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 * 3. The names of the authors may not be used to endorse or promote 20 * products derived from this software without specific prior written 21 * permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/malloc.h> 43 #include <sys/bio.h> 44 #include <sys/ktr.h> 45 #include <sys/proc.h> 46 #include <sys/stack.h> 47 48 #include <sys/errno.h> 49 #include <geom/geom.h> 50 #include <geom/geom_int.h> 51 #include <sys/devicestat.h> 52 53 #include <vm/uma.h> 54 55 static struct g_bioq g_bio_run_down; 56 static struct g_bioq g_bio_run_up; 57 static struct g_bioq g_bio_run_task; 58 59 static u_int pace; 60 static uma_zone_t biozone; 61 62 #include <machine/atomic.h> 63 64 static void 65 g_bioq_lock(struct g_bioq *bq) 66 { 67 68 mtx_lock(&bq->bio_queue_lock); 69 } 70 71 static void 72 g_bioq_unlock(struct g_bioq *bq) 73 { 74 75 mtx_unlock(&bq->bio_queue_lock); 76 } 77 78 #if 0 79 static void 80 g_bioq_destroy(struct g_bioq *bq) 81 { 82 83 mtx_destroy(&bq->bio_queue_lock); 84 } 85 #endif 86 87 static void 88 g_bioq_init(struct g_bioq *bq) 89 { 90 91 TAILQ_INIT(&bq->bio_queue); 92 mtx_init(&bq->bio_queue_lock, "bio queue", NULL, MTX_DEF); 93 } 94 95 static struct bio * 96 g_bioq_first(struct g_bioq *bq) 97 { 98 struct bio *bp; 99 100 bp = TAILQ_FIRST(&bq->bio_queue); 101 if (bp != NULL) { 102 KASSERT((bp->bio_flags & BIO_ONQUEUE), 103 ("Bio not on queue bp=%p target %p", bp, bq)); 104 bp->bio_flags &= ~BIO_ONQUEUE; 105 TAILQ_REMOVE(&bq->bio_queue, bp, bio_queue); 106 bq->bio_queue_length--; 107 } 108 return (bp); 109 } 110 111 struct bio * 112 g_new_bio(void) 113 { 114 struct bio *bp; 115 116 bp = uma_zalloc(biozone, M_NOWAIT | M_ZERO); 117 #ifdef KTR 118 if (KTR_COMPILE & KTR_GEOM) { 119 struct stack st; 120 121 CTR1(KTR_GEOM, "g_new_bio(): %p", bp); 122 stack_save(&st); 123 CTRSTACK(KTR_GEOM, &st, 3, 0); 124 } 125 #endif 126 return (bp); 127 } 128 129 struct bio * 130 g_alloc_bio(void) 131 { 132 struct bio *bp; 133 134 bp = uma_zalloc(biozone, M_WAITOK | M_ZERO); 135 #ifdef KTR 136 if (KTR_COMPILE & KTR_GEOM) { 137 struct stack st; 138 139 CTR1(KTR_GEOM, "g_alloc_bio(): %p", bp); 140 stack_save(&st); 141 CTRSTACK(KTR_GEOM, &st, 3, 0); 142 } 143 #endif 144 return (bp); 145 } 146 147 void 148 g_destroy_bio(struct bio *bp) 149 { 150 #ifdef KTR 151 if (KTR_COMPILE & KTR_GEOM) { 152 struct stack st; 153 154 CTR1(KTR_GEOM, "g_destroy_bio(): %p", bp); 155 stack_save(&st); 156 CTRSTACK(KTR_GEOM, &st, 3, 0); 157 } 158 #endif 159 uma_zfree(biozone, bp); 160 } 161 162 struct bio * 163 g_clone_bio(struct bio *bp) 164 { 165 struct bio *bp2; 166 167 bp2 = uma_zalloc(biozone, M_NOWAIT | M_ZERO); 168 if (bp2 != NULL) { 169 bp2->bio_parent = bp; 170 bp2->bio_cmd = bp->bio_cmd; 171 bp2->bio_length = bp->bio_length; 172 bp2->bio_offset = bp->bio_offset; 173 bp2->bio_data = bp->bio_data; 174 bp2->bio_attribute = bp->bio_attribute; 175 bp->bio_children++; 176 } 177 #ifdef KTR 178 if (KTR_COMPILE & KTR_GEOM) { 179 struct stack st; 180 181 CTR2(KTR_GEOM, "g_clone_bio(%p): %p", bp, bp2); 182 stack_save(&st); 183 CTRSTACK(KTR_GEOM, &st, 3, 0); 184 } 185 #endif 186 return(bp2); 187 } 188 189 struct bio * 190 g_duplicate_bio(struct bio *bp) 191 { 192 struct bio *bp2; 193 194 bp2 = uma_zalloc(biozone, M_WAITOK | M_ZERO); 195 bp2->bio_parent = bp; 196 bp2->bio_cmd = bp->bio_cmd; 197 bp2->bio_length = bp->bio_length; 198 bp2->bio_offset = bp->bio_offset; 199 bp2->bio_data = bp->bio_data; 200 bp2->bio_attribute = bp->bio_attribute; 201 bp->bio_children++; 202 #ifdef KTR 203 if (KTR_COMPILE & KTR_GEOM) { 204 struct stack st; 205 206 CTR2(KTR_GEOM, "g_duplicate_bio(%p): %p", bp, bp2); 207 stack_save(&st); 208 CTRSTACK(KTR_GEOM, &st, 3, 0); 209 } 210 #endif 211 return(bp2); 212 } 213 214 void 215 g_io_init() 216 { 217 218 g_bioq_init(&g_bio_run_down); 219 g_bioq_init(&g_bio_run_up); 220 g_bioq_init(&g_bio_run_task); 221 biozone = uma_zcreate("g_bio", sizeof (struct bio), 222 NULL, NULL, 223 NULL, NULL, 224 0, 0); 225 } 226 227 int 228 g_io_getattr(const char *attr, struct g_consumer *cp, int *len, void *ptr) 229 { 230 struct bio *bp; 231 int error; 232 233 g_trace(G_T_BIO, "bio_getattr(%s)", attr); 234 bp = g_alloc_bio(); 235 bp->bio_cmd = BIO_GETATTR; 236 bp->bio_done = NULL; 237 bp->bio_attribute = attr; 238 bp->bio_length = *len; 239 bp->bio_data = ptr; 240 g_io_request(bp, cp); 241 error = biowait(bp, "ggetattr"); 242 *len = bp->bio_completed; 243 g_destroy_bio(bp); 244 return (error); 245 } 246 247 static int 248 g_io_check(struct bio *bp) 249 { 250 struct g_consumer *cp; 251 struct g_provider *pp; 252 253 cp = bp->bio_from; 254 pp = bp->bio_to; 255 256 /* Fail if access counters dont allow the operation */ 257 switch(bp->bio_cmd) { 258 case BIO_READ: 259 case BIO_GETATTR: 260 if (cp->acr == 0) 261 return (EPERM); 262 break; 263 case BIO_WRITE: 264 case BIO_DELETE: 265 if (cp->acw == 0) 266 return (EPERM); 267 break; 268 default: 269 return (EPERM); 270 } 271 /* if provider is marked for error, don't disturb. */ 272 if (pp->error) 273 return (pp->error); 274 275 switch(bp->bio_cmd) { 276 case BIO_READ: 277 case BIO_WRITE: 278 case BIO_DELETE: 279 /* Zero sectorsize is a probably lack of media */ 280 if (pp->sectorsize == 0) 281 return (ENXIO); 282 /* Reject I/O not on sector boundary */ 283 if (bp->bio_offset % pp->sectorsize) 284 return (EINVAL); 285 /* Reject I/O not integral sector long */ 286 if (bp->bio_length % pp->sectorsize) 287 return (EINVAL); 288 /* Reject requests before or past the end of media. */ 289 if (bp->bio_offset < 0) 290 return (EIO); 291 if (bp->bio_offset > pp->mediasize) 292 return (EIO); 293 break; 294 default: 295 break; 296 } 297 return (0); 298 } 299 300 void 301 g_io_request(struct bio *bp, struct g_consumer *cp) 302 { 303 struct g_provider *pp; 304 305 KASSERT(cp != NULL, ("NULL cp in g_io_request")); 306 KASSERT(bp != NULL, ("NULL bp in g_io_request")); 307 KASSERT(bp->bio_data != NULL, ("NULL bp->data in g_io_request")); 308 pp = cp->provider; 309 KASSERT(pp != NULL, ("consumer not attached in g_io_request")); 310 #ifdef DIAGNOSTIC 311 KASSERT(bp->bio_driver1 == NULL, 312 ("bio_driver1 used by the consumer (geom %s)", cp->geom->name)); 313 KASSERT(bp->bio_driver2 == NULL, 314 ("bio_driver2 used by the consumer (geom %s)", cp->geom->name)); 315 KASSERT(bp->bio_pflags == 0, 316 ("bio_pflags used by the consumer (geom %s)", cp->geom->name)); 317 /* 318 * Remember consumer's private fields, so we can detect if they were 319 * modified by the provider. 320 */ 321 bp->_bio_caller1 = bp->bio_caller1; 322 bp->_bio_caller2 = bp->bio_caller2; 323 bp->_bio_cflags = bp->bio_cflags; 324 #endif 325 326 if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) { 327 KASSERT(bp->bio_offset % cp->provider->sectorsize == 0, 328 ("wrong offset %jd for sectorsize %u", 329 bp->bio_offset, cp->provider->sectorsize)); 330 KASSERT(bp->bio_length % cp->provider->sectorsize == 0, 331 ("wrong length %jd for sectorsize %u", 332 bp->bio_length, cp->provider->sectorsize)); 333 } 334 335 g_trace(G_T_BIO, "bio_request(%p) from %p(%s) to %p(%s) cmd %d", 336 bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd); 337 338 bp->bio_from = cp; 339 bp->bio_to = pp; 340 bp->bio_error = 0; 341 bp->bio_completed = 0; 342 343 KASSERT(!(bp->bio_flags & BIO_ONQUEUE), 344 ("Bio already on queue bp=%p", bp)); 345 bp->bio_flags |= BIO_ONQUEUE; 346 347 binuptime(&bp->bio_t0); 348 349 /* 350 * The statistics collection is lockless, as such, but we 351 * can not update one instance of the statistics from more 352 * than one thread at a time, so grab the lock first. 353 */ 354 g_bioq_lock(&g_bio_run_down); 355 if (g_collectstats & 1) 356 devstat_start_transaction(pp->stat, &bp->bio_t0); 357 if (g_collectstats & 2) 358 devstat_start_transaction(cp->stat, &bp->bio_t0); 359 360 pp->nstart++; 361 cp->nstart++; 362 TAILQ_INSERT_TAIL(&g_bio_run_down.bio_queue, bp, bio_queue); 363 g_bio_run_down.bio_queue_length++; 364 g_bioq_unlock(&g_bio_run_down); 365 366 /* Pass it on down. */ 367 wakeup(&g_wait_down); 368 } 369 370 void 371 g_io_deliver(struct bio *bp, int error) 372 { 373 struct g_consumer *cp; 374 struct g_provider *pp; 375 376 KASSERT(bp != NULL, ("NULL bp in g_io_deliver")); 377 pp = bp->bio_to; 378 KASSERT(pp != NULL, ("NULL bio_to in g_io_deliver")); 379 #ifdef DIAGNOSTIC 380 KASSERT(bp->bio_caller1 == bp->_bio_caller1, 381 ("bio_caller1 used by the provider %s", pp->name)); 382 KASSERT(bp->bio_caller2 == bp->_bio_caller2, 383 ("bio_caller2 used by the provider %s", pp->name)); 384 KASSERT(bp->bio_cflags == bp->_bio_cflags, 385 ("bio_cflags used by the provider %s", pp->name)); 386 #endif 387 cp = bp->bio_from; 388 if (cp == NULL) { 389 bp->bio_error = error; 390 bp->bio_done(bp); 391 return; 392 } 393 KASSERT(cp != NULL, ("NULL bio_from in g_io_deliver")); 394 KASSERT(cp->geom != NULL, ("NULL bio_from->geom in g_io_deliver")); 395 KASSERT(bp->bio_completed >= 0, ("bio_completed can't be less than 0")); 396 KASSERT(bp->bio_completed <= bp->bio_length, 397 ("bio_completed can't be greater than bio_length")); 398 399 g_trace(G_T_BIO, 400 "g_io_deliver(%p) from %p(%s) to %p(%s) cmd %d error %d off %jd len %jd", 401 bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd, error, 402 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length); 403 404 KASSERT(!(bp->bio_flags & BIO_ONQUEUE), 405 ("Bio already on queue bp=%p", bp)); 406 407 /* 408 * XXX: next two doesn't belong here 409 */ 410 bp->bio_bcount = bp->bio_length; 411 bp->bio_resid = bp->bio_bcount - bp->bio_completed; 412 413 /* 414 * The statistics collection is lockless, as such, but we 415 * can not update one instance of the statistics from more 416 * than one thread at a time, so grab the lock first. 417 */ 418 g_bioq_lock(&g_bio_run_up); 419 if (g_collectstats & 1) 420 devstat_end_transaction_bio(pp->stat, bp); 421 if (g_collectstats & 2) 422 devstat_end_transaction_bio(cp->stat, bp); 423 424 cp->nend++; 425 pp->nend++; 426 if (error != ENOMEM) { 427 bp->bio_error = error; 428 TAILQ_INSERT_TAIL(&g_bio_run_up.bio_queue, bp, bio_queue); 429 bp->bio_flags |= BIO_ONQUEUE; 430 g_bio_run_up.bio_queue_length++; 431 g_bioq_unlock(&g_bio_run_up); 432 wakeup(&g_wait_up); 433 return; 434 } 435 g_bioq_unlock(&g_bio_run_up); 436 437 if (bootverbose) 438 printf("ENOMEM %p on %p(%s)\n", bp, pp, pp->name); 439 bp->bio_children = 0; 440 bp->bio_inbed = 0; 441 g_io_request(bp, cp); 442 pace++; 443 return; 444 } 445 446 void 447 g_io_schedule_down(struct thread *tp __unused) 448 { 449 struct bio *bp; 450 off_t excess; 451 int error; 452 453 for(;;) { 454 g_bioq_lock(&g_bio_run_down); 455 bp = g_bioq_first(&g_bio_run_down); 456 if (bp == NULL) { 457 CTR0(KTR_GEOM, "g_down going to sleep"); 458 msleep(&g_wait_down, &g_bio_run_down.bio_queue_lock, 459 PRIBIO | PDROP, "-", hz/10); 460 continue; 461 } 462 CTR0(KTR_GEOM, "g_down has work to do"); 463 g_bioq_unlock(&g_bio_run_down); 464 if (pace > 0) { 465 CTR1(KTR_GEOM, "g_down pacing self (pace %d)", pace); 466 msleep(&error, NULL, PRIBIO, "g_down", hz/10); 467 pace--; 468 } 469 error = g_io_check(bp); 470 if (error) { 471 CTR3(KTR_GEOM, "g_down g_io_check on bp %p provider " 472 "%s returned %d", bp, bp->bio_to->name, error); 473 g_io_deliver(bp, error); 474 continue; 475 } 476 CTR2(KTR_GEOM, "g_down processing bp %p provider %s", bp, 477 bp->bio_to->name); 478 switch (bp->bio_cmd) { 479 case BIO_READ: 480 case BIO_WRITE: 481 case BIO_DELETE: 482 /* Truncate requests to the end of providers media. */ 483 /* 484 * XXX: What if we truncate because of offset being 485 * bad, not length? 486 */ 487 excess = bp->bio_offset + bp->bio_length; 488 if (excess > bp->bio_to->mediasize) { 489 excess -= bp->bio_to->mediasize; 490 bp->bio_length -= excess; 491 if (excess > 0) 492 CTR3(KTR_GEOM, "g_down truncated bio " 493 "%p provider %s by %d", bp, 494 bp->bio_to->name, excess); 495 } 496 /* Deliver zero length transfers right here. */ 497 if (bp->bio_length == 0) { 498 g_io_deliver(bp, 0); 499 CTR2(KTR_GEOM, "g_down terminated 0-length " 500 "bp %p provider %s", bp, bp->bio_to->name); 501 continue; 502 } 503 break; 504 default: 505 break; 506 } 507 THREAD_NO_SLEEPING(); 508 CTR4(KTR_GEOM, "g_down starting bp %p provider %s off %ld " 509 "len %ld", bp, bp->bio_to->name, bp->bio_offset, 510 bp->bio_length); 511 bp->bio_to->geom->start(bp); 512 THREAD_SLEEPING_OK(); 513 } 514 } 515 516 void 517 bio_taskqueue(struct bio *bp, bio_task_t *func, void *arg) 518 { 519 bp->bio_task = func; 520 bp->bio_task_arg = arg; 521 /* 522 * The taskqueue is actually just a second queue off the "up" 523 * queue, so we use the same lock. 524 */ 525 g_bioq_lock(&g_bio_run_up); 526 KASSERT(!(bp->bio_flags & BIO_ONQUEUE), 527 ("Bio already on queue bp=%p target taskq", bp)); 528 bp->bio_flags |= BIO_ONQUEUE; 529 TAILQ_INSERT_TAIL(&g_bio_run_task.bio_queue, bp, bio_queue); 530 g_bio_run_task.bio_queue_length++; 531 wakeup(&g_wait_up); 532 g_bioq_unlock(&g_bio_run_up); 533 } 534 535 536 void 537 g_io_schedule_up(struct thread *tp __unused) 538 { 539 struct bio *bp; 540 for(;;) { 541 g_bioq_lock(&g_bio_run_up); 542 bp = g_bioq_first(&g_bio_run_task); 543 if (bp != NULL) { 544 g_bioq_unlock(&g_bio_run_up); 545 THREAD_NO_SLEEPING(); 546 CTR1(KTR_GEOM, "g_up processing task bp %p", bp); 547 bp->bio_task(bp->bio_task_arg); 548 THREAD_SLEEPING_OK(); 549 continue; 550 } 551 bp = g_bioq_first(&g_bio_run_up); 552 if (bp != NULL) { 553 g_bioq_unlock(&g_bio_run_up); 554 THREAD_NO_SLEEPING(); 555 CTR4(KTR_GEOM, "g_up biodone bp %p provider %s off " 556 "%ld len %ld", bp, bp->bio_to->name, 557 bp->bio_offset, bp->bio_length); 558 biodone(bp); 559 THREAD_SLEEPING_OK(); 560 continue; 561 } 562 CTR0(KTR_GEOM, "g_up going to sleep"); 563 msleep(&g_wait_up, &g_bio_run_up.bio_queue_lock, 564 PRIBIO | PDROP, "-", hz/10); 565 } 566 } 567 568 void * 569 g_read_data(struct g_consumer *cp, off_t offset, off_t length, int *error) 570 { 571 struct bio *bp; 572 void *ptr; 573 int errorc; 574 575 KASSERT(length > 0 && length >= cp->provider->sectorsize && 576 length <= MAXPHYS, ("g_read_data(): invalid length %jd", 577 (intmax_t)length)); 578 579 bp = g_alloc_bio(); 580 bp->bio_cmd = BIO_READ; 581 bp->bio_done = NULL; 582 bp->bio_offset = offset; 583 bp->bio_length = length; 584 ptr = g_malloc(length, M_WAITOK); 585 bp->bio_data = ptr; 586 g_io_request(bp, cp); 587 errorc = biowait(bp, "gread"); 588 if (error != NULL) 589 *error = errorc; 590 g_destroy_bio(bp); 591 if (errorc) { 592 g_free(ptr); 593 ptr = NULL; 594 } 595 return (ptr); 596 } 597 598 int 599 g_write_data(struct g_consumer *cp, off_t offset, void *ptr, off_t length) 600 { 601 struct bio *bp; 602 int error; 603 604 KASSERT(length > 0 && length >= cp->provider->sectorsize && 605 length <= MAXPHYS, ("g_write_data(): invalid length %jd", 606 (intmax_t)length)); 607 608 bp = g_alloc_bio(); 609 bp->bio_cmd = BIO_WRITE; 610 bp->bio_done = NULL; 611 bp->bio_offset = offset; 612 bp->bio_length = length; 613 bp->bio_data = ptr; 614 g_io_request(bp, cp); 615 error = biowait(bp, "gwrite"); 616 g_destroy_bio(bp); 617 return (error); 618 } 619 620 void 621 g_print_bio(struct bio *bp) 622 { 623 const char *pname, *cmd = NULL; 624 625 if (bp->bio_to != NULL) 626 pname = bp->bio_to->name; 627 else 628 pname = "[unknown]"; 629 630 switch (bp->bio_cmd) { 631 case BIO_GETATTR: 632 cmd = "GETATTR"; 633 printf("%s[%s(attr=%s)]", pname, cmd, bp->bio_attribute); 634 return; 635 case BIO_READ: 636 cmd = "READ"; 637 case BIO_WRITE: 638 if (cmd == NULL) 639 cmd = "WRITE"; 640 case BIO_DELETE: 641 if (cmd == NULL) 642 cmd = "DELETE"; 643 printf("%s[%s(offset=%jd, length=%jd)]", pname, cmd, 644 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length); 645 return; 646 default: 647 cmd = "UNKNOWN"; 648 printf("%s[%s()]", pname, cmd); 649 return; 650 } 651 /* NOTREACHED */ 652 } 653