2 * Copyright (c) 2002 Poul-Henning Kamp
3 * Copyright (c) 2002 Networks Associates Technology, Inc.
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.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
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
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
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
46 #include <sys/stack.h>
48 #include <sys/errno.h>
49 #include <geom/geom.h>
50 #include <geom/geom_int.h>
51 #include <sys/devicestat.h>
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;
60 static uma_zone_t biozone;
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.
68 static TAILQ_HEAD(g_classifier_tailq, g_classifier_hook)
69 g_classifier_tailq = TAILQ_HEAD_INITIALIZER(g_classifier_tailq);
71 #include <machine/atomic.h>
74 g_bioq_lock(struct g_bioq *bq)
77 mtx_lock(&bq->bio_queue_lock);
81 g_bioq_unlock(struct g_bioq *bq)
84 mtx_unlock(&bq->bio_queue_lock);
89 g_bioq_destroy(struct g_bioq *bq)
92 mtx_destroy(&bq->bio_queue_lock);
97 g_bioq_init(struct g_bioq *bq)
100 TAILQ_INIT(&bq->bio_queue);
101 mtx_init(&bq->bio_queue_lock, "bio queue", NULL, MTX_DEF);
105 g_bioq_first(struct g_bioq *bq)
109 bp = TAILQ_FIRST(&bq->bio_queue);
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--;
125 bp = uma_zalloc(biozone, M_NOWAIT | M_ZERO);
127 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
130 CTR1(KTR_GEOM, "g_new_bio(): %p", bp);
132 CTRSTACK(KTR_GEOM, &st, 3, 0);
143 bp = uma_zalloc(biozone, M_WAITOK | M_ZERO);
145 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
148 CTR1(KTR_GEOM, "g_alloc_bio(): %p", bp);
150 CTRSTACK(KTR_GEOM, &st, 3, 0);
157 g_destroy_bio(struct bio *bp)
160 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
163 CTR1(KTR_GEOM, "g_destroy_bio(): %p", bp);
165 CTRSTACK(KTR_GEOM, &st, 3, 0);
168 uma_zfree(biozone, bp);
172 g_clone_bio(struct bio *bp)
176 bp2 = uma_zalloc(biozone, M_NOWAIT | M_ZERO);
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;
190 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
193 CTR2(KTR_GEOM, "g_clone_bio(%p): %p", bp, bp2);
195 CTRSTACK(KTR_GEOM, &st, 3, 0);
202 g_duplicate_bio(struct bio *bp)
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;
215 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
218 CTR2(KTR_GEOM, "g_duplicate_bio(%p): %p", bp, bp2);
220 CTRSTACK(KTR_GEOM, &st, 3, 0);
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),
240 g_io_getattr(const char *attr, struct g_consumer *cp, int *len, void *ptr)
245 g_trace(G_T_BIO, "bio_getattr(%s)", attr);
247 bp->bio_cmd = BIO_GETATTR;
249 bp->bio_attribute = attr;
250 bp->bio_length = *len;
252 g_io_request(bp, cp);
253 error = biowait(bp, "ggetattr");
254 *len = bp->bio_completed;
260 g_io_flush(struct g_consumer *cp)
265 g_trace(G_T_BIO, "bio_flush(%s)", cp->provider->name);
267 bp->bio_cmd = BIO_FLUSH;
269 bp->bio_attribute = NULL;
270 bp->bio_offset = cp->provider->mediasize;
273 g_io_request(bp, cp);
274 error = biowait(bp, "gflush");
280 g_io_check(struct bio *bp)
282 struct g_consumer *cp;
283 struct g_provider *pp;
288 /* Fail if access counters dont allow the operation */
289 switch(bp->bio_cmd) {
304 /* if provider is marked for error, don't disturb. */
308 switch(bp->bio_cmd) {
312 /* Zero sectorsize or mediasize is probably a lack of media. */
313 if (pp->sectorsize == 0 || pp->mediasize == 0)
315 /* Reject I/O not on sector boundary */
316 if (bp->bio_offset % pp->sectorsize)
318 /* Reject I/O not integral sector long */
319 if (bp->bio_length % pp->sectorsize)
321 /* Reject requests before or past the end of media. */
322 if (bp->bio_offset < 0)
324 if (bp->bio_offset > pp->mediasize)
334 * bio classification support.
336 * g_register_classifier() and g_unregister_classifier()
337 * are used to add/remove a classifier from the list.
338 * The list is protected using the g_bio_run_down lock,
339 * because the classifiers are called in this path.
341 * g_io_request() passes bio's that are not already classified
342 * (i.e. those with bio_classifier1 == NULL) to g_run_classifiers().
343 * Classifiers can store their result in the two fields
344 * bio_classifier1 and bio_classifier2.
345 * A classifier that updates one of the fields should
346 * return a non-zero value.
347 * If no classifier updates the field, g_run_classifiers() sets
348 * bio_classifier1 = BIO_NOTCLASSIFIED to avoid further calls.
352 g_register_classifier(struct g_classifier_hook *hook)
355 g_bioq_lock(&g_bio_run_down);
356 TAILQ_INSERT_TAIL(&g_classifier_tailq, hook, link);
357 g_bioq_unlock(&g_bio_run_down);
363 g_unregister_classifier(struct g_classifier_hook *hook)
365 struct g_classifier_hook *entry;
367 g_bioq_lock(&g_bio_run_down);
368 TAILQ_FOREACH(entry, &g_classifier_tailq, link) {
370 TAILQ_REMOVE(&g_classifier_tailq, hook, link);
374 g_bioq_unlock(&g_bio_run_down);
378 g_run_classifiers(struct bio *bp)
380 struct g_classifier_hook *hook;
383 TAILQ_FOREACH(hook, &g_classifier_tailq, link)
384 classified |= hook->func(hook->arg, bp);
387 bp->bio_classifier1 = BIO_NOTCLASSIFIED;
391 g_io_request(struct bio *bp, struct g_consumer *cp)
393 struct g_provider *pp;
396 KASSERT(cp != NULL, ("NULL cp in g_io_request"));
397 KASSERT(bp != NULL, ("NULL bp in g_io_request"));
399 KASSERT(pp != NULL, ("consumer not attached in g_io_request"));
401 KASSERT(bp->bio_driver1 == NULL,
402 ("bio_driver1 used by the consumer (geom %s)", cp->geom->name));
403 KASSERT(bp->bio_driver2 == NULL,
404 ("bio_driver2 used by the consumer (geom %s)", cp->geom->name));
405 KASSERT(bp->bio_pflags == 0,
406 ("bio_pflags used by the consumer (geom %s)", cp->geom->name));
408 * Remember consumer's private fields, so we can detect if they were
409 * modified by the provider.
411 bp->_bio_caller1 = bp->bio_caller1;
412 bp->_bio_caller2 = bp->bio_caller2;
413 bp->_bio_cflags = bp->bio_cflags;
416 if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_GETATTR)) {
417 KASSERT(bp->bio_data != NULL,
418 ("NULL bp->data in g_io_request(cmd=%hhu)", bp->bio_cmd));
420 if (bp->bio_cmd & (BIO_DELETE|BIO_FLUSH)) {
421 KASSERT(bp->bio_data == NULL,
422 ("non-NULL bp->data in g_io_request(cmd=%hhu)",
425 if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) {
426 KASSERT(bp->bio_offset % cp->provider->sectorsize == 0,
427 ("wrong offset %jd for sectorsize %u",
428 bp->bio_offset, cp->provider->sectorsize));
429 KASSERT(bp->bio_length % cp->provider->sectorsize == 0,
430 ("wrong length %jd for sectorsize %u",
431 bp->bio_length, cp->provider->sectorsize));
434 g_trace(G_T_BIO, "bio_request(%p) from %p(%s) to %p(%s) cmd %d",
435 bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd);
440 bp->bio_completed = 0;
442 KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
443 ("Bio already on queue bp=%p", bp));
444 bp->bio_flags |= BIO_ONQUEUE;
446 binuptime(&bp->bio_t0);
449 * The statistics collection is lockless, as such, but we
450 * can not update one instance of the statistics from more
451 * than one thread at a time, so grab the lock first.
453 * We also use the lock to protect the list of classifiers.
455 g_bioq_lock(&g_bio_run_down);
457 if (!TAILQ_EMPTY(&g_classifier_tailq) && !bp->bio_classifier1)
458 g_run_classifiers(bp);
460 if (g_collectstats & 1)
461 devstat_start_transaction(pp->stat, &bp->bio_t0);
462 if (g_collectstats & 2)
463 devstat_start_transaction(cp->stat, &bp->bio_t0);
467 first = TAILQ_EMPTY(&g_bio_run_down.bio_queue);
468 TAILQ_INSERT_TAIL(&g_bio_run_down.bio_queue, bp, bio_queue);
469 g_bio_run_down.bio_queue_length++;
470 g_bioq_unlock(&g_bio_run_down);
472 /* Pass it on down. */
474 wakeup(&g_wait_down);
478 g_io_deliver(struct bio *bp, int error)
480 struct g_consumer *cp;
481 struct g_provider *pp;
484 KASSERT(bp != NULL, ("NULL bp in g_io_deliver"));
486 KASSERT(pp != NULL, ("NULL bio_to in g_io_deliver"));
489 bp->bio_error = error;
493 KASSERT(cp != NULL, ("NULL bio_from in g_io_deliver"));
494 KASSERT(cp->geom != NULL, ("NULL bio_from->geom in g_io_deliver"));
497 * Some classes - GJournal in particular - can modify bio's
498 * private fields while the bio is in transit; G_GEOM_VOLATILE_BIO
499 * flag means it's an expected behaviour for that particular geom.
501 if ((cp->geom->flags & G_GEOM_VOLATILE_BIO) == 0) {
502 KASSERT(bp->bio_caller1 == bp->_bio_caller1,
503 ("bio_caller1 used by the provider %s", pp->name));
504 KASSERT(bp->bio_caller2 == bp->_bio_caller2,
505 ("bio_caller2 used by the provider %s", pp->name));
506 KASSERT(bp->bio_cflags == bp->_bio_cflags,
507 ("bio_cflags used by the provider %s", pp->name));
510 KASSERT(bp->bio_completed >= 0, ("bio_completed can't be less than 0"));
511 KASSERT(bp->bio_completed <= bp->bio_length,
512 ("bio_completed can't be greater than bio_length"));
515 "g_io_deliver(%p) from %p(%s) to %p(%s) cmd %d error %d off %jd len %jd",
516 bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd, error,
517 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
519 KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
520 ("Bio already on queue bp=%p", bp));
523 * XXX: next two doesn't belong here
525 bp->bio_bcount = bp->bio_length;
526 bp->bio_resid = bp->bio_bcount - bp->bio_completed;
529 * The statistics collection is lockless, as such, but we
530 * can not update one instance of the statistics from more
531 * than one thread at a time, so grab the lock first.
533 g_bioq_lock(&g_bio_run_up);
534 if (g_collectstats & 1)
535 devstat_end_transaction_bio(pp->stat, bp);
536 if (g_collectstats & 2)
537 devstat_end_transaction_bio(cp->stat, bp);
541 if (error != ENOMEM) {
542 bp->bio_error = error;
543 first = TAILQ_EMPTY(&g_bio_run_up.bio_queue);
544 TAILQ_INSERT_TAIL(&g_bio_run_up.bio_queue, bp, bio_queue);
545 bp->bio_flags |= BIO_ONQUEUE;
546 g_bio_run_up.bio_queue_length++;
547 g_bioq_unlock(&g_bio_run_up);
552 g_bioq_unlock(&g_bio_run_up);
555 printf("ENOMEM %p on %p(%s)\n", bp, pp, pp->name);
556 bp->bio_children = 0;
558 g_io_request(bp, cp);
564 g_io_schedule_down(struct thread *tp __unused)
571 g_bioq_lock(&g_bio_run_down);
572 bp = g_bioq_first(&g_bio_run_down);
574 CTR0(KTR_GEOM, "g_down going to sleep");
575 msleep(&g_wait_down, &g_bio_run_down.bio_queue_lock,
576 PRIBIO | PDROP, "-", 0);
579 CTR0(KTR_GEOM, "g_down has work to do");
580 g_bioq_unlock(&g_bio_run_down);
582 CTR1(KTR_GEOM, "g_down pacing self (pace %d)", pace);
583 pause("g_down", hz/10);
586 error = g_io_check(bp);
588 CTR3(KTR_GEOM, "g_down g_io_check on bp %p provider "
589 "%s returned %d", bp, bp->bio_to->name, error);
590 g_io_deliver(bp, error);
593 CTR2(KTR_GEOM, "g_down processing bp %p provider %s", bp,
595 switch (bp->bio_cmd) {
599 /* Truncate requests to the end of providers media. */
601 * XXX: What if we truncate because of offset being
604 excess = bp->bio_offset + bp->bio_length;
605 if (excess > bp->bio_to->mediasize) {
606 excess -= bp->bio_to->mediasize;
607 bp->bio_length -= excess;
609 CTR3(KTR_GEOM, "g_down truncated bio "
610 "%p provider %s by %d", bp,
611 bp->bio_to->name, excess);
613 /* Deliver zero length transfers right here. */
614 if (bp->bio_length == 0) {
616 CTR2(KTR_GEOM, "g_down terminated 0-length "
617 "bp %p provider %s", bp, bp->bio_to->name);
624 THREAD_NO_SLEEPING();
625 CTR4(KTR_GEOM, "g_down starting bp %p provider %s off %ld "
626 "len %ld", bp, bp->bio_to->name, bp->bio_offset,
628 bp->bio_to->geom->start(bp);
629 THREAD_SLEEPING_OK();
634 bio_taskqueue(struct bio *bp, bio_task_t *func, void *arg)
637 bp->bio_task_arg = arg;
639 * The taskqueue is actually just a second queue off the "up"
640 * queue, so we use the same lock.
642 g_bioq_lock(&g_bio_run_up);
643 KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
644 ("Bio already on queue bp=%p target taskq", bp));
645 bp->bio_flags |= BIO_ONQUEUE;
646 TAILQ_INSERT_TAIL(&g_bio_run_task.bio_queue, bp, bio_queue);
647 g_bio_run_task.bio_queue_length++;
649 g_bioq_unlock(&g_bio_run_up);
654 g_io_schedule_up(struct thread *tp __unused)
658 g_bioq_lock(&g_bio_run_up);
659 bp = g_bioq_first(&g_bio_run_task);
661 g_bioq_unlock(&g_bio_run_up);
662 THREAD_NO_SLEEPING();
663 CTR1(KTR_GEOM, "g_up processing task bp %p", bp);
664 bp->bio_task(bp->bio_task_arg);
665 THREAD_SLEEPING_OK();
668 bp = g_bioq_first(&g_bio_run_up);
670 g_bioq_unlock(&g_bio_run_up);
671 THREAD_NO_SLEEPING();
672 CTR4(KTR_GEOM, "g_up biodone bp %p provider %s off "
673 "%jd len %ld", bp, bp->bio_to->name,
674 bp->bio_offset, bp->bio_length);
676 THREAD_SLEEPING_OK();
679 CTR0(KTR_GEOM, "g_up going to sleep");
680 msleep(&g_wait_up, &g_bio_run_up.bio_queue_lock,
681 PRIBIO | PDROP, "-", 0);
686 g_read_data(struct g_consumer *cp, off_t offset, off_t length, int *error)
692 KASSERT(length > 0 && length >= cp->provider->sectorsize &&
693 length <= MAXPHYS, ("g_read_data(): invalid length %jd",
697 bp->bio_cmd = BIO_READ;
699 bp->bio_offset = offset;
700 bp->bio_length = length;
701 ptr = g_malloc(length, M_WAITOK);
703 g_io_request(bp, cp);
704 errorc = biowait(bp, "gread");
716 g_write_data(struct g_consumer *cp, off_t offset, void *ptr, off_t length)
721 KASSERT(length > 0 && length >= cp->provider->sectorsize &&
722 length <= MAXPHYS, ("g_write_data(): invalid length %jd",
726 bp->bio_cmd = BIO_WRITE;
728 bp->bio_offset = offset;
729 bp->bio_length = length;
731 g_io_request(bp, cp);
732 error = biowait(bp, "gwrite");
738 g_delete_data(struct g_consumer *cp, off_t offset, off_t length)
743 KASSERT(length > 0 && length >= cp->provider->sectorsize,
744 ("g_delete_data(): invalid length %jd", (intmax_t)length));
747 bp->bio_cmd = BIO_DELETE;
749 bp->bio_offset = offset;
750 bp->bio_length = length;
752 g_io_request(bp, cp);
753 error = biowait(bp, "gdelete");
759 g_print_bio(struct bio *bp)
761 const char *pname, *cmd = NULL;
763 if (bp->bio_to != NULL)
764 pname = bp->bio_to->name;
768 switch (bp->bio_cmd) {
771 printf("%s[%s(attr=%s)]", pname, cmd, bp->bio_attribute);
775 printf("%s[%s]", pname, cmd);
785 printf("%s[%s(offset=%jd, length=%jd)]", pname, cmd,
786 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
790 printf("%s[%s()]", pname, cmd);