geom_io.c revision 136755 
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: head/sys/geom/geom_io.c 136755 2004-10-21 18:35:24Z rwatson $");
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
46#include <sys/errno.h>
47#include <geom/geom.h>
48#include <geom/geom_int.h>
49#include <sys/devicestat.h>
50
51#include <vm/uma.h>
52
53static struct g_bioq g_bio_run_down;
54static struct g_bioq g_bio_run_up;
55static struct g_bioq g_bio_run_task;
56
57static u_int pace;
58static uma_zone_t	biozone;
59
60#include <machine/atomic.h>
61
62static void
63g_bioq_lock(struct g_bioq *bq)
64{
65
66	mtx_lock(&bq->bio_queue_lock);
67}
68
69static void
70g_bioq_unlock(struct g_bioq *bq)
71{
72
73	mtx_unlock(&bq->bio_queue_lock);
74}
75
76#if 0
77static void
78g_bioq_destroy(struct g_bioq *bq)
79{
80
81	mtx_destroy(&bq->bio_queue_lock);
82}
83#endif
84
85static void
86g_bioq_init(struct g_bioq *bq)
87{
88
89	TAILQ_INIT(&bq->bio_queue);
90	mtx_init(&bq->bio_queue_lock, "bio queue", NULL, MTX_DEF);
91}
92
93static struct bio *
94g_bioq_first(struct g_bioq *bq)
95{
96	struct bio *bp;
97
98	bp = TAILQ_FIRST(&bq->bio_queue);
99	if (bp != NULL) {
100		KASSERT((bp->bio_flags & BIO_ONQUEUE),
101		    ("Bio not on queue bp=%p target %p", bp, bq));
102		bp->bio_flags &= ~BIO_ONQUEUE;
103		TAILQ_REMOVE(&bq->bio_queue, bp, bio_queue);
104		bq->bio_queue_length--;
105	}
106	return (bp);
107}
108
109struct bio *
110g_new_bio(void)
111{
112	struct bio *bp;
113
114	bp = uma_zalloc(biozone, M_NOWAIT | M_ZERO);
115	return (bp);
116}
117
118struct bio *
119g_alloc_bio(void)
120{
121	struct bio *bp;
122
123	bp = uma_zalloc(biozone, M_WAITOK | M_ZERO);
124	return (bp);
125}
126
127void
128g_destroy_bio(struct bio *bp)
129{
130
131	uma_zfree(biozone, bp);
132}
133
134struct bio *
135g_clone_bio(struct bio *bp)
136{
137	struct bio *bp2;
138
139	bp2 = uma_zalloc(biozone, M_NOWAIT | M_ZERO);
140	if (bp2 != NULL) {
141		bp2->bio_parent = bp;
142		bp2->bio_cmd = bp->bio_cmd;
143		bp2->bio_length = bp->bio_length;
144		bp2->bio_offset = bp->bio_offset;
145		bp2->bio_data = bp->bio_data;
146		bp2->bio_attribute = bp->bio_attribute;
147		bp->bio_children++;
148	}
149	return(bp2);
150}
151
152void
153g_io_init()
154{
155
156	g_bioq_init(&g_bio_run_down);
157	g_bioq_init(&g_bio_run_up);
158	g_bioq_init(&g_bio_run_task);
159	biozone = uma_zcreate("g_bio", sizeof (struct bio),
160	    NULL, NULL,
161	    NULL, NULL,
162	    0, 0);
163}
164
165int
166g_io_getattr(const char *attr, struct g_consumer *cp, int *len, void *ptr)
167{
168	struct bio *bp;
169	int error;
170
171	g_trace(G_T_BIO, "bio_getattr(%s)", attr);
172	bp = g_alloc_bio();
173	bp->bio_cmd = BIO_GETATTR;
174	bp->bio_done = NULL;
175	bp->bio_attribute = attr;
176	bp->bio_length = *len;
177	bp->bio_data = ptr;
178	g_io_request(bp, cp);
179	error = biowait(bp, "ggetattr");
180	*len = bp->bio_completed;
181	g_destroy_bio(bp);
182	return (error);
183}
184
185static int
186g_io_check(struct bio *bp)
187{
188	struct g_consumer *cp;
189	struct g_provider *pp;
190
191	cp = bp->bio_from;
192	pp = bp->bio_to;
193
194	/* Fail if access counters dont allow the operation */
195	switch(bp->bio_cmd) {
196	case BIO_READ:
197	case BIO_GETATTR:
198		if (cp->acr == 0)
199			return (EPERM);
200		break;
201	case BIO_WRITE:
202	case BIO_DELETE:
203		if (cp->acw == 0)
204			return (EPERM);
205		break;
206	default:
207		return (EPERM);
208	}
209	/* if provider is marked for error, don't disturb. */
210	if (pp->error)
211		return (pp->error);
212
213	switch(bp->bio_cmd) {
214	case BIO_READ:
215	case BIO_WRITE:
216	case BIO_DELETE:
217		/* Zero sectorsize is a probably lack of media */
218		if (pp->sectorsize == 0)
219			return (ENXIO);
220		/* Reject I/O not on sector boundary */
221		if (bp->bio_offset % pp->sectorsize)
222			return (EINVAL);
223		/* Reject I/O not integral sector long */
224		if (bp->bio_length % pp->sectorsize)
225			return (EINVAL);
226		/* Reject requests before or past the end of media. */
227		if (bp->bio_offset < 0)
228			return (EIO);
229		if (bp->bio_offset > pp->mediasize)
230			return (EIO);
231		break;
232	default:
233		break;
234	}
235	return (0);
236}
237
238void
239g_io_request(struct bio *bp, struct g_consumer *cp)
240{
241	struct g_provider *pp;
242
243	KASSERT(cp != NULL, ("NULL cp in g_io_request"));
244	KASSERT(bp != NULL, ("NULL bp in g_io_request"));
245	KASSERT(bp->bio_data != NULL, ("NULL bp->data in g_io_request"));
246	pp = cp->provider;
247	KASSERT(pp != NULL, ("consumer not attached in g_io_request"));
248
249	if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) {
250		KASSERT(bp->bio_offset % cp->provider->sectorsize == 0,
251		    ("wrong offset %jd for sectorsize %u",
252		    bp->bio_offset, cp->provider->sectorsize));
253		KASSERT(bp->bio_length % cp->provider->sectorsize == 0,
254		    ("wrong length %jd for sectorsize %u",
255		    bp->bio_length, cp->provider->sectorsize));
256	}
257
258	g_trace(G_T_BIO, "bio_request(%p) from %p(%s) to %p(%s) cmd %d",
259	    bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd);
260
261	bp->bio_from = cp;
262	bp->bio_to = pp;
263	bp->bio_error = 0;
264	bp->bio_completed = 0;
265
266	KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
267	    ("Bio already on queue bp=%p", bp));
268	bp->bio_flags |= BIO_ONQUEUE;
269
270	binuptime(&bp->bio_t0);
271	if (g_collectstats & 4)
272		g_bioq_lock(&g_bio_run_down);
273	if (g_collectstats & 1)
274		devstat_start_transaction(pp->stat, &bp->bio_t0);
275	if (g_collectstats & 2)
276		devstat_start_transaction(cp->stat, &bp->bio_t0);
277
278	if (!(g_collectstats & 4))
279		g_bioq_lock(&g_bio_run_down);
280	pp->nstart++;
281	cp->nstart++;
282	TAILQ_INSERT_TAIL(&g_bio_run_down.bio_queue, bp, bio_queue);
283	g_bio_run_down.bio_queue_length++;
284	g_bioq_unlock(&g_bio_run_down);
285
286	/* Pass it on down. */
287	wakeup(&g_wait_down);
288}
289
290void
291g_io_deliver(struct bio *bp, int error)
292{
293	struct g_consumer *cp;
294	struct g_provider *pp;
295
296	KASSERT(bp != NULL, ("NULL bp in g_io_deliver"));
297	pp = bp->bio_to;
298	KASSERT(pp != NULL, ("NULL bio_to in g_io_deliver"));
299	cp = bp->bio_from;
300	if (cp == NULL) {
301		bp->bio_error = error;
302		bp->bio_done(bp);
303		return;
304	}
305	KASSERT(cp != NULL, ("NULL bio_from in g_io_deliver"));
306	KASSERT(cp->geom != NULL, ("NULL bio_from->geom in g_io_deliver"));
307	KASSERT(bp->bio_completed >= 0, ("bio_completed can't be less than 0"));
308	KASSERT(bp->bio_completed <= bp->bio_length,
309	    ("bio_completed can't be greater than bio_length"));
310
311	g_trace(G_T_BIO,
312"g_io_deliver(%p) from %p(%s) to %p(%s) cmd %d error %d off %jd len %jd",
313	    bp, cp, cp->geom->name, pp, pp->name, bp->bio_cmd, error,
314	    (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
315
316	KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
317	    ("Bio already on queue bp=%p", bp));
318
319	/*
320	 * XXX: next two doesn't belong here
321	 */
322	bp->bio_bcount = bp->bio_length;
323	bp->bio_resid = bp->bio_bcount - bp->bio_completed;
324
325	if (g_collectstats & 4)
326		g_bioq_lock(&g_bio_run_up);
327	if (g_collectstats & 1)
328		devstat_end_transaction_bio(pp->stat, bp);
329	if (g_collectstats & 2)
330		devstat_end_transaction_bio(cp->stat, bp);
331	if (!(g_collectstats & 4))
332		g_bioq_lock(&g_bio_run_up);
333	cp->nend++;
334	pp->nend++;
335	if (error != ENOMEM) {
336		bp->bio_error = error;
337		TAILQ_INSERT_TAIL(&g_bio_run_up.bio_queue, bp, bio_queue);
338		bp->bio_flags |= BIO_ONQUEUE;
339		g_bio_run_up.bio_queue_length++;
340		g_bioq_unlock(&g_bio_run_up);
341		wakeup(&g_wait_up);
342		return;
343	}
344	g_bioq_unlock(&g_bio_run_up);
345
346	if (bootverbose)
347		printf("ENOMEM %p on %p(%s)\n", bp, pp, pp->name);
348	bp->bio_children = 0;
349	bp->bio_inbed = 0;
350	g_io_request(bp, cp);
351	pace++;
352	return;
353}
354
355void
356g_io_schedule_down(struct thread *tp __unused)
357{
358	struct bio *bp;
359	off_t excess;
360	int error;
361#ifdef WITNESS
362	struct mtx mymutex;
363
364	bzero(&mymutex, sizeof mymutex);
365	mtx_init(&mymutex, "g_xdown", NULL, MTX_DEF);
366#endif
367
368	for(;;) {
369		g_bioq_lock(&g_bio_run_down);
370		bp = g_bioq_first(&g_bio_run_down);
371		if (bp == NULL) {
372			CTR0(KTR_GEOM, "g_down going to sleep");
373			msleep(&g_wait_down, &g_bio_run_down.bio_queue_lock,
374			    PRIBIO | PDROP, "-", hz/10);
375			continue;
376		}
377		CTR0(KTR_GEOM, "g_down has work to do");
378		g_bioq_unlock(&g_bio_run_down);
379		if (pace > 0) {
380			CTR1(KTR_GEOM, "g_down pacing self (pace %d)", pace);
381			msleep(&error, NULL, PRIBIO, "g_down", hz/10);
382			pace--;
383		}
384		error = g_io_check(bp);
385		if (error) {
386			CTR3(KTR_GEOM, "g_down g_io_check on bp %p provider "
387			    "%s returned %d", bp, bp->bio_to->name, error);
388			g_io_deliver(bp, error);
389			continue;
390		}
391		CTR2(KTR_GEOM, "g_down processing bp %p provider %s", bp,
392		    bp->bio_to->name);
393		switch (bp->bio_cmd) {
394		case BIO_READ:
395		case BIO_WRITE:
396		case BIO_DELETE:
397			/* Truncate requests to the end of providers media. */
398			/*
399			 * XXX: What if we truncate because of offset being
400			 * bad, not length?
401			 */
402			excess = bp->bio_offset + bp->bio_length;
403			if (excess > bp->bio_to->mediasize) {
404				excess -= bp->bio_to->mediasize;
405				bp->bio_length -= excess;
406				if (excess > 0)
407					CTR3(KTR_GEOM, "g_down truncated bio "
408					    "%p provider %s by %d", bp,
409					    bp->bio_to->name, excess);
410			}
411			/* Deliver zero length transfers right here. */
412			if (bp->bio_length == 0) {
413				g_io_deliver(bp, 0);
414				CTR2(KTR_GEOM, "g_down terminated 0-length "
415				    "bp %p provider %s", bp, bp->bio_to->name);
416				continue;
417			}
418			break;
419		default:
420			break;
421		}
422#ifdef WITNESS
423		mtx_lock(&mymutex);
424#endif
425		CTR4(KTR_GEOM, "g_down starting bp %p provider %s off %ld "
426		    "len %ld", bp, bp->bio_to->name, bp->bio_offset,
427		    bp->bio_length);
428		bp->bio_to->geom->start(bp);
429#ifdef WITNESS
430		mtx_unlock(&mymutex);
431#endif
432	}
433}
434
435void
436bio_taskqueue(struct bio *bp, bio_task_t *func, void *arg)
437{
438	bp->bio_task = func;
439	bp->bio_task_arg = arg;
440	/*
441	 * The taskqueue is actually just a second queue off the "up"
442	 * queue, so we use the same lock.
443	 */
444	g_bioq_lock(&g_bio_run_up);
445	KASSERT(!(bp->bio_flags & BIO_ONQUEUE),
446	    ("Bio already on queue bp=%p target taskq", bp));
447	bp->bio_flags |= BIO_ONQUEUE;
448	TAILQ_INSERT_TAIL(&g_bio_run_task.bio_queue, bp, bio_queue);
449	g_bio_run_task.bio_queue_length++;
450	wakeup(&g_wait_up);
451	g_bioq_unlock(&g_bio_run_up);
452}
453
454
455void
456g_io_schedule_up(struct thread *tp __unused)
457{
458	struct bio *bp;
459#ifdef WITNESS
460	struct mtx mymutex;
461
462	bzero(&mymutex, sizeof mymutex);
463	mtx_init(&mymutex, "g_xup", NULL, MTX_DEF);
464#endif
465	for(;;) {
466		g_bioq_lock(&g_bio_run_up);
467		bp = g_bioq_first(&g_bio_run_task);
468		if (bp != NULL) {
469			g_bioq_unlock(&g_bio_run_up);
470#ifdef WITNESS
471			mtx_lock(&mymutex);
472#endif
473			CTR1(KTR_GEOM, "g_up processing task bp %p", bp);
474			bp->bio_task(bp->bio_task_arg);
475#ifdef WITNESS
476			mtx_unlock(&mymutex);
477#endif
478			continue;
479		}
480		bp = g_bioq_first(&g_bio_run_up);
481		if (bp != NULL) {
482			g_bioq_unlock(&g_bio_run_up);
483#ifdef WITNESS
484			mtx_lock(&mymutex);
485#endif
486			CTR4(KTR_GEOM, "g_up biodone bp %p provider %s off "
487			    "%ld len %ld", bp, bp->bio_to->name,
488			    bp->bio_offset, bp->bio_length);
489			biodone(bp);
490#ifdef WITNESS
491			mtx_unlock(&mymutex);
492#endif
493			continue;
494		}
495		CTR0(KTR_GEOM, "g_up going to sleep");
496		msleep(&g_wait_up, &g_bio_run_up.bio_queue_lock,
497		    PRIBIO | PDROP, "-", hz/10);
498	}
499}
500
501void *
502g_read_data(struct g_consumer *cp, off_t offset, off_t length, int *error)
503{
504	struct bio *bp;
505	void *ptr;
506	int errorc;
507
508	KASSERT(length > 0 && length >= cp->provider->sectorsize &&
509	    length <= MAXPHYS, ("g_read_data(): invalid length %jd",
510	    (intmax_t)length));
511
512	bp = g_alloc_bio();
513	bp->bio_cmd = BIO_READ;
514	bp->bio_done = NULL;
515	bp->bio_offset = offset;
516	bp->bio_length = length;
517	ptr = g_malloc(length, M_WAITOK);
518	bp->bio_data = ptr;
519	g_io_request(bp, cp);
520	errorc = biowait(bp, "gread");
521	if (error != NULL)
522		*error = errorc;
523	g_destroy_bio(bp);
524	if (errorc) {
525		g_free(ptr);
526		ptr = NULL;
527	}
528	return (ptr);
529}
530
531int
532g_write_data(struct g_consumer *cp, off_t offset, void *ptr, off_t length)
533{
534	struct bio *bp;
535	int error;
536
537	KASSERT(length > 0 && length >= cp->provider->sectorsize &&
538	    length <= MAXPHYS, ("g_write_data(): invalid length %jd",
539	    (intmax_t)length));
540
541	bp = g_alloc_bio();
542	bp->bio_cmd = BIO_WRITE;
543	bp->bio_done = NULL;
544	bp->bio_offset = offset;
545	bp->bio_length = length;
546	bp->bio_data = ptr;
547	g_io_request(bp, cp);
548	error = biowait(bp, "gwrite");
549	g_destroy_bio(bp);
550	return (error);
551}
552
553void
554g_print_bio(struct bio *bp)
555{
556	const char *pname, *cmd = NULL;
557
558	if (bp->bio_to != NULL)
559		pname = bp->bio_to->name;
560	else
561		pname = "[unknown]";
562
563	switch (bp->bio_cmd) {
564	case BIO_GETATTR:
565		cmd = "GETATTR";
566		printf("%s[%s(attr=%s)]", pname, cmd, bp->bio_attribute);
567		return;
568	case BIO_READ:
569		cmd = "READ";
570	case BIO_WRITE:
571		if (cmd == NULL)
572			cmd = "WRITE";
573	case BIO_DELETE:
574		if (cmd == NULL)
575			cmd = "DELETE";
576		printf("%s[%s(offset=%jd, length=%jd)]", pname, cmd,
577		    (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length);
578		return;
579	default:
580		cmd = "UNKNOWN";
581		printf("%s[%s()]", pname, cmd);
582		return;
583	}
584	/* NOTREACHED */
585}
586