g_bde_work.c revision 108558
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 108558 2003-01-02 19:29:49Z 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/stdint.h> 62#include <sys/bio.h> 63#include <sys/lock.h> 64#include <sys/mutex.h> 65#include <sys/queue.h> 66#include <sys/malloc.h> 67#include <sys/systm.h> 68#include <sys/kernel.h> 69#include <sys/sysctl.h> 70#include <sys/time.h> 71#include <sys/proc.h> 72#include <sys/kthread.h> 73 74#include <crypto/rijndael/rijndael.h> 75#include <crypto/sha2/sha2.h> 76#include <geom/geom.h> 77#include <geom/bde/g_bde.h> 78 79static void g_bde_delete_sector(struct g_bde_softc *wp, struct g_bde_sector *sp); 80static struct g_bde_sector * g_bde_new_sector(struct g_bde_work *wp, u_int len); 81static void g_bde_release_sector(struct g_bde_work *wp, struct g_bde_sector *sp); 82static struct g_bde_sector *g_bde_get_sector(struct g_bde_work *wp, off_t offset); 83static int g_bde_start_read(struct g_bde_sector *sp); 84static void g_bde_purge_sector(struct g_bde_softc *sc, int fraction); 85 86/* 87 * Work item allocation. 88 * 89 * C++ would call these constructors and destructors. 90 */ 91static u_int g_bde_nwork; 92SYSCTL_UINT(_debug, OID_AUTO, gbde_nwork, CTLFLAG_RD, &g_bde_nwork, 0, ""); 93 94static struct g_bde_work * 95g_bde_new_work(struct g_bde_softc *sc) 96{ 97 struct g_bde_work *wp; 98 99 wp = g_malloc(sizeof *wp, M_NOWAIT | M_ZERO); 100 if (wp == NULL) 101 return (wp); 102 wp->state = SETUP; 103 wp->softc = sc; 104 g_bde_nwork++; 105 sc->nwork++; 106 TAILQ_INSERT_TAIL(&sc->worklist, wp, list); 107 return (wp); 108} 109 110static void 111g_bde_delete_work(struct g_bde_work *wp) 112{ 113 struct g_bde_softc *sc; 114 115 sc = wp->softc; 116 g_bde_nwork--; 117 sc->nwork--; 118 TAILQ_REMOVE(&sc->worklist, wp, list); 119 g_free(wp); 120} 121 122/* 123 * Sector buffer allocation 124 * 125 * These two functions allocate and free back variable sized sector buffers 126 */ 127 128static u_int g_bde_nsect; 129SYSCTL_UINT(_debug, OID_AUTO, gbde_nsect, CTLFLAG_RD, &g_bde_nsect, 0, ""); 130 131static void 132g_bde_delete_sector(struct g_bde_softc *sc, struct g_bde_sector *sp) 133{ 134 135 g_bde_nsect--; 136 sc->nsect--; 137 if (sp->malloc) 138 g_free(sp->data); 139 g_free(sp); 140} 141 142static struct g_bde_sector * 143g_bde_new_sector(struct g_bde_work *wp, u_int len) 144{ 145 struct g_bde_sector *sp; 146 147 sp = g_malloc(sizeof *sp, M_NOWAIT | M_ZERO); 148 if (sp == NULL) 149 return (sp); 150 if (len > 0) { 151 sp->data = g_malloc(len, M_NOWAIT | M_ZERO); 152 if (sp->data == NULL) { 153 g_free(sp); 154 return (NULL); 155 } 156 sp->malloc = 1; 157 } 158 g_bde_nsect++; 159 wp->softc->nsect++; 160 sp->size = len; 161 sp->softc = wp->softc; 162 sp->ref = 1; 163 sp->owner = wp; 164 sp->offset = wp->so; 165 sp->state = JUNK; 166 return (sp); 167} 168 169/* 170 * Skey sector cache. 171 * 172 * Nothing prevents two separate I/O requests from addressing the same zone 173 * and thereby needing the same skey sector. We therefore need to sequence 174 * I/O operations to the skey sectors. A certain amount of caching is also 175 * desirable, although the extent of benefit from this is not at this point 176 * determined. 177 * 178 * XXX: GEOM may be able to grow a generic caching facility at some point 179 * XXX: to support such needs. 180 */ 181 182static u_int g_bde_ncache; 183SYSCTL_UINT(_debug, OID_AUTO, gbde_ncache, CTLFLAG_RD, &g_bde_ncache, 0, ""); 184 185static void 186g_bde_purge_one_sector(struct g_bde_softc *sc, struct g_bde_sector *sp) 187{ 188 189 g_trace(G_T_TOPOLOGY, "g_bde_purge_one_sector(%p, %p)", sc, sp); 190 if (sp->ref != 0) 191 return; 192 TAILQ_REMOVE(&sc->freelist, sp, list); 193 g_bde_ncache--; 194 sc->ncache--; 195 bzero(sp->data, sp->size); 196 g_bde_delete_sector(sc, sp); 197} 198 199static struct g_bde_sector * 200g_bde_get_sector(struct g_bde_work *wp, off_t offset) 201{ 202 struct g_bde_sector *sp; 203 struct g_bde_softc *sc; 204 205 g_trace(G_T_TOPOLOGY, "g_bde_get_sector(%p, %jd)", wp, (intmax_t)offset); 206 sc = wp->softc; 207 208 if (malloc_last_fail() < g_bde_ncache) 209 g_bde_purge_sector(sc, -1); 210 211 sp = TAILQ_FIRST(&sc->freelist); 212 if (sp != NULL && sp->ref == 0 && sp->used + 300 < time_uptime) 213 g_bde_purge_one_sector(sc, sp); 214 215 TAILQ_FOREACH(sp, &sc->freelist, list) { 216 if (sp->offset == offset) 217 break; 218 } 219 if (sp != NULL) { 220 sp->ref++; 221 KASSERT(sp->offset == offset, ("wrong offset")); 222 KASSERT(sp->softc == wp->softc, ("wrong softc")); 223 if (sp->ref == 1) 224 sp->owner = wp; 225 } else { 226 if (malloc_last_fail() < g_bde_ncache) { 227 TAILQ_FOREACH(sp, &sc->freelist, list) 228 if (sp->ref == 0) 229 break; 230 } 231 if (sp == NULL && !TAILQ_EMPTY(&sc->freelist)) 232 sp = TAILQ_FIRST(&sc->freelist); 233 if (sp != NULL && sp->ref > 0) 234 sp = NULL; 235 if (sp == NULL) { 236 g_bde_ncache++; 237 sc->ncache++; 238 sp = g_bde_new_sector(wp, sc->sectorsize); 239 if (sp != NULL) { 240 TAILQ_INSERT_TAIL(&sc->freelist, sp, list); 241 sp->malloc = 2; 242 } 243 } 244 if (sp != NULL) { 245 sp->offset = offset; 246 sp->softc = wp->softc; 247 sp->ref = 1; 248 sp->owner = wp; 249 sp->state = JUNK; 250 sp->error = 0; 251 } 252 } 253 if (sp != NULL) { 254 TAILQ_REMOVE(&sc->freelist, sp, list); 255 TAILQ_INSERT_TAIL(&sc->freelist, sp, list); 256 } 257 wp->ksp = sp; 258 if (sp == NULL) { 259 g_bde_purge_sector(sc, -1); 260 sp = g_bde_get_sector(wp, offset); 261 } 262 if (sp != NULL) 263 sp->used = time_uptime; 264 KASSERT(sp != NULL, ("get_sector failed")); 265 return(sp); 266} 267 268static void 269g_bde_release_sector(struct g_bde_work *wp, struct g_bde_sector *sp) 270{ 271 struct g_bde_softc *sc; 272 struct g_bde_work *wp2; 273 274 g_trace(G_T_TOPOLOGY, "g_bde_release_sector(%p)", sp); 275 KASSERT(sp->malloc == 2, ("Wrong sector released")); 276 sc = sp->softc; 277 KASSERT(sc != NULL, ("NULL sp->softc")); 278 KASSERT(wp == sp->owner, ("Releasing, not owner")); 279 sp->owner = NULL; 280 wp->ksp = NULL; 281 sp->ref--; 282 if (sp->ref > 0) { 283 TAILQ_REMOVE(&sc->freelist, sp, list); 284 TAILQ_INSERT_TAIL(&sc->freelist, sp, list); 285 TAILQ_FOREACH(wp2, &sc->worklist, list) { 286 if (wp2->ksp == sp) { 287 KASSERT(wp2 != wp, ("Self-reowning")); 288 sp->owner = wp2; 289 wakeup(sp->softc); 290 break; 291 } 292 } 293 KASSERT(wp2 != NULL, ("Failed to pick up owner for %p\n", sp)); 294 } else if (sp->error != 0) { 295 sp->offset = ~0; 296 sp->error = 0; 297 sp->state = JUNK; 298 } 299 TAILQ_REMOVE(&sc->freelist, sp, list); 300 TAILQ_INSERT_HEAD(&sc->freelist, sp, list); 301} 302 303static void 304g_bde_purge_sector(struct g_bde_softc *sc, int fraction) 305{ 306 struct g_bde_sector *sp; 307 int n; 308 309 g_trace(G_T_TOPOLOGY, "g_bde_purge_sector(%p)", sc); 310 if (fraction > 0) 311 n = sc->ncache / fraction + 1; 312 else 313 n = g_bde_ncache - malloc_last_fail(); 314 if (n < 0) 315 return; 316 if (n > sc->ncache) 317 n = sc->ncache; 318 while(n--) { 319 TAILQ_FOREACH(sp, &sc->freelist, list) { 320 if (sp->ref != 0) 321 continue; 322 TAILQ_REMOVE(&sc->freelist, sp, list); 323 g_bde_ncache--; 324 sc->ncache--; 325 bzero(sp->data, sp->size); 326 g_bde_delete_sector(sc, sp); 327 break; 328 } 329 } 330} 331 332static struct g_bde_sector * 333g_bde_read_sector(struct g_bde_softc *sc, struct g_bde_work *wp, off_t offset) 334{ 335 struct g_bde_sector *sp; 336 337 g_trace(G_T_TOPOLOGY, "g_bde_read_sector(%p)", wp); 338 sp = g_bde_get_sector(wp, offset); 339 if (sp == NULL) 340 return (sp); 341 if (sp->owner != wp) 342 return (sp); 343 if (sp->state == VALID) 344 return (sp); 345 if (g_bde_start_read(sp) == 0) 346 return (sp); 347 g_bde_release_sector(wp, sp); 348 return (NULL); 349} 350 351/* 352 * Contribute to the completion of the original bio request. 353 * 354 * We have no simple way to tell how many bits the original bio request has 355 * been segmented into, so the easiest way to determine when we can deliver 356 * it is to keep track of the number of bytes we have completed. We keep 357 * track of any errors underway and latch onto the first one. 358 * 359 * We always report "nothing done" in case of error, because random bits here 360 * and there may be completed and returning a number of completed bytes does 361 * not convey any useful information about which bytes they were. If some 362 * piece of broken code somewhere interprets this to mean that nothing has 363 * changed on the underlying media they deserve the lossage headed for them. 364 * 365 * A single mutex per g_bde instance is used to prevent contention. 366 */ 367 368static void 369g_bde_contribute(struct bio *bp, off_t bytes, int error) 370{ 371 struct g_bde_softc *sc; 372 373 g_trace(G_T_TOPOLOGY, "g_bde_contribute bp %p bytes %jd error %d", 374 bp, (intmax_t)bytes, error); 375 sc = bp->bio_driver1; 376 if (bp->bio_error == 0) 377 bp->bio_error = error; 378 bp->bio_completed += bytes; 379 KASSERT(bp->bio_completed <= bp->bio_length, ("Too large contribution")); 380 if (bp->bio_completed == bp->bio_length) { 381 if (bp->bio_error != 0) 382 bp->bio_completed = 0; 383 g_io_deliver(bp, bp->bio_error); 384 } 385} 386 387/* 388 * A write operation has finished. When we have all expected cows in the 389 * barn close the door and call it a day. 390 */ 391 392static void 393g_bde_write_done(struct bio *bp) 394{ 395 struct g_bde_sector *sp; 396 struct g_bde_work *wp; 397 struct g_bde_softc *sc; 398 399 sp = bp->bio_caller1; 400 sc = bp->bio_caller2; 401 mtx_lock(&sc->worklist_mutex); 402 KASSERT(sp != NULL, ("NULL sp")); 403 KASSERT(sc != NULL, ("NULL sc")); 404 KASSERT(sp->owner != NULL, ("NULL sp->owner")); 405 g_trace(G_T_TOPOLOGY, "g_bde_write_done(%p)", sp); 406 sp->error = bp->bio_error; 407 g_destroy_bio(bp); 408 wp = sp->owner; 409 if (wp->error == 0) 410 wp->error = sp->error; 411 412 if (wp->bp->bio_cmd == BIO_DELETE) { 413 KASSERT(sp == wp->sp, ("trashed delete op")); 414 g_bde_contribute(wp->bp, wp->length, wp->error); 415 g_bde_delete_sector(sc, sp); 416 g_bde_delete_work(wp); 417 mtx_unlock(&sc->worklist_mutex); 418 return; 419 } 420 421 KASSERT(wp->bp->bio_cmd == BIO_WRITE, ("Confused in g_bde_write_done()")); 422 KASSERT(sp == wp->sp || sp == wp->ksp, ("trashed write op")); 423 if (wp->sp == sp) { 424 g_bde_delete_sector(sc, wp->sp); 425 wp->sp = NULL; 426 } else { 427 sp->state = VALID; 428 } 429 if (wp->sp == NULL && wp->ksp != NULL && wp->ksp->state == VALID) { 430 g_bde_contribute(wp->bp, wp->length, wp->error); 431 g_bde_release_sector(wp, wp->ksp); 432 g_bde_delete_work(wp); 433 } 434 mtx_unlock(&sc->worklist_mutex); 435 return; 436} 437 438/* 439 * Send a write request for the given sector down the pipeline. 440 */ 441 442static int 443g_bde_start_write(struct g_bde_sector *sp) 444{ 445 struct bio *bp; 446 struct g_bde_softc *sc; 447 448 g_trace(G_T_TOPOLOGY, "g_bde_start_write(%p)", sp); 449 sc = sp->softc; 450 KASSERT(sc != NULL, ("NULL sc in g_bde_start_write")); 451 KASSERT(sp->owner != NULL, ("NULL sp->owner in g_bde_start_write")); 452 bp = g_new_bio(); 453 if (bp == NULL) 454 return (ENOMEM); 455 bp->bio_cmd = BIO_WRITE; 456 bp->bio_offset = sp->offset; 457 bp->bio_data = sp->data; 458 bp->bio_length = sp->size; 459 bp->bio_done = g_bde_write_done; 460 bp->bio_caller1 = sp; 461 bp->bio_caller2 = sc; 462 sp->state = IO; 463 g_io_request(bp, sc->consumer); 464 return(0); 465} 466 467/* 468 * A read operation has finished. Mark the sector no longer iobusy and 469 * wake up the worker thread and let it do its thing. 470 */ 471 472static void 473g_bde_read_done(struct bio *bp) 474{ 475 struct g_bde_sector *sp; 476 struct g_bde_softc *sc; 477 478 sp = bp->bio_caller1; 479 g_trace(G_T_TOPOLOGY, "g_bde_read_done(%p)", sp); 480 sc = bp->bio_caller2; 481 mtx_lock(&sc->worklist_mutex); 482 sp->error = bp->bio_error; 483 sp->state = VALID; 484 wakeup(sc); 485 g_destroy_bio(bp); 486 mtx_unlock(&sc->worklist_mutex); 487} 488 489/* 490 * Send a read request for the given sector down the pipeline. 491 */ 492 493static int 494g_bde_start_read(struct g_bde_sector *sp) 495{ 496 struct bio *bp; 497 struct g_bde_softc *sc; 498 499 g_trace(G_T_TOPOLOGY, "g_bde_start_read(%p)", sp); 500 sc = sp->softc; 501 KASSERT(sc != NULL, ("Null softc in sp %p", sp)); 502 bp = g_new_bio(); 503 if (bp == NULL) 504 return (ENOMEM); 505 bp->bio_cmd = BIO_READ; 506 bp->bio_offset = sp->offset; 507 bp->bio_data = sp->data; 508 bp->bio_length = sp->size; 509 bp->bio_done = g_bde_read_done; 510 bp->bio_caller1 = sp; 511 bp->bio_caller2 = sc; 512 sp->state = IO; 513 g_io_request(bp, sc->consumer); 514 return(0); 515} 516 517/* 518 * The worker thread. 519 * 520 * The up/down path of GEOM is not allowed to sleep or do any major work 521 * so we use this thread to do the actual crypto operations and to push 522 * the state engine onwards. 523 * 524 * XXX: if we switch to the src/sys/opencrypt hardware assisted encryption 525 * XXX: using a thread here is probably not needed. 526 */ 527 528void 529g_bde_worker(void *arg) 530{ 531 struct g_bde_softc *sc; 532 struct g_bde_work *wp; 533 struct g_geom *gp; 534 int busy, error; 535 536 gp = arg; 537 sc = gp->softc; 538 539 mtx_lock(&sc->worklist_mutex); 540 for (;;) { 541 busy = 0; 542 g_trace(G_T_TOPOLOGY, "g_bde_worker scan"); 543 TAILQ_FOREACH(wp, &sc->worklist, list) { 544 KASSERT(wp != NULL, ("NULL wp")); 545 KASSERT(wp->softc != NULL, ("NULL wp->softc")); 546 if (wp->state != WAIT) 547 continue; /* Not interesting here */ 548 549 KASSERT(wp->bp != NULL, ("NULL wp->bp")); 550 KASSERT(wp->sp != NULL, ("NULL wp->sp")); 551 552 if (wp->ksp != NULL) { 553 if (wp->ksp->owner != wp) 554 continue; 555 if (wp->ksp->state == IO) 556 continue; 557 KASSERT(wp->ksp->state == VALID, 558 ("Illegal sector state (JUNK ?)")); 559 } 560 561 if (wp->bp->bio_cmd == BIO_READ && wp->sp->state != VALID) 562 continue; 563 564 if (wp->ksp != NULL && wp->ksp->error != 0) { 565 g_bde_contribute(wp->bp, wp->length, 566 wp->ksp->error); 567 g_bde_delete_sector(sc, wp->sp); 568 g_bde_release_sector(wp, wp->ksp); 569 g_bde_delete_work(wp); 570 busy++; 571 break; 572 } 573 switch(wp->bp->bio_cmd) { 574 case BIO_READ: 575 if (wp->ksp != NULL && wp->sp->error == 0) { 576 mtx_unlock(&sc->worklist_mutex); 577 g_bde_crypt_read(wp); 578 mtx_lock(&sc->worklist_mutex); 579 } 580 g_bde_contribute(wp->bp, wp->length, 581 wp->sp->error); 582 g_bde_delete_sector(sc, wp->sp); 583 if (wp->ksp != NULL) 584 g_bde_release_sector(wp, wp->ksp); 585 g_bde_delete_work(wp); 586 break; 587 case BIO_WRITE: 588 wp->state = FINISH; 589 KASSERT(wp->sp->owner == wp, ("Write not owner sp")); 590 KASSERT(wp->ksp->owner == wp, ("Write not owner ksp")); 591 mtx_unlock(&sc->worklist_mutex); 592 g_bde_crypt_write(wp); 593 mtx_lock(&sc->worklist_mutex); 594 g_bde_start_write(wp->sp); 595 g_bde_start_write(wp->ksp); 596 break; 597 case BIO_DELETE: 598 wp->state = FINISH; 599 mtx_unlock(&sc->worklist_mutex); 600 g_bde_crypt_delete(wp); 601 mtx_lock(&sc->worklist_mutex); 602 g_bde_start_write(wp->sp); 603 break; 604 } 605 busy++; 606 break; 607 } 608 if (!busy) { 609 /* 610 * We don't look for our death-warrant until we are 611 * idle. Shouldn't make a difference in practice. 612 */ 613 if (sc->dead) 614 break; 615 g_trace(G_T_TOPOLOGY, "g_bde_worker sleep"); 616 error = msleep(sc, &sc->worklist_mutex, 617 PRIBIO, "g_bde", hz); 618 if (error == EWOULDBLOCK) { 619 /* 620 * Loose our skey cache in an orderly fashion. 621 * The exact rate can be tuned to be less 622 * aggressive if this is desirable. 10% per 623 * second means that the cache is gone in a 624 * few minutes. 625 */ 626 g_bde_purge_sector(sc, 10); 627 } 628 } 629 } 630 g_trace(G_T_TOPOLOGY, "g_bde_worker die"); 631 g_bde_purge_sector(sc, 1); 632 KASSERT(sc->nwork == 0, ("Dead but %d work remaining", sc->nwork)); 633 KASSERT(sc->ncache == 0, ("Dead but %d cache remaining", sc->ncache)); 634 KASSERT(sc->nsect == 0, ("Dead but %d sect remaining", sc->nsect)); 635 mtx_unlock(&sc->worklist_mutex); 636 sc->dead = 2; 637 wakeup(sc); 638 mtx_lock(&Giant); 639 kthread_exit(0); 640} 641 642/* 643 * g_bde_start1 has chopped the incoming request up so all the requests 644 * we see here are inside a single zone. Map the data and key locations 645 * grab the buffers we need and fire off the first volley of read requests. 646 */ 647 648static void 649g_bde_start2(struct g_bde_work *wp) 650{ 651 struct g_bde_softc *sc; 652 653 KASSERT(wp != NULL, ("NULL wp in g_bde_start2")); 654 KASSERT(wp->softc != NULL, ("NULL wp->softc")); 655 g_trace(G_T_TOPOLOGY, "g_bde_start2(%p)", wp); 656 sc = wp->softc; 657 if (wp->bp->bio_cmd == BIO_READ) { 658 wp->sp = g_bde_new_sector(wp, 0); 659 if (wp->sp == NULL) { 660 g_bde_contribute(wp->bp, wp->length, ENOMEM); 661 g_bde_delete_work(wp); 662 return; 663 } 664 wp->sp->size = wp->length; 665 wp->sp->data = wp->data; 666 if (g_bde_start_read(wp->sp) != 0) { 667 g_bde_contribute(wp->bp, wp->length, ENOMEM); 668 g_bde_delete_sector(sc, wp->sp); 669 g_bde_delete_work(wp); 670 return; 671 } 672 g_bde_read_sector(sc, wp, wp->kso); 673 if (wp->ksp == NULL) 674 wp->error = ENOMEM; 675 } else if (wp->bp->bio_cmd == BIO_DELETE) { 676 wp->sp = g_bde_new_sector(wp, wp->length); 677 if (wp->sp == NULL) { 678 g_bde_contribute(wp->bp, wp->length, ENOMEM); 679 g_bde_delete_work(wp); 680 return; 681 } 682 } else if (wp->bp->bio_cmd == BIO_WRITE) { 683 wp->sp = g_bde_new_sector(wp, wp->length); 684 if (wp->sp == NULL) { 685 g_bde_contribute(wp->bp, wp->length, ENOMEM); 686 g_bde_delete_work(wp); 687 return; 688 } 689 g_bde_read_sector(sc, wp, wp->kso); 690 if (wp->ksp == NULL) { 691 g_bde_contribute(wp->bp, wp->length, ENOMEM); 692 g_bde_delete_sector(sc, wp->sp); 693 g_bde_delete_work(wp); 694 return; 695 } 696 } else { 697 KASSERT(0 == 1, 698 ("Wrong bio_cmd %d in g_bde_start2", wp->bp->bio_cmd)); 699 } 700 701 wp->state = WAIT; 702 wakeup(sc); 703} 704 705/* 706 * Create a sequence of work structures, and have g_bde_map_sector() determine 707 * how long they each can be. Feed them to g_bde_start2(). 708 */ 709 710void 711g_bde_start1(struct bio *bp) 712{ 713 struct g_bde_softc *sc; 714 struct g_bde_work *wp; 715 off_t done; 716 717 sc = bp->bio_to->geom->softc; 718 bp->bio_driver1 = sc; 719 720 mtx_lock(&sc->worklist_mutex); 721 for(done = 0; done < bp->bio_length; ) { 722 wp = g_bde_new_work(sc); 723 if (wp == NULL) { 724 g_io_deliver(bp, ENOMEM); 725 mtx_unlock(&sc->worklist_mutex); 726 return; 727 } 728 wp->bp = bp; 729 wp->offset = bp->bio_offset + done; 730 wp->data = bp->bio_data + done; 731 wp->length = bp->bio_length - done; 732 g_bde_map_sector(wp); 733 done += wp->length; 734 g_bde_start2(wp); 735 } 736 mtx_unlock(&sc->worklist_mutex); 737 return; 738} 739