomic_set_int(&xop->collect[clindex].flags, 1086 HAMMER2_XOP_FIFO_RUN); 1087 break; 1088 } 1089 hammer2_spin_unex(&pmp->xop_spin); 1090 1091 return xop; 1092 } 1093 1094 /* 1095 * Remove the completed XOP from the queue, clear HAMMER2_XOP_FIFO_RUN. 1096 * 1097 * NOTE! Xops run concurrently for each cluster index. 1098 */ 1099 static 1100 void 1101 hammer2_xop_dequeue(hammer2_thread_t *thr, hammer2_xop_head_t *xop) 1102 { 1103 hammer2_pfs_t *pmp = thr->pmp; 1104 int clindex = thr->clindex; 1105 1106 hammer2_spin_ex(&pmp->xop_spin); 1107 TAILQ_REMOVE(&thr->xopq, xop, collect[clindex].entry); 1108 atomic_clear_int(&xop->collect[clindex].flags, 1109 HAMMER2_XOP_FIFO_RUN); 1110 hammer2_spin_unex(&pmp->xop_spin); 1111 if (TAILQ_FIRST(&thr->xopq)) 1112 hammer2_thr_signal(thr, HAMMER2_THREAD_XOPQ); 1113 } 1114 1115 /* 1116 * Primary management thread for xops support. Each node has several such 1117 * threads which replicate front-end operations on cluster nodes. 1118 * 1119 * XOPS thread node operations, allowing the function to focus on a single 1120 * node in the cluster after validating the operation with the cluster. 1121 * This is primarily what prevents dead or stalled nodes from stalling 1122 * the front-end. 1123 */ 1124 void 1125 hammer2_primary_xops_thread(void *arg) 1126 { 1127 hammer2_thread_t *thr = arg; 1128 hammer2_pfs_t *pmp; 1129 hammer2_xop_head_t *xop; 1130 uint64_t mask; 1131 uint32_t flags; 1132 uint32_t nflags; 1133 hammer2_xop_desc_t *last_desc = NULL; 1134 1135 pmp = thr->pmp; 1136 /*xgrp = &pmp->xop_groups[thr->repidx]; not needed */ 1137 mask = 1LLU << thr->clindex; 1138 1139 for (;;) { 1140 flags = thr->flags; 1141 1142 /* 1143 * Handle stop request 1144 */ 1145 if (flags & HAMMER2_THREAD_STOP) 1146 break; 1147 1148 /* 1149 * Handle freeze request 1150 */ 1151 if (flags & HAMMER2_THREAD_FREEZE) { 1152 hammer2_thr_signal2(thr, HAMMER2_THREAD_FROZEN, 1153 HAMMER2_THREAD_FREEZE); 1154 continue; 1155 } 1156 1157 if (flags & HAMMER2_THREAD_UNFREEZE) { 1158 hammer2_thr_signal2(thr, 0, 1159 HAMMER2_THREAD_FROZEN | 1160 HAMMER2_THREAD_UNFREEZE); 1161 continue; 1162 } 1163 1164 /* 1165 * Force idle if frozen until unfrozen or stopped. 1166 */ 1167 if (flags & HAMMER2_THREAD_FROZEN) { 1168 hammer2_thr_wait_any(thr, 1169 HAMMER2_THREAD_UNFREEZE | 1170 HAMMER2_THREAD_STOP, 1171 0); 1172 continue; 1173 } 1174 1175 /* 1176 * Reset state on REMASTER request 1177 */ 1178 if (flags & HAMMER2_THREAD_REMASTER) { 1179 hammer2_thr_signal2(thr, 0, HAMMER2_THREAD_REMASTER); 1180 /* reset state here */ 1181 continue; 1182 } 1183 1184 /* 1185 * Process requests. Each request can be multi-queued. 1186 * 1187 * If we get behind and the frontend VOP is no longer active, 1188 * we retire the request without processing it. The callback 1189 * may also abort processing if the frontend VOP becomes 1190 * inactive. 1191 */ 1192 if (flags & HAMMER2_THREAD_XOPQ) { 1193 nflags = flags & ~HAMMER2_THREAD_XOPQ; 1194 if (!atomic_cmpset_int(&thr->flags, flags, nflags)) 1195 continue; 1196 flags = nflags; 1197 /* fall through */ 1198 } 1199 while ((xop = hammer2_xop_next(thr)) != NULL) { 1200 if (hammer2_xop_active(xop)) { 1201 last_desc = xop->desc; 1202 xop->desc->storage_func((hammer2_xop_t *)xop, 1203 thr->scratch, 1204 thr->clindex); 1205 hammer2_xop_dequeue(thr, xop); 1206 hammer2_xop_retire(xop, mask); 1207 } else { 1208 last_desc = xop->desc; 1209 hammer2_xop_feed(xop, NULL, thr->clindex, 1210 ECONNABORTED); 1211 hammer2_xop_dequeue(thr, xop); 1212 hammer2_xop_retire(xop, mask); 1213 } 1214 } 1215 1216 /* Don't wait, this is a XOP caller thread in makefs */ 1217 break; 1218 1219 /* 1220 * Wait for event, interlock using THREAD_WAITING and 1221 * THREAD_SIGNAL. 1222 * 1223 * For robustness poll on a 30-second interval, but nominally 1224 * expect to be woken up. 1225 */ 1226 nflags = flags | HAMMER2_THREAD_WAITING; 1227 1228 tsleep_interlock(&thr->flags, 0); 1229 if (atomic_cmpset_int(&thr->flags, flags, nflags)) { 1230 tsleep(&thr->flags, PINTERLOCKED, "h2idle", hz*30); 1231 } 1232 } 1233 1234 #if 0 1235 /* 1236 * Cleanup / termination 1237 */ 1238 while ((xop = TAILQ_FIRST(&thr->xopq)) != NULL) { 1239 kprintf("hammer2_thread: aborting xop %s\n", xop->desc->id); 1240 TAILQ_REMOVE(&thr->xopq, xop, 1241 collect[thr->clindex].entry); 1242 hammer2_xop_retire(xop, mask); 1243 } 1244 #endif 1245 thr->td = NULL; 1246 hammer2_thr_signal(thr, HAMMER2_THREAD_STOPPED); 1247 /* thr structure can go invalid after this point */ 1248 } 1249