g_bde_work.c revision 114088
1/*- 2 * Copyright (c) 2002 Poul-Henning Kamp 3 * Copyright (c) 2002 Networks Associates Technology, Inc. 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp 7 * and NAI Labs, the Security Research Division of Network Associates, Inc. 8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 9 * DARPA CHATS research program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 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 the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: head/sys/geom/bde/g_bde_work.c 114088 2003-04-26 21:40:26Z phk $ 33 * 34 * This source file contains the state-engine which makes things happen in the 35 * right order. 36 * 37 * Outline: 38 * 1) g_bde_start1() 39 * Break the struct bio into multiple work packets one per zone. 40 * 2) g_bde_start2() 41 * Setup the necessary sector buffers and start those read operations 42 * which we can start at this time and put the item on the work-list. 43 * 3) g_bde_worker() 44 * Scan the work-list for items which are ready for crypto processing 45 * and call the matching crypto function in g_bde_crypt.c and schedule 46 * any writes needed. Read operations finish here by releasing the 47 * sector buffers and delivering the original bio request. 48 * 4) g_bde_write_done() 49 * Release sector buffers and deliver the original bio request. 50 * 51 * Because of the C-scope rules, the functions are almost perfectly in the 52 * opposite order in this source file. 53 * 54 * XXX: A switch to the hardware assisted crypto in src/sys/opencrypto will add 55 * XXX: additional states to this state-engine. Since no hardware available 56 * XXX: at this time has AES support, implementing this has been postponed 57 * XXX: until such time as it would result in a benefit. 58 */ 59 60#include <sys/param.h> 61#include <sys/bio.h> 62#include <sys/lock.h> 63#include <sys/mutex.h> 64#include <sys/queue.h> 65#include <sys/malloc.h> 66#include <sys/systm.h> 67#include <sys/kernel.h> 68#include <sys/sysctl.h> 69#include <sys/proc.h> 70#include <sys/kthread.h> 71 72#include <crypto/rijndael/rijndael.h> 73#include <crypto/sha2/sha2.h> 74#include <geom/geom.h> 75#include <geom/bde/g_bde.h> 76 77static void g_bde_delete_sector(struct g_bde_softc *wp, struct g_bde_sector *sp); 78static struct g_bde_sector * g_bde_new_sector(struct g_bde_work *wp, u_int len); 79static void g_bde_release_keysector(struct g_bde_work *wp); 80static struct g_bde_sector *g_bde_get_sector(struct g_bde_work *wp, off_t offset); 81static int g_bde_start_read(struct g_bde_sector *sp); 82static void g_bde_purge_sector(struct g_bde_softc *sc, int fraction); 83 84/* 85 * Work item allocation. 86 * 87 * C++ would call these constructors and destructors. 88 */ 89static u_int g_bde_nwork; 90SYSCTL_UINT(_debug, OID_AUTO, gbde_nwork, CTLFLAG_RD, &g_bde_nwork, 0, ""); 91 92static struct g_bde_work * 93g_bde_new_work(struct g_bde_softc *sc) 94{ 95 struct g_bde_work *wp; 96 97 wp = g_malloc(sizeof *wp, M_NOWAIT | M_ZERO); 98 if (wp == NULL) 99 return (wp); 100 wp->state = SETUP; 101 wp->softc = sc; 102 g_bde_nwork++; 103 sc->nwork++; 104 TAILQ_INSERT_TAIL(&sc->worklist, wp, list); 105 return (wp); 106} 107 108static void 109g_bde_delete_work(struct g_bde_work *wp) 110{ 111 struct g_bde_softc *sc; 112 113 sc = wp->softc; 114 g_bde_nwork--; 115 sc->nwork--; 116 TAILQ_REMOVE(&sc->worklist, wp, list); 117 g_free(wp); 118} 119 120/* 121 * Sector buffer allocation 122 * 123 * These two functions allocate and free back variable sized sector buffers 124 */ 125 126static u_int g_bde_nsect; 127SYSCTL_UINT(_debug, OID_AUTO, gbde_nsect, CTLFLAG_RD, &g_bde_nsect, 0, ""); 128 129static void 130g_bde_delete_sector(struct g_bde_softc *sc, struct g_bde_sector *sp) 131{ 132 133 g_bde_nsect--; 134 sc->nsect--; 135 if (sp->malloc) 136 g_free(sp->data); 137 g_free(sp); 138} 139 140static struct g_bde_sector * 141g_bde_new_sector(struct g_bde_work *wp, u_int len) 142{ 143 struct g_bde_sector *sp; 144 145 sp = g_malloc(sizeof *sp, M_NOWAIT | M_ZERO); 146 if (sp == NULL) 147 return (sp); 148 if (len > 0) { 149 sp->data = g_malloc(len, M_NOWAIT | M_ZERO); 150 if (sp->data == NULL) { 151 g_free(sp); 152 return (NULL); 153 } 154 sp->malloc = 1; 155 } 156 g_bde_nsect++; 157 wp->softc->nsect++; 158 sp->size = len; 159 sp->softc = wp->softc; 160 sp->ref = 1; 161 sp->owner = wp; 162 sp->offset = wp->so; 163 sp->state = JUNK; 164 return (sp); 165} 166 167/* 168 * Skey sector cache. 169 * 170 * Nothing prevents two separate I/O requests from addressing the same zone 171 * and thereby needing the same skey sector. We therefore need to sequence 172 * I/O operations to the skey sectors. A certain amount of caching is also 173 * desirable, although the extent of benefit from this is not at this point 174 * determined. 175 * 176 * XXX: GEOM may be able to grow a generic caching facility at some point 177 * XXX: to support such needs. 178 */ 179 180static u_int g_bde_ncache; 181SYSCTL_UINT(_debug, OID_AUTO, gbde_ncache, CTLFLAG_RD, &g_bde_ncache, 0, ""); 182 183static void 184g_bde_purge_one_sector(struct g_bde_softc *sc, struct g_bde_sector *sp) 185{ 186 187 g_trace(G_T_TOPOLOGY, "g_bde_purge_one_sector(%p, %p)", sc, sp); 188 if (sp->ref != 0) 189 return; 190 TAILQ_REMOVE(&sc->freelist, sp, list); 191 g_bde_ncache--; 192 sc->ncache--; 193 bzero(sp->data, sp->size); 194 g_bde_delete_sector(sc, sp); 195} 196 197static struct g_bde_sector * 198g_bde_get_sector(struct g_bde_work *wp, off_t offset) 199{ 200 struct g_bde_sector *sp; 201 struct g_bde_softc *sc; 202 203 g_trace(G_T_TOPOLOGY, "g_bde_get_sector(%p, %jd)", wp, (intmax_t)offset); 204 sc = wp->softc; 205 206 if (malloc_last_fail() < g_bde_ncache) 207 g_bde_purge_sector(sc, -1); 208 209 sp = TAILQ_FIRST(&sc->freelist); 210 if (sp != NULL && sp->ref == 0 && sp->used + 300 < time_uptime) 211 g_bde_purge_one_sector(sc, sp); 212 213 TAILQ_FOREACH(sp, &sc->freelist, list) { 214 if (sp->offset == offset) 215 break; 216 } 217 if (sp != NULL) { 218 sp->ref++; 219 KASSERT(sp->offset == offset, ("wrong offset")); 220 KASSERT(sp->softc == wp->softc, ("wrong softc")); 221 if (sp->ref == 1) 222 sp->owner = wp; 223 } else { 224 if (malloc_last_fail() < g_bde_ncache) { 225 TAILQ_FOREACH(sp, &sc->freelist, list) 226 if (sp->ref == 0) 227 break; 228 } 229 if (sp == NULL && !TAILQ_EMPTY(&sc->freelist)) 230 sp = TAILQ_FIRST(&sc->freelist); 231 if (sp != NULL && sp->ref > 0) 232 sp = NULL; 233 if (sp == NULL) { 234 sp = g_bde_new_sector(wp, sc->sectorsize); 235 if (sp != NULL) { 236 g_bde_ncache++; 237 sc->ncache++; 238 TAILQ_INSERT_TAIL(&sc->freelist, sp, list); 239 sp->malloc = 2; 240 } 241 } 242 if (sp != NULL) { 243 sp->offset = offset; 244 sp->softc = wp->softc; 245 sp->ref = 1; 246 sp->owner = wp; 247 sp->state = JUNK; 248 sp->error = 0; 249 } 250 } 251 if (sp != NULL) { 252 TAILQ_REMOVE(&sc->freelist, sp, list); 253 TAILQ_INSERT_TAIL(&sc->freelist, sp, list); 254 sp->used = time_uptime; 255 } 256 wp->ksp = sp; 257 if (sp == NULL) { 258 g_bde_purge_sector(sc, -1); 259 } 260 return(sp); 261} 262 263static void 264g_bde_release_keysector(struct g_bde_work *wp) 265{ 266 struct g_bde_softc *sc; 267 struct g_bde_work *wp2; 268 struct g_bde_sector *sp; 269 270 sp = wp->ksp; 271 g_trace(G_T_TOPOLOGY, "g_bde_release_keysector(%p)", sp); 272 KASSERT(sp->malloc == 2, ("Wrong sector released")); 273 sc = sp->softc; 274 KASSERT(sc != NULL, ("NULL sp->softc")); 275 KASSERT(wp == sp->owner, ("Releasing, not owner")); 276 sp->owner = NULL; 277 wp->ksp = NULL; 278 sp->ref--; 279 if (sp->ref > 0) { 280 TAILQ_REMOVE(&sc->freelist, sp, list); 281 TAILQ_INSERT_TAIL(&sc->freelist, sp, list); 282 TAILQ_FOREACH(wp2, &sc->worklist, list) { 283 if (wp2->ksp == sp) { 284 KASSERT(wp2 != wp, ("Self-reowning")); 285 sp->owner = wp2; 286 wakeup(sp->softc); 287 break; 288 } 289 } 290 KASSERT(wp2 != NULL, ("Failed to pick up owner for %p\n", sp)); 291 } else if (sp->error != 0) { 292 sp->offset = ~0; 293 sp->error = 0; 294 sp->state = JUNK; 295 } 296 TAILQ_REMOVE(&sc->freelist, sp, list); 297 TAILQ_INSERT_HEAD(&sc->freelist, sp, list); 298} 299 300static void 301g_bde_purge_sector(struct g_bde_softc *sc, int fraction) 302{ 303 struct g_bde_sector *sp; 304 int n; 305 306 g_trace(G_T_TOPOLOGY, "g_bde_purge_sector(%p)", sc); 307 if (fraction > 0) 308 n = sc->ncache / fraction + 1; 309 else 310 n = g_bde_ncache - malloc_last_fail(); 311 if (n < 0) 312 return; 313 if (n > sc->ncache) 314 n = sc->ncache; 315 while(n--) { 316 TAILQ_FOREACH(sp, &sc->freelist, list) { 317 if (sp->ref != 0) 318 continue; 319 TAILQ_REMOVE(&sc->freelist, sp, list); 320 g_bde_ncache--; 321 sc->ncache--; 322 bzero(sp->data, sp->size); 323 g_bde_delete_sector(sc, sp); 324 break; 325 } 326 } 327} 328 329static struct g_bde_sector * 330g_bde_read_keysector(struct g_bde_softc *sc, struct g_bde_work *wp) 331{ 332 struct g_bde_sector *sp; 333 334 g_trace(G_T_TOPOLOGY, "g_bde_read_keysector(%p)", wp); 335 sp = g_bde_get_sector(wp, wp->kso); 336 if (sp == NULL) 337 sp = g_bde_get_sector(wp, wp->kso); 338 if (sp == NULL) 339 return (sp); 340 if (sp->owner != wp) 341 return (sp); 342 if (sp->state == VALID) 343 return (sp); 344 if (g_bde_start_read(sp) == 0) 345 return (sp); 346 g_bde_release_keysector(wp); 347 return (NULL); 348} 349 350/* 351 * Contribute to the completion of the original bio request. 352 * 353 * We have no simple way to tell how many bits the original bio request has 354 * been segmented into, so the easiest way to determine when we can deliver 355 * it is to keep track of the number of bytes we have completed. We keep 356 * track of any errors underway and latch onto the first one. 357 * 358 * We always report "nothing done" in case of error, because random bits here 359 * and there may be completed and returning a number of completed bytes does 360 * not convey any useful information about which bytes they were. If some 361 * piece of broken code somewhere interprets this to mean that nothing has 362 * changed on the underlying media they deserve the lossage headed for them. 363 * 364 * A single mutex per g_bde instance is used to prevent contention. 365 */ 366 367static void 368g_bde_contribute(struct bio *bp, off_t bytes, int error) 369{ 370 struct g_bde_softc *sc; 371 372 g_trace(G_T_TOPOLOGY, "g_bde_contribute bp %p bytes %jd error %d", 373 bp, (intmax_t)bytes, error); 374 sc = bp->bio_driver1; 375 if (bp->bio_error == 0) 376 bp->bio_error = error; 377 bp->bio_completed += bytes; 378 KASSERT(bp->bio_completed <= bp->bio_length, ("Too large contribution")); 379 if (bp->bio_completed == bp->bio_length) { 380 if (bp->bio_error != 0) 381 bp->bio_completed = 0; 382 g_io_deliver(bp, bp->bio_error); 383 } 384} 385 386/* 387 * A write operation has finished. When we have all expected cows in the 388 * barn close the door and call it a day. 389 */ 390 391static void 392g_bde_write_done(struct bio *bp) 393{ 394 struct g_bde_sector *sp; 395 struct g_bde_work *wp; 396 struct g_bde_softc *sc; 397 398 sp = bp->bio_caller1; 399 sc = bp->bio_caller2; 400 mtx_lock(&sc->worklist_mutex); 401 KASSERT(sp != NULL, ("NULL sp")); 402 KASSERT(sc != NULL, ("NULL sc")); 403 KASSERT(sp->owner != NULL, ("NULL sp->owner")); 404 g_trace(G_T_TOPOLOGY, "g_bde_write_done(%p)", sp); 405 sp->error = bp->bio_error; 406 g_destroy_bio(bp); 407 wp = sp->owner; 408 if (wp->error == 0) 409 wp->error = sp->error; 410 411 if (wp->bp->bio_cmd == BIO_DELETE) { 412 KASSERT(sp == wp->sp, ("trashed delete op")); 413 g_bde_contribute(wp->bp, wp->length, wp->error); 414 g_bde_delete_sector(sc, sp); 415 g_bde_delete_work(wp); 416 mtx_unlock(&sc->worklist_mutex); 417 return; 418 } 419 420 KASSERT(wp->bp->bio_cmd == BIO_WRITE, ("Confused in g_bde_write_done()")); 421 KASSERT(sp == wp->sp || sp == wp->ksp, ("trashed write op")); 422 if (wp->sp == sp) { 423 g_bde_delete_sector(sc, wp->sp); 424 wp->sp = NULL; 425 } else { 426 sp->state = VALID; 427 } 428 if (wp->sp == NULL && wp->ksp != NULL && wp->ksp->state == VALID) { 429 g_bde_contribute(wp->bp, wp->length, wp->error); 430 g_bde_release_keysector(wp); 431 g_bde_delete_work(wp); 432 } 433 mtx_unlock(&sc->worklist_mutex); 434 return; 435} 436 437/* 438 * Send a write request for the given sector down the pipeline. 439 */ 440 441static int 442g_bde_start_write(struct g_bde_sector *sp) 443{ 444 struct bio *bp; 445 struct g_bde_softc *sc; 446 447 g_trace(G_T_TOPOLOGY, "g_bde_start_write(%p)", sp); 448 sc = sp->softc; 449 KASSERT(sc != NULL, ("NULL sc in g_bde_start_write")); 450 KASSERT(sp->owner != NULL, ("NULL sp->owner in g_bde_start_write")); 451 bp = g_new_bio(); 452 if (bp == NULL) 453 return (ENOMEM); 454 bp->bio_cmd = BIO_WRITE; 455 bp->bio_offset = sp->offset; 456 bp->bio_data = sp->data; 457 bp->bio_length = sp->size; 458 bp->bio_done = g_bde_write_done; 459 bp->bio_caller1 = sp; 460 bp->bio_caller2 = sc; 461 sp->state = IO; 462 g_io_request(bp, sc->consumer); 463 return(0); 464} 465 466/* 467 * A read operation has finished. Mark the sector no longer iobusy and 468 * wake up the worker thread and let it do its thing. 469 */ 470 471static void 472g_bde_read_done(struct bio *bp) 473{ 474 struct g_bde_sector *sp; 475 struct g_bde_softc *sc; 476 477 sp = bp->bio_caller1; 478 g_trace(G_T_TOPOLOGY, "g_bde_read_done(%p)", sp); 479 sc = bp->bio_caller2; 480 mtx_lock(&sc->worklist_mutex); 481 sp->error = bp->bio_error; 482 sp->state = VALID; 483 wakeup(sc); 484 g_destroy_bio(bp); 485 mtx_unlock(&sc->worklist_mutex); 486} 487 488/* 489 * Send a read request for the given sector down the pipeline. 490 */ 491 492static int 493g_bde_start_read(struct g_bde_sector *sp) 494{ 495 struct bio *bp; 496 struct g_bde_softc *sc; 497 498 g_trace(G_T_TOPOLOGY, "g_bde_start_read(%p)", sp); 499 sc = sp->softc; 500 KASSERT(sc != NULL, ("Null softc in sp %p", sp)); 501 bp = g_new_bio(); 502 if (bp == NULL) 503 return (ENOMEM); 504 bp->bio_cmd = BIO_READ; 505 bp->bio_offset = sp->offset; 506 bp->bio_data = sp->data; 507 bp->bio_length = sp->size; 508 bp->bio_done = g_bde_read_done; 509 bp->bio_caller1 = sp; 510 bp->bio_caller2 = sc; 511 sp->state = IO; 512 g_io_request(bp, sc->consumer); 513 return(0); 514} 515 516/* 517 * The worker thread. 518 * 519 * The up/down path of GEOM is not allowed to sleep or do any major work 520 * so we use this thread to do the actual crypto operations and to push 521 * the state engine onwards. 522 * 523 * XXX: if we switch to the src/sys/opencrypt hardware assisted encryption 524 * XXX: using a thread here is probably not needed. 525 */ 526 527void 528g_bde_worker(void *arg) 529{ 530 struct g_bde_softc *sc; 531 struct g_bde_work *wp; 532 struct g_geom *gp; 533 int busy, error; 534 535 gp = arg; 536 sc = gp->softc; 537 538 mtx_lock(&sc->worklist_mutex); 539 for (;;) { 540 busy = 0; 541 g_trace(G_T_TOPOLOGY, "g_bde_worker scan"); 542 TAILQ_FOREACH(wp, &sc->worklist, list) { 543 KASSERT(wp != NULL, ("NULL wp")); 544 KASSERT(wp->softc != NULL, ("NULL wp->softc")); 545 if (wp->state != WAIT) 546 continue; /* Not interesting here */ 547 548 KASSERT(wp->bp != NULL, ("NULL wp->bp")); 549 KASSERT(wp->sp != NULL, ("NULL wp->sp")); 550 551 if (wp->ksp != NULL) { 552 if (wp->ksp->owner != wp) 553 continue; 554 if (wp->ksp->state == IO) 555 continue; 556 KASSERT(wp->ksp->state == VALID, 557 ("Illegal sector state (JUNK ?)")); 558 } 559 560 if (wp->bp->bio_cmd == BIO_READ && wp->sp->state != VALID) 561 continue; 562 563 if (wp->ksp != NULL && wp->ksp->error != 0) { 564 g_bde_contribute(wp->bp, wp->length, 565 wp->ksp->error); 566 g_bde_delete_sector(sc, wp->sp); 567 g_bde_release_keysector(wp); 568 g_bde_delete_work(wp); 569 busy++; 570 break; 571 } 572 switch(wp->bp->bio_cmd) { 573 case BIO_READ: 574 if (wp->ksp == NULL) { 575 KASSERT(wp->error != 0, 576 ("BIO_READ, no ksp and no error")); 577 g_bde_contribute(wp->bp, wp->length, 578 wp->error); 579 } else { 580 if (wp->sp->error == 0) { 581 mtx_unlock(&sc->worklist_mutex); 582 g_bde_crypt_read(wp); 583 mtx_lock(&sc->worklist_mutex); 584 } 585 g_bde_contribute(wp->bp, wp->length, 586 wp->sp->error); 587 } 588 g_bde_delete_sector(sc, wp->sp); 589 if (wp->ksp != NULL) 590 g_bde_release_keysector(wp); 591 g_bde_delete_work(wp); 592 break; 593 case BIO_WRITE: 594 wp->state = FINISH; 595 KASSERT(wp->sp->owner == wp, ("Write not owner sp")); 596 KASSERT(wp->ksp->owner == wp, ("Write not owner ksp")); 597 mtx_unlock(&sc->worklist_mutex); 598 g_bde_crypt_write(wp); 599 mtx_lock(&sc->worklist_mutex); 600 error = g_bde_start_write(wp->sp); 601 if (error) { 602 g_bde_contribute(wp->bp, wp->length, error); 603 g_bde_release_keysector(wp); 604 g_bde_delete_sector(sc, wp->sp); 605 g_bde_delete_work(wp); 606 break; 607 } 608 error = g_bde_start_write(wp->ksp); 609 if (wp->error == 0) 610 wp->error = error; 611 break; 612 case BIO_DELETE: 613 wp->state = FINISH; 614 mtx_unlock(&sc->worklist_mutex); 615 g_bde_crypt_delete(wp); 616 mtx_lock(&sc->worklist_mutex); 617 g_bde_start_write(wp->sp); 618 break; 619 } 620 busy++; 621 break; 622 } 623 if (!busy) { 624 /* 625 * We don't look for our death-warrant until we are 626 * idle. Shouldn't make a difference in practice. 627 */ 628 if (sc->dead) 629 break; 630 g_trace(G_T_TOPOLOGY, "g_bde_worker sleep"); 631 error = msleep(sc, &sc->worklist_mutex, 632 PRIBIO, "g_bde", hz); 633 if (error == EWOULDBLOCK) { 634 /* 635 * Loose our skey cache in an orderly fashion. 636 * The exact rate can be tuned to be less 637 * aggressive if this is desirable. 10% per 638 * second means that the cache is gone in a 639 * few minutes. 640 */ 641 g_bde_purge_sector(sc, 10); 642 } 643 } 644 } 645 g_trace(G_T_TOPOLOGY, "g_bde_worker die"); 646 g_bde_purge_sector(sc, 1); 647 KASSERT(sc->nwork == 0, ("Dead but %d work remaining", sc->nwork)); 648 KASSERT(sc->ncache == 0, ("Dead but %d cache remaining", sc->ncache)); 649 KASSERT(sc->nsect == 0, ("Dead but %d sect remaining", sc->nsect)); 650 mtx_unlock(&sc->worklist_mutex); 651 sc->dead = 2; 652 wakeup(sc); 653 mtx_lock(&Giant); 654 kthread_exit(0); 655} 656 657/* 658 * g_bde_start1 has chopped the incoming request up so all the requests 659 * we see here are inside a single zone. Map the data and key locations 660 * grab the buffers we need and fire off the first volley of read requests. 661 */ 662 663static void 664g_bde_start2(struct g_bde_work *wp) 665{ 666 struct g_bde_softc *sc; 667 668 KASSERT(wp != NULL, ("NULL wp in g_bde_start2")); 669 KASSERT(wp->softc != NULL, ("NULL wp->softc")); 670 g_trace(G_T_TOPOLOGY, "g_bde_start2(%p)", wp); 671 sc = wp->softc; 672 if (wp->bp->bio_cmd == BIO_READ) { 673 wp->sp = g_bde_new_sector(wp, 0); 674 if (wp->sp == NULL) { 675 g_bde_contribute(wp->bp, wp->length, ENOMEM); 676 g_bde_delete_work(wp); 677 return; 678 } 679 wp->sp->size = wp->length; 680 wp->sp->data = wp->data; 681 if (g_bde_start_read(wp->sp) != 0) { 682 g_bde_contribute(wp->bp, wp->length, ENOMEM); 683 g_bde_delete_sector(sc, wp->sp); 684 g_bde_delete_work(wp); 685 return; 686 } 687 g_bde_read_keysector(sc, wp); 688 if (wp->ksp == NULL) 689 wp->error = ENOMEM; 690 } else if (wp->bp->bio_cmd == BIO_DELETE) { 691 wp->sp = g_bde_new_sector(wp, wp->length); 692 if (wp->sp == NULL) { 693 g_bde_contribute(wp->bp, wp->length, ENOMEM); 694 g_bde_delete_work(wp); 695 return; 696 } 697 } else if (wp->bp->bio_cmd == BIO_WRITE) { 698 wp->sp = g_bde_new_sector(wp, wp->length); 699 if (wp->sp == NULL) { 700 g_bde_contribute(wp->bp, wp->length, ENOMEM); 701 g_bde_delete_work(wp); 702 return; 703 } 704 g_bde_read_keysector(sc, wp); 705 if (wp->ksp == NULL) { 706 g_bde_contribute(wp->bp, wp->length, ENOMEM); 707 g_bde_delete_sector(sc, wp->sp); 708 g_bde_delete_work(wp); 709 return; 710 } 711 } else { 712 KASSERT(0 == 1, 713 ("Wrong bio_cmd %d in g_bde_start2", wp->bp->bio_cmd)); 714 } 715 716 wp->state = WAIT; 717 wakeup(sc); 718} 719 720/* 721 * Create a sequence of work structures, and have g_bde_map_sector() determine 722 * how long they each can be. Feed them to g_bde_start2(). 723 */ 724 725void 726g_bde_start1(struct bio *bp) 727{ 728 struct g_bde_softc *sc; 729 struct g_bde_work *wp; 730 off_t done; 731 732 sc = bp->bio_to->geom->softc; 733 bp->bio_driver1 = sc; 734 735 mtx_lock(&sc->worklist_mutex); 736 for(done = 0; done < bp->bio_length; ) { 737 wp = g_bde_new_work(sc); 738 if (wp != NULL) { 739 wp->bp = bp; 740 wp->offset = bp->bio_offset + done; 741 wp->data = bp->bio_data + done; 742 wp->length = bp->bio_length - done; 743 g_bde_map_sector(wp); 744 done += wp->length; 745 g_bde_start2(wp); 746 } 747 if (wp == NULL || bp->bio_error != 0) { 748 g_bde_contribute(bp, bp->bio_length - done, ENOMEM); 749 break; 750 } 751 } 752 mtx_unlock(&sc->worklist_mutex); 753 return; 754} 755