xref: /freebsd/sys/geom/geom_io.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
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 /*
63  * The head of the list of classifiers used in g_io_request.
64  * Use g_register_classifier() and g_unregister_classifier()
65  * to add/remove entries to the list.
66  * Classifiers are invoked in registration order.
67  */
68 static TAILQ_HEAD(g_classifier_tailq, g_classifier_hook)
69     g_classifier_tailq = TAILQ_HEAD_INITIALIZER(g_classifier_tailq);
70 
71 #include <machine/atomic.h>
72 
73 static void
74 g_bioq_lock(struct g_bioq *bq)
75 {
76 
77 	mtx_lock(&bq->bio_queue_lock);
78 }
79 
80 static void
81 g_bioq_unlock(struct g_bioq *bq)
82 {
83 
84 	mtx_unlock(&bq->bio_queue_lock);
85 }
86 
87 #if 0
88 static void
89 g_bioq_destroy(struct g_bioq *bq)
90 {
91 
92 	mtx_destroy(&bq->bio_queue_lock);
93 }
94 #endif
95 
96 static void
97 g_bioq_init(struct g_bioq *bq)
98 {
99 
100 	TAILQ_INIT(&bq->bio_queue);
101 	mtx_init(&bq->bio_queue_lock, "bio queue", NULL, MTX_DEF);
102 }
103 
104 static struct bio *
105 g_bioq_first(struct g_bioq *bq)
106 {
107 	struct bio *bp;
108 
109 	bp = TAILQ_FIRST(&bq->bio_queue);
110 	if (bp != NULL) {
111 		KASSERT((bp->bio_flags & BIO_ONQUEUE),
112 		    ("Bio not on queue bp=%p target %p", bp, bq));
113 		bp->bio_flags &= ~BIO_ONQUEUE;
114 		TAILQ_REMOVE(&bq->bio_queue, bp, bio_queue);
115 		bq->bio_queue_length--;
116 	}
117 	return (bp);
118 }
119 
120 struct bio *
121 g_new_bio(void)
122 {
123 	struct bio *bp;
124 
125 	bp = uma_zalloc(biozone, M_NOWAIT | M_ZERO);
126 #ifdef KTR
127 	if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
128 		struct stack st;
129 
130 		CTR1(KTR_GEOM, "g_new_bio(): %p", bp);
131 		stack_save(&st);
132 		CTRSTACK(KTR_GEOM, &st, 3, 0);
133 	}
134 #endif
135 	return (bp);
136 }
137 
138 struct bio *
139 g_alloc_bio(void)
140 {
141 	struct bio *bp;
142 
143 	bp = uma_zalloc(biozone, M_WAITOK | M_ZERO);
144 #ifdef KTR
145 	if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
146 		struct stack st;
147 
148 		CTR1(KTR_GEOM, "g_alloc_bio(): %p", bp);
149 		stack_save(&st);
150 		CTRSTACK(KTR_GEOM, &st, 3, 0);
151 	}
152 #endif
153 	return (bp);
154 }
155 
156 void
157 g_destroy_bio(struct bio *bp)
158 {
159 #ifdef KTR
160 	if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
161 		struct stack st;
162 
163 		CTR1(KTR_GEOM, "g_destroy_bio(): %p", bp);
164 		stack_save(&st);
165 		CTRSTACK(KTR_GEOM, &st, 3, 0);
166 	}
167 #endif
168 	uma_zfree(biozone, bp);
169 }
170 
171 struct bio *
172 g_clone_bio(struct bio *bp)
173 {
174 	struct bio *bp2;
175 
176 	bp2 = uma_zalloc(biozone, M_NOWAIT | M_ZERO);
177 	if (bp2 != NULL) {
178 		bp2->bio_parent = bp;
179 		bp2->bio_cmd = bp->bio_cmd;
180 		bp2->bio_length = bp->bio_length;
181 		bp2->bio_offset = bp->bio_offset;
182 		bp2->bio_data = bp->bio_data;
183 		bp2->bio_attribute = bp->bio_attribute;
184 		/* Inherit classification info from the parent */
185 		bp2->bio_classifier1 = bp->bio_classifier1;
186 		bp2->bio_classifier2 = bp->bio_classifier2;
187 		bp->bio_children++;
188 	}
189 #ifdef KTR
190 	if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
191 		struct stack st;
192 
193 		CTR2(KTR_GEOM, "g_clone_bio(%p): %p", bp, bp2);
194 		stack_save(&st);
195 		CTRSTACK(KTR_GEOM, &st, 3, 0);
196 	}
197 #endif
198 	return(bp2);
199 }
200 
201 struct bio *
202 g_duplicate_bio(struct bio *bp)
203 {
204 	struct bio *bp2;
205 
206 	bp2 = uma_zalloc(biozone, M_WAITOK | M_ZERO);
207 	bp2->bio_parent = bp;
208 	bp2->bio_cmd = bp->bio_cmd;
209 	bp2->bio_length = bp->bio_length;
210 	bp2->bio_offset = bp->bio_offset;
211 	bp2->bio_data = bp->bio_data;
212 	bp2->bio_attribute = bp->bio_attribute;
213 	bp->bio_children++;
214 #ifdef KTR
215 	if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
216 		struct stack st;
217 
218 		CTR2(KTR_GEOM, "g_duplicate_bio(%p): %p", bp, bp2);
219 		stack_save(&st);
220 		CTRSTACK(KTR_GEOM, &st, 3, 0);
221 	}
222 #endif
223 	return(bp2);
224 }
225 
226 void
227 g_io_init()
228 {
229 
230 	g_bioq_init(&g_bio_run_down);
231 	g_bioq_init(&g_bio_run_up);
232 	g_bioq_init(&g_bio_run_task);
233 	biozone = uma_zcreate("g_bio", sizeof (struct bio),
234 	    NULL, NULL,
235 	    NULL, NULL,
236 	    0, 0);
237 }
238 
239 int
240 g_io_getattr(const char *attr, struct g_consumer *cp, int *len, void *ptr)
241 {
242 	struct bio *bp;
243 	int error;
244 
245 	g_trace(G_T_BIO, "bio_getattr(%s)", attr);
246 	bp = g_alloc_bio();
247 	bp->bio_cmd = BIO_GETATTR;
248 	bp->bio_done = NULL;
249 	bp->bio_attribute = attr;
250 	bp->bio_length = *len;
251 	bp->bio_data = ptr;
252 	g_io_request(bp, cp);
253 	error = biowait(bp, "ggetattr");
254 	*len = bp->bio_completed;
255 	g_destroy_bio(bp);
256 	return (error);
257 }
258 
259 int
260 g_io_flush(struct g_consumer *cp)
261 {
262 	struct bio *bp;
263 	int error;
264 
265 	g_trace(G_T_BIO, "bio_flush(%s)", cp->provider->name);
266 	bp = g_alloc_bio();
267 	bp->bio_cmd = BIO_FLUSH;
268 	bp->bio_flags |= BIO_ORDERED;
269 	bp->bio_done = NULL;
270 	bp->bio_attribute = NULL;
271 	bp->bio_offset = cp->provider->mediasize;
272 	bp->bio_length = 0;
273 	bp->bio_data = NULL;
274 	g_io_request(bp, cp);
275 	error = biowait(bp, "gflush");
276 	g_destroy_bio(bp);
277 	return (error);
278 }
279 
280 static int
281 g_io_check(struct bio *bp)
282 {
283 	struct g_consumer *cp;
284 	struct g_provider *pp;
285 
286 	cp = bp->bio_from;
287 	pp = bp->bio_to;
288 
289 	/* Fail if access counters dont allow the operation */
290 	switch(bp->bio_cmd) {
291 	case BIO_READ:
292 	case BIO_GETATTR:
293 		if (cp->acr == 0)
294 			return (EPERM);
295 		break;
296 	case BIO_WRITE:
297 	case BIO_DELETE:
298 	case BIO_FLUSH:
299 		if (cp->acw == 0)
300 			return (EPERM);
301 		break;
302 	default:
303 		return (EPERM);
304 	}
305 	/* if provider is marked for error, don't disturb. */
306 	if (pp->error)
307 		return (pp->error);
308 
309 	switch(bp->bio_cmd) {
310 	case BIO_READ:
311 	case BIO_WRITE:
312 	case BIO_DELETE:
313 		/* Zero sectorsize or mediasize is probably a lack of media. */
314 		if (pp->sectorsize == 0 || pp->mediasize == 0)
315 			return (ENXIO);
316 		/* Reject I/O not on sector boundary */
317 		if (bp->bio_offset % pp->sectorsize)
318 			return (EINVAL);
319 		/* Reject I/O not integral sector long */
320 		if (bp->bio_length % pp->sectorsize)
321 			return (EINVAL);
322 		/* Reject requests before or past the end of media. */
323 		if (bp->bio_offset < 0)
324 			return (EIO);
325 		if (bp->bio_offset > pp->mediasize)
326 			return (EIO);
327 		break;
328 	default:
329 		break;
330 	}
331 	return (0);
332 }
333 
334 /*
335  * bio classification support.
336  *
337  * g_register_classifier() and g_unregister_classifier()
338  * are used to add/remove a classifier from the list.
339  * The list is protected using the g_bio_run_down lock,
340  * because the classifiers are called in this path.
341  *
342  * g_io_request() passes bio's that are not already classified
343  * (i.e. those with bio_classifier1 == NULL) to g_run_classifiers().
344  * Classifiers can store their result in the two fields
345  * bio_classifier1 and bio_classifier2.
346  * A classifier that updates one of the fields should
347  * return a non-zero value.
348  * If no classifier updates the field, g_run_classifiers() sets
349  * bio_classifier1 = BIO_NOTCLASSIFIED to avoid further calls.
350  */
351 
352 int
353 g_register_classifier(struct g_classifier_hook *hook)
354 {
355 
356 	g_bioq_lock(&g_bio_run_down);
357 	TAILQ_INSERT_TAIL(&g_classifier_tailq, hook, link);
358 	g_bioq_unlock(&g_bio_run_down);
359 
360 	return (0);
361 }
362 
363 void
364 g_unregister_classifier(struct g_classifier_hook *hook)
365 {
366 	struct g_classifier_hook *entry;
367 
368 	g_bioq_lock(&g_bio_run_down);
369 	TAILQ_FOREACH(entry, &g_classifier_tailq, link) {
370 		if (entry == hook) {
371 			TAILQ_REMOVE(&g_classifier_tailq, hook, link);
372 			break;
373 		}
374 	}
375 	g_bioq_unlock(&g_bio_run_down);
376 }
377 
378 static void
379 g_run_classifiers(struct bio *bp)
380 {
381 	struct g_classifier_hook *hook;
382 	int classified = 0;
383 
384 	TAILQ_FOREACH(hook, &g_classifier_tailq, link)
385 		classified |= hook->func(hook->arg, bp);
386 
387 	if (!classified)
388 		bp->bio_classifier1 = BIO_NOTCLASSIFIED;
389 }
390 
391 void
392 g_io_request(struct bio *bp, struct g_consumer *cp)
393 {
394 	struct g_provider *pp;
395 	int first;
396 
397 	KASSERT(cp != NULL, ("NULL cp in g_io_request"));
398 	KASSERT(bp != NULL, ("NULL bp in g_io_request"));
399 	pp = cp->provider;
400 	KASSERT(pp != NULL, ("consumer not attached in g_io_request"));
401 #ifdef DIAGNOSTIC
402 	KASSERT(bp->bio_driver1 == NULL,
403 	    ("bio_driver1 used by the consumer (geom %s)", cp->geom->name));
404 	KASSERT(bp->bio_driver2 == NULL,
405 	    ("bio_driver2 used by the consumer (geom %s)", cp->geom->name));
406 	KASSERT(bp->bio_pflags == 0,
407 	    ("bio_pflags used by the consumer (geom %s)", cp->geom->name));
408 	/*
409 	 * Remember consumer's private fields, so we can detect if they were
410 	 * modified by the provider.
411 	 */
412 	bp->_bio_caller1 = bp->bio_caller1;
413 	bp->_bio_caller2 = bp->bio_caller2;
414 	bp->_bio_cflags = bp->bio_cflags;
415 #endif
416 
417 	if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_GETATTR)) {
418 		KASSERT(bp->bio_data != NULL,
419 		    ("NULL bp->data in g_io_request(cmd=%hhu)", bp->bio_cmd));
420 	}
421 	if (bp->bio_cmd & (BIO_DELETE|BIO_FLUSH)) {
422 		KASSERT(bp->bio_data == NULL,
423 		    ("non-NULL bp->data in g_io_request(cmd=%hhu)",
424 		    bp->bio_cmd));
425 	}
426 	if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) {
427 		KASSERT(bp->bio_offset % cp->provider->sectorsize == 0,
428 		    ("wrong offset %jd for sectorsize %u",
429 		    bp->bio_offset, cp->provider->sectorsize));
430 		KASSERT(bp->bio_length % cp->provider->sectorsize == 0,
431 		    ("wrong length %jd for sectorsize %u",
432 		    bp->bio_length, cp->provider->sectorsize));
433 	}
434 
435 	g_trace(G_T_BIO, "bio_request(%p) from %p(%s) to %p(%s) cmd %d",
436 	    bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd);
437 
438 	bp->bio_from = cp;
439 	bp->bio_to = pp;
440 	bp->bio_error = 0;
441 	bp->bio_completed = 0;
442 
443 	KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
444 	    ("Bio already on queue bp=%p", bp));
445 	bp->bio_flags |= BIO_ONQUEUE;
446 
447 	if (g_collectstats)
448 		binuptime(&bp->bio_t0);
449 	else
450 		getbinuptime(&bp->bio_t0);
451 
452 	/*
453 	 * The statistics collection is lockless, as such, but we
454 	 * can not update one instance of the statistics from more
455 	 * than one thread at a time, so grab the lock first.
456 	 *
457 	 * We also use the lock to protect the list of classifiers.
458 	 */
459 	g_bioq_lock(&g_bio_run_down);
460 
461 	if (!TAILQ_EMPTY(&g_classifier_tailq) && !bp->bio_classifier1)
462 		g_run_classifiers(bp);
463 
464 	if (g_collectstats & 1)
465 		devstat_start_transaction(pp->stat, &bp->bio_t0);
466 	if (g_collectstats & 2)
467 		devstat_start_transaction(cp->stat, &bp->bio_t0);
468 
469 	pp->nstart++;
470 	cp->nstart++;
471 	first = TAILQ_EMPTY(&g_bio_run_down.bio_queue);
472 	TAILQ_INSERT_TAIL(&g_bio_run_down.bio_queue, bp, bio_queue);
473 	g_bio_run_down.bio_queue_length++;
474 	g_bioq_unlock(&g_bio_run_down);
475 
476 	/* Pass it on down. */
477 	if (first)
478 		wakeup(&g_wait_down);
479 }
480 
481 void
482 g_io_deliver(struct bio *bp, int error)
483 {
484 	struct g_consumer *cp;
485 	struct g_provider *pp;
486 	int first;
487 
488 	KASSERT(bp != NULL, ("NULL bp in g_io_deliver"));
489 	pp = bp->bio_to;
490 	KASSERT(pp != NULL, ("NULL bio_to in g_io_deliver"));
491 	cp = bp->bio_from;
492 	if (cp == NULL) {
493 		bp->bio_error = error;
494 		bp->bio_done(bp);
495 		return;
496 	}
497 	KASSERT(cp != NULL, ("NULL bio_from in g_io_deliver"));
498 	KASSERT(cp->geom != NULL, ("NULL bio_from->geom in g_io_deliver"));
499 #ifdef DIAGNOSTIC
500 	/*
501 	 * Some classes - GJournal in particular - can modify bio's
502 	 * private fields while the bio is in transit; G_GEOM_VOLATILE_BIO
503 	 * flag means it's an expected behaviour for that particular geom.
504 	 */
505 	if ((cp->geom->flags & G_GEOM_VOLATILE_BIO) == 0) {
506 		KASSERT(bp->bio_caller1 == bp->_bio_caller1,
507 		    ("bio_caller1 used by the provider %s", pp->name));
508 		KASSERT(bp->bio_caller2 == bp->_bio_caller2,
509 		    ("bio_caller2 used by the provider %s", pp->name));
510 		KASSERT(bp->bio_cflags == bp->_bio_cflags,
511 		    ("bio_cflags used by the provider %s", pp->name));
512 	}
513 #endif
514 	KASSERT(bp->bio_completed >= 0, ("bio_completed can't be less than 0"));
515 	KASSERT(bp->bio_completed <= bp->bio_length,
516 	    ("bio_completed can't be greater than bio_length"));
517 
518 	g_trace(G_T_BIO,
519 "g_io_deliver(%p) from %p(%s) to %p(%s) cmd %d error %d off %jd len %jd",
520 	    bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd, error,
521 	    (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
522 
523 	KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
524 	    ("Bio already on queue bp=%p", bp));
525 
526 	/*
527 	 * XXX: next two doesn't belong here
528 	 */
529 	bp->bio_bcount = bp->bio_length;
530 	bp->bio_resid = bp->bio_bcount - bp->bio_completed;
531 
532 	/*
533 	 * The statistics collection is lockless, as such, but we
534 	 * can not update one instance of the statistics from more
535 	 * than one thread at a time, so grab the lock first.
536 	 */
537 	g_bioq_lock(&g_bio_run_up);
538 	if (g_collectstats & 1)
539 		devstat_end_transaction_bio(pp->stat, bp);
540 	if (g_collectstats & 2)
541 		devstat_end_transaction_bio(cp->stat, bp);
542 
543 	cp->nend++;
544 	pp->nend++;
545 	if (error != ENOMEM) {
546 		bp->bio_error = error;
547 		first = TAILQ_EMPTY(&g_bio_run_up.bio_queue);
548 		TAILQ_INSERT_TAIL(&g_bio_run_up.bio_queue, bp, bio_queue);
549 		bp->bio_flags |= BIO_ONQUEUE;
550 		g_bio_run_up.bio_queue_length++;
551 		g_bioq_unlock(&g_bio_run_up);
552 		if (first)
553 			wakeup(&g_wait_up);
554 		return;
555 	}
556 	g_bioq_unlock(&g_bio_run_up);
557 
558 	if (bootverbose)
559 		printf("ENOMEM %p on %p(%s)\n", bp, pp, pp->name);
560 	bp->bio_children = 0;
561 	bp->bio_inbed = 0;
562 	g_io_request(bp, cp);
563 	pace++;
564 	return;
565 }
566 
567 void
568 g_io_schedule_down(struct thread *tp __unused)
569 {
570 	struct bio *bp;
571 	off_t excess;
572 	int error;
573 
574 	for(;;) {
575 		g_bioq_lock(&g_bio_run_down);
576 		bp = g_bioq_first(&g_bio_run_down);
577 		if (bp == NULL) {
578 			CTR0(KTR_GEOM, "g_down going to sleep");
579 			msleep(&g_wait_down, &g_bio_run_down.bio_queue_lock,
580 			    PRIBIO | PDROP, "-", 0);
581 			continue;
582 		}
583 		CTR0(KTR_GEOM, "g_down has work to do");
584 		g_bioq_unlock(&g_bio_run_down);
585 		if (pace > 0) {
586 			CTR1(KTR_GEOM, "g_down pacing self (pace %d)", pace);
587 			pause("g_down", hz/10);
588 			pace--;
589 		}
590 		error = g_io_check(bp);
591 		if (error) {
592 			CTR3(KTR_GEOM, "g_down g_io_check on bp %p provider "
593 			    "%s returned %d", bp, bp->bio_to->name, error);
594 			g_io_deliver(bp, error);
595 			continue;
596 		}
597 		CTR2(KTR_GEOM, "g_down processing bp %p provider %s", bp,
598 		    bp->bio_to->name);
599 		switch (bp->bio_cmd) {
600 		case BIO_READ:
601 		case BIO_WRITE:
602 		case BIO_DELETE:
603 			/* Truncate requests to the end of providers media. */
604 			/*
605 			 * XXX: What if we truncate because of offset being
606 			 * bad, not length?
607 			 */
608 			excess = bp->bio_offset + bp->bio_length;
609 			if (excess > bp->bio_to->mediasize) {
610 				excess -= bp->bio_to->mediasize;
611 				bp->bio_length -= excess;
612 				if (excess > 0)
613 					CTR3(KTR_GEOM, "g_down truncated bio "
614 					    "%p provider %s by %d", bp,
615 					    bp->bio_to->name, excess);
616 			}
617 			/* Deliver zero length transfers right here. */
618 			if (bp->bio_length == 0) {
619 				g_io_deliver(bp, 0);
620 				CTR2(KTR_GEOM, "g_down terminated 0-length "
621 				    "bp %p provider %s", bp, bp->bio_to->name);
622 				continue;
623 			}
624 			break;
625 		default:
626 			break;
627 		}
628 		THREAD_NO_SLEEPING();
629 		CTR4(KTR_GEOM, "g_down starting bp %p provider %s off %ld "
630 		    "len %ld", bp, bp->bio_to->name, bp->bio_offset,
631 		    bp->bio_length);
632 		bp->bio_to->geom->start(bp);
633 		THREAD_SLEEPING_OK();
634 	}
635 }
636 
637 void
638 bio_taskqueue(struct bio *bp, bio_task_t *func, void *arg)
639 {
640 	bp->bio_task = func;
641 	bp->bio_task_arg = arg;
642 	/*
643 	 * The taskqueue is actually just a second queue off the "up"
644 	 * queue, so we use the same lock.
645 	 */
646 	g_bioq_lock(&g_bio_run_up);
647 	KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
648 	    ("Bio already on queue bp=%p target taskq", bp));
649 	bp->bio_flags |= BIO_ONQUEUE;
650 	TAILQ_INSERT_TAIL(&g_bio_run_task.bio_queue, bp, bio_queue);
651 	g_bio_run_task.bio_queue_length++;
652 	wakeup(&g_wait_up);
653 	g_bioq_unlock(&g_bio_run_up);
654 }
655 
656 
657 void
658 g_io_schedule_up(struct thread *tp __unused)
659 {
660 	struct bio *bp;
661 	for(;;) {
662 		g_bioq_lock(&g_bio_run_up);
663 		bp = g_bioq_first(&g_bio_run_task);
664 		if (bp != NULL) {
665 			g_bioq_unlock(&g_bio_run_up);
666 			THREAD_NO_SLEEPING();
667 			CTR1(KTR_GEOM, "g_up processing task bp %p", bp);
668 			bp->bio_task(bp->bio_task_arg);
669 			THREAD_SLEEPING_OK();
670 			continue;
671 		}
672 		bp = g_bioq_first(&g_bio_run_up);
673 		if (bp != NULL) {
674 			g_bioq_unlock(&g_bio_run_up);
675 			THREAD_NO_SLEEPING();
676 			CTR4(KTR_GEOM, "g_up biodone bp %p provider %s off "
677 			    "%jd len %ld", bp, bp->bio_to->name,
678 			    bp->bio_offset, bp->bio_length);
679 			biodone(bp);
680 			THREAD_SLEEPING_OK();
681 			continue;
682 		}
683 		CTR0(KTR_GEOM, "g_up going to sleep");
684 		msleep(&g_wait_up, &g_bio_run_up.bio_queue_lock,
685 		    PRIBIO | PDROP, "-", 0);
686 	}
687 }
688 
689 void *
690 g_read_data(struct g_consumer *cp, off_t offset, off_t length, int *error)
691 {
692 	struct bio *bp;
693 	void *ptr;
694 	int errorc;
695 
696 	KASSERT(length > 0 && length >= cp->provider->sectorsize &&
697 	    length <= MAXPHYS, ("g_read_data(): invalid length %jd",
698 	    (intmax_t)length));
699 
700 	bp = g_alloc_bio();
701 	bp->bio_cmd = BIO_READ;
702 	bp->bio_done = NULL;
703 	bp->bio_offset = offset;
704 	bp->bio_length = length;
705 	ptr = g_malloc(length, M_WAITOK);
706 	bp->bio_data = ptr;
707 	g_io_request(bp, cp);
708 	errorc = biowait(bp, "gread");
709 	if (error != NULL)
710 		*error = errorc;
711 	g_destroy_bio(bp);
712 	if (errorc) {
713 		g_free(ptr);
714 		ptr = NULL;
715 	}
716 	return (ptr);
717 }
718 
719 int
720 g_write_data(struct g_consumer *cp, off_t offset, void *ptr, off_t length)
721 {
722 	struct bio *bp;
723 	int error;
724 
725 	KASSERT(length > 0 && length >= cp->provider->sectorsize &&
726 	    length <= MAXPHYS, ("g_write_data(): invalid length %jd",
727 	    (intmax_t)length));
728 
729 	bp = g_alloc_bio();
730 	bp->bio_cmd = BIO_WRITE;
731 	bp->bio_done = NULL;
732 	bp->bio_offset = offset;
733 	bp->bio_length = length;
734 	bp->bio_data = ptr;
735 	g_io_request(bp, cp);
736 	error = biowait(bp, "gwrite");
737 	g_destroy_bio(bp);
738 	return (error);
739 }
740 
741 int
742 g_delete_data(struct g_consumer *cp, off_t offset, off_t length)
743 {
744 	struct bio *bp;
745 	int error;
746 
747 	KASSERT(length > 0 && length >= cp->provider->sectorsize,
748 	    ("g_delete_data(): invalid length %jd", (intmax_t)length));
749 
750 	bp = g_alloc_bio();
751 	bp->bio_cmd = BIO_DELETE;
752 	bp->bio_done = NULL;
753 	bp->bio_offset = offset;
754 	bp->bio_length = length;
755 	bp->bio_data = NULL;
756 	g_io_request(bp, cp);
757 	error = biowait(bp, "gdelete");
758 	g_destroy_bio(bp);
759 	return (error);
760 }
761 
762 void
763 g_print_bio(struct bio *bp)
764 {
765 	const char *pname, *cmd = NULL;
766 
767 	if (bp->bio_to != NULL)
768 		pname = bp->bio_to->name;
769 	else
770 		pname = "[unknown]";
771 
772 	switch (bp->bio_cmd) {
773 	case BIO_GETATTR:
774 		cmd = "GETATTR";
775 		printf("%s[%s(attr=%s)]", pname, cmd, bp->bio_attribute);
776 		return;
777 	case BIO_FLUSH:
778 		cmd = "FLUSH";
779 		printf("%s[%s]", pname, cmd);
780 		return;
781 	case BIO_READ:
782 		cmd = "READ";
783 		break;
784 	case BIO_WRITE:
785 		cmd = "WRITE";
786 		break;
787 	case BIO_DELETE:
788 		cmd = "DELETE";
789 		break;
790 	default:
791 		cmd = "UNKNOWN";
792 		printf("%s[%s()]", pname, cmd);
793 		return;
794 	}
795 	printf("%s[%s(offset=%jd, length=%jd)]", pname, cmd,
796 	    (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
797 }
798