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;
181 * BIO_ORDERED flag may be used by disk drivers to enforce
182 * ordering restrictions, so this flag needs to be cloned.
183 * Other bio flags are not suitable for cloning.
185 bp2->bio_flags = bp->bio_flags & BIO_ORDERED;
186 bp2->bio_length = bp->bio_length;
187 bp2->bio_offset = bp->bio_offset;
188 bp2->bio_data = bp->bio_data;
189 bp2->bio_attribute = bp->bio_attribute;
190 /* Inherit classification info from the parent */
191 bp2->bio_classifier1 = bp->bio_classifier1;
192 bp2->bio_classifier2 = bp->bio_classifier2;
196 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
199 CTR2(KTR_GEOM, "g_clone_bio(%p): %p", bp, bp2);
201 CTRSTACK(KTR_GEOM, &st, 3, 0);
208 g_duplicate_bio(struct bio *bp)
212 bp2 = uma_zalloc(biozone, M_WAITOK | M_ZERO);
213 bp2->bio_parent = bp;
214 bp2->bio_cmd = bp->bio_cmd;
215 bp2->bio_length = bp->bio_length;
216 bp2->bio_offset = bp->bio_offset;
217 bp2->bio_data = bp->bio_data;
218 bp2->bio_attribute = bp->bio_attribute;
221 if ((KTR_COMPILE & KTR_GEOM) && (ktr_mask & KTR_GEOM)) {
224 CTR2(KTR_GEOM, "g_duplicate_bio(%p): %p", bp, bp2);
226 CTRSTACK(KTR_GEOM, &st, 3, 0);
236 g_bioq_init(&g_bio_run_down);
237 g_bioq_init(&g_bio_run_up);
238 g_bioq_init(&g_bio_run_task);
239 biozone = uma_zcreate("g_bio", sizeof (struct bio),
246 g_io_getattr(const char *attr, struct g_consumer *cp, int *len, void *ptr)
251 g_trace(G_T_BIO, "bio_getattr(%s)", attr);
253 bp->bio_cmd = BIO_GETATTR;
255 bp->bio_attribute = attr;
256 bp->bio_length = *len;
258 g_io_request(bp, cp);
259 error = biowait(bp, "ggetattr");
260 *len = bp->bio_completed;
266 g_io_flush(struct g_consumer *cp)
271 g_trace(G_T_BIO, "bio_flush(%s)", cp->provider->name);
273 bp->bio_cmd = BIO_FLUSH;
274 bp->bio_flags |= BIO_ORDERED;
276 bp->bio_attribute = NULL;
277 bp->bio_offset = cp->provider->mediasize;
280 g_io_request(bp, cp);
281 error = biowait(bp, "gflush");
287 g_io_check(struct bio *bp)
289 struct g_consumer *cp;
290 struct g_provider *pp;
295 /* Fail if access counters dont allow the operation */
296 switch(bp->bio_cmd) {
311 /* if provider is marked for error, don't disturb. */
314 if (cp->flags & G_CF_ORPHAN)
317 switch(bp->bio_cmd) {
321 /* Zero sectorsize or mediasize is probably a lack of media. */
322 if (pp->sectorsize == 0 || pp->mediasize == 0)
324 /* Reject I/O not on sector boundary */
325 if (bp->bio_offset % pp->sectorsize)
327 /* Reject I/O not integral sector long */
328 if (bp->bio_length % pp->sectorsize)
330 /* Reject requests before or past the end of media. */
331 if (bp->bio_offset < 0)
333 if (bp->bio_offset > pp->mediasize)
343 * bio classification support.
345 * g_register_classifier() and g_unregister_classifier()
346 * are used to add/remove a classifier from the list.
347 * The list is protected using the g_bio_run_down lock,
348 * because the classifiers are called in this path.
350 * g_io_request() passes bio's that are not already classified
351 * (i.e. those with bio_classifier1 == NULL) to g_run_classifiers().
352 * Classifiers can store their result in the two fields
353 * bio_classifier1 and bio_classifier2.
354 * A classifier that updates one of the fields should
355 * return a non-zero value.
356 * If no classifier updates the field, g_run_classifiers() sets
357 * bio_classifier1 = BIO_NOTCLASSIFIED to avoid further calls.
361 g_register_classifier(struct g_classifier_hook *hook)
364 g_bioq_lock(&g_bio_run_down);
365 TAILQ_INSERT_TAIL(&g_classifier_tailq, hook, link);
366 g_bioq_unlock(&g_bio_run_down);
372 g_unregister_classifier(struct g_classifier_hook *hook)
374 struct g_classifier_hook *entry;
376 g_bioq_lock(&g_bio_run_down);
377 TAILQ_FOREACH(entry, &g_classifier_tailq, link) {
379 TAILQ_REMOVE(&g_classifier_tailq, hook, link);
383 g_bioq_unlock(&g_bio_run_down);
387 g_run_classifiers(struct bio *bp)
389 struct g_classifier_hook *hook;
392 TAILQ_FOREACH(hook, &g_classifier_tailq, link)
393 classified |= hook->func(hook->arg, bp);
396 bp->bio_classifier1 = BIO_NOTCLASSIFIED;
400 g_io_request(struct bio *bp, struct g_consumer *cp)
402 struct g_provider *pp;
405 KASSERT(cp != NULL, ("NULL cp in g_io_request"));
406 KASSERT(bp != NULL, ("NULL bp in g_io_request"));
408 KASSERT(pp != NULL, ("consumer not attached in g_io_request"));
410 KASSERT(bp->bio_driver1 == NULL,
411 ("bio_driver1 used by the consumer (geom %s)", cp->geom->name));
412 KASSERT(bp->bio_driver2 == NULL,
413 ("bio_driver2 used by the consumer (geom %s)", cp->geom->name));
414 KASSERT(bp->bio_pflags == 0,
415 ("bio_pflags used by the consumer (geom %s)", cp->geom->name));
417 * Remember consumer's private fields, so we can detect if they were
418 * modified by the provider.
420 bp->_bio_caller1 = bp->bio_caller1;
421 bp->_bio_caller2 = bp->bio_caller2;
422 bp->_bio_cflags = bp->bio_cflags;
425 if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_GETATTR)) {
426 KASSERT(bp->bio_data != NULL,
427 ("NULL bp->data in g_io_request(cmd=%hhu)", bp->bio_cmd));
429 if (bp->bio_cmd & (BIO_DELETE|BIO_FLUSH)) {
430 KASSERT(bp->bio_data == NULL,
431 ("non-NULL bp->data in g_io_request(cmd=%hhu)",
434 if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) {
435 KASSERT(bp->bio_offset % cp->provider->sectorsize == 0,
436 ("wrong offset %jd for sectorsize %u",
437 bp->bio_offset, cp->provider->sectorsize));
438 KASSERT(bp->bio_length % cp->provider->sectorsize == 0,
439 ("wrong length %jd for sectorsize %u",
440 bp->bio_length, cp->provider->sectorsize));
443 g_trace(G_T_BIO, "bio_request(%p) from %p(%s) to %p(%s) cmd %d",
444 bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd);
449 bp->bio_completed = 0;
451 KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
452 ("Bio already on queue bp=%p", bp));
453 bp->bio_flags |= BIO_ONQUEUE;
456 binuptime(&bp->bio_t0);
458 getbinuptime(&bp->bio_t0);
461 * The statistics collection is lockless, as such, but we
462 * can not update one instance of the statistics from more
463 * than one thread at a time, so grab the lock first.
465 * We also use the lock to protect the list of classifiers.
467 g_bioq_lock(&g_bio_run_down);
469 if (!TAILQ_EMPTY(&g_classifier_tailq) && !bp->bio_classifier1)
470 g_run_classifiers(bp);
472 if (g_collectstats & 1)
473 devstat_start_transaction(pp->stat, &bp->bio_t0);
474 if (g_collectstats & 2)
475 devstat_start_transaction(cp->stat, &bp->bio_t0);
479 first = TAILQ_EMPTY(&g_bio_run_down.bio_queue);
480 TAILQ_INSERT_TAIL(&g_bio_run_down.bio_queue, bp, bio_queue);
481 g_bio_run_down.bio_queue_length++;
482 g_bioq_unlock(&g_bio_run_down);
484 /* Pass it on down. */
486 wakeup(&g_wait_down);
490 g_io_deliver(struct bio *bp, int error)
492 struct g_consumer *cp;
493 struct g_provider *pp;
496 KASSERT(bp != NULL, ("NULL bp in g_io_deliver"));
498 KASSERT(pp != NULL, ("NULL bio_to in g_io_deliver"));
501 bp->bio_error = error;
505 KASSERT(cp != NULL, ("NULL bio_from in g_io_deliver"));
506 KASSERT(cp->geom != NULL, ("NULL bio_from->geom in g_io_deliver"));
509 * Some classes - GJournal in particular - can modify bio's
510 * private fields while the bio is in transit; G_GEOM_VOLATILE_BIO
511 * flag means it's an expected behaviour for that particular geom.
513 if ((cp->geom->flags & G_GEOM_VOLATILE_BIO) == 0) {
514 KASSERT(bp->bio_caller1 == bp->_bio_caller1,
515 ("bio_caller1 used by the provider %s", pp->name));
516 KASSERT(bp->bio_caller2 == bp->_bio_caller2,
517 ("bio_caller2 used by the provider %s", pp->name));
518 KASSERT(bp->bio_cflags == bp->_bio_cflags,
519 ("bio_cflags used by the provider %s", pp->name));
522 KASSERT(bp->bio_completed >= 0, ("bio_completed can't be less than 0"));
523 KASSERT(bp->bio_completed <= bp->bio_length,
524 ("bio_completed can't be greater than bio_length"));
527 "g_io_deliver(%p) from %p(%s) to %p(%s) cmd %d error %d off %jd len %jd",
528 bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd, error,
529 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
531 KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
532 ("Bio already on queue bp=%p", bp));
535 * XXX: next two doesn't belong here
537 bp->bio_bcount = bp->bio_length;
538 bp->bio_resid = bp->bio_bcount - bp->bio_completed;
541 * The statistics collection is lockless, as such, but we
542 * can not update one instance of the statistics from more
543 * than one thread at a time, so grab the lock first.
545 g_bioq_lock(&g_bio_run_up);
546 if (g_collectstats & 1)
547 devstat_end_transaction_bio(pp->stat, bp);
548 if (g_collectstats & 2)
549 devstat_end_transaction_bio(cp->stat, bp);
553 if (error != ENOMEM) {
554 bp->bio_error = error;
555 first = TAILQ_EMPTY(&g_bio_run_up.bio_queue);
556 TAILQ_INSERT_TAIL(&g_bio_run_up.bio_queue, bp, bio_queue);
557 bp->bio_flags |= BIO_ONQUEUE;
558 g_bio_run_up.bio_queue_length++;
559 g_bioq_unlock(&g_bio_run_up);
564 g_bioq_unlock(&g_bio_run_up);
567 printf("ENOMEM %p on %p(%s)\n", bp, pp, pp->name);
568 bp->bio_children = 0;
570 bp->bio_driver1 = NULL;
571 bp->bio_driver2 = NULL;
573 g_io_request(bp, cp);
579 g_io_schedule_down(struct thread *tp __unused)
586 g_bioq_lock(&g_bio_run_down);
587 bp = g_bioq_first(&g_bio_run_down);
589 CTR0(KTR_GEOM, "g_down going to sleep");
590 msleep(&g_wait_down, &g_bio_run_down.bio_queue_lock,
591 PRIBIO | PDROP, "-", 0);
594 CTR0(KTR_GEOM, "g_down has work to do");
595 g_bioq_unlock(&g_bio_run_down);
597 CTR1(KTR_GEOM, "g_down pacing self (pace %d)", pace);
598 pause("g_down", hz/10);
601 error = g_io_check(bp);
603 CTR3(KTR_GEOM, "g_down g_io_check on bp %p provider "
604 "%s returned %d", bp, bp->bio_to->name, error);
605 g_io_deliver(bp, error);
608 CTR2(KTR_GEOM, "g_down processing bp %p provider %s", bp,
610 switch (bp->bio_cmd) {
614 /* Truncate requests to the end of providers media. */
616 * XXX: What if we truncate because of offset being
619 excess = bp->bio_offset + bp->bio_length;
620 if (excess > bp->bio_to->mediasize) {
621 excess -= bp->bio_to->mediasize;
622 bp->bio_length -= excess;
624 CTR3(KTR_GEOM, "g_down truncated bio "
625 "%p provider %s by %d", bp,
626 bp->bio_to->name, excess);
628 /* Deliver zero length transfers right here. */
629 if (bp->bio_length == 0) {
631 CTR2(KTR_GEOM, "g_down terminated 0-length "
632 "bp %p provider %s", bp, bp->bio_to->name);
639 THREAD_NO_SLEEPING();
640 CTR4(KTR_GEOM, "g_down starting bp %p provider %s off %ld "
641 "len %ld", bp, bp->bio_to->name, bp->bio_offset,
643 bp->bio_to->geom->start(bp);
644 THREAD_SLEEPING_OK();
649 bio_taskqueue(struct bio *bp, bio_task_t *func, void *arg)
652 bp->bio_task_arg = arg;
654 * The taskqueue is actually just a second queue off the "up"
655 * queue, so we use the same lock.
657 g_bioq_lock(&g_bio_run_up);
658 KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
659 ("Bio already on queue bp=%p target taskq", bp));
660 bp->bio_flags |= BIO_ONQUEUE;
661 TAILQ_INSERT_TAIL(&g_bio_run_task.bio_queue, bp, bio_queue);
662 g_bio_run_task.bio_queue_length++;
664 g_bioq_unlock(&g_bio_run_up);
669 g_io_schedule_up(struct thread *tp __unused)
673 g_bioq_lock(&g_bio_run_up);
674 bp = g_bioq_first(&g_bio_run_task);
676 g_bioq_unlock(&g_bio_run_up);
677 THREAD_NO_SLEEPING();
678 CTR1(KTR_GEOM, "g_up processing task bp %p", bp);
679 bp->bio_task(bp->bio_task_arg);
680 THREAD_SLEEPING_OK();
683 bp = g_bioq_first(&g_bio_run_up);
685 g_bioq_unlock(&g_bio_run_up);
686 THREAD_NO_SLEEPING();
687 CTR4(KTR_GEOM, "g_up biodone bp %p provider %s off "
688 "%jd len %ld", bp, bp->bio_to->name,
689 bp->bio_offset, bp->bio_length);
691 THREAD_SLEEPING_OK();
694 CTR0(KTR_GEOM, "g_up going to sleep");
695 msleep(&g_wait_up, &g_bio_run_up.bio_queue_lock,
696 PRIBIO | PDROP, "-", 0);
701 g_read_data(struct g_consumer *cp, off_t offset, off_t length, int *error)
707 KASSERT(length > 0 && length >= cp->provider->sectorsize &&
708 length <= MAXPHYS, ("g_read_data(): invalid length %jd",
712 bp->bio_cmd = BIO_READ;
714 bp->bio_offset = offset;
715 bp->bio_length = length;
716 ptr = g_malloc(length, M_WAITOK);
718 g_io_request(bp, cp);
719 errorc = biowait(bp, "gread");
731 g_write_data(struct g_consumer *cp, off_t offset, void *ptr, off_t length)
736 KASSERT(length > 0 && length >= cp->provider->sectorsize &&
737 length <= MAXPHYS, ("g_write_data(): invalid length %jd",
741 bp->bio_cmd = BIO_WRITE;
743 bp->bio_offset = offset;
744 bp->bio_length = length;
746 g_io_request(bp, cp);
747 error = biowait(bp, "gwrite");
753 g_delete_data(struct g_consumer *cp, off_t offset, off_t length)
758 KASSERT(length > 0 && length >= cp->provider->sectorsize,
759 ("g_delete_data(): invalid length %jd", (intmax_t)length));
762 bp->bio_cmd = BIO_DELETE;
764 bp->bio_offset = offset;
765 bp->bio_length = length;
767 g_io_request(bp, cp);
768 error = biowait(bp, "gdelete");
774 g_print_bio(struct bio *bp)
776 const char *pname, *cmd = NULL;
778 if (bp->bio_to != NULL)
779 pname = bp->bio_to->name;
783 switch (bp->bio_cmd) {
786 printf("%s[%s(attr=%s)]", pname, cmd, bp->bio_attribute);
790 printf("%s[%s]", pname, cmd);
803 printf("%s[%s()]", pname, cmd);
806 printf("%s[%s(offset=%jd, length=%jd)]", pname, cmd,
807 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);