xref: /freebsd/sys/dev/virtio/block/virtio_blk.c (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /* Driver for VirtIO block devices. */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/bio.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/sglist.h>
41 #include <sys/sysctl.h>
42 #include <sys/lock.h>
43 #include <sys/mutex.h>
44 #include <sys/queue.h>
45 
46 #include <geom/geom.h>
47 #include <geom/geom_disk.h>
48 
49 #include <machine/bus.h>
50 #include <machine/resource.h>
51 #include <sys/bus.h>
52 #include <sys/rman.h>
53 
54 #include <dev/virtio/virtio.h>
55 #include <dev/virtio/virtqueue.h>
56 #include <dev/virtio/block/virtio_blk.h>
57 
58 #include "virtio_if.h"
59 
60 struct vtblk_request {
61 	struct virtio_blk_outhdr	 vbr_hdr;
62 	struct bio			*vbr_bp;
63 	uint8_t				 vbr_ack;
64 	TAILQ_ENTRY(vtblk_request)	 vbr_link;
65 };
66 
67 enum vtblk_cache_mode {
68 	VTBLK_CACHE_WRITETHROUGH,
69 	VTBLK_CACHE_WRITEBACK,
70 	VTBLK_CACHE_MAX
71 };
72 
73 struct vtblk_softc {
74 	device_t		 vtblk_dev;
75 	struct mtx		 vtblk_mtx;
76 	uint64_t		 vtblk_features;
77 	uint32_t		 vtblk_flags;
78 #define VTBLK_FLAG_INDIRECT	0x0001
79 #define VTBLK_FLAG_READONLY	0x0002
80 #define VTBLK_FLAG_DETACH	0x0004
81 #define VTBLK_FLAG_SUSPEND	0x0008
82 #define VTBLK_FLAG_BARRIER	0x0010
83 #define VTBLK_FLAG_WC_CONFIG	0x0020
84 
85 	struct virtqueue	*vtblk_vq;
86 	struct sglist		*vtblk_sglist;
87 	struct disk		*vtblk_disk;
88 
89 	struct bio_queue_head	 vtblk_bioq;
90 	TAILQ_HEAD(, vtblk_request)
91 				 vtblk_req_free;
92 	TAILQ_HEAD(, vtblk_request)
93 				 vtblk_req_ready;
94 	struct vtblk_request	*vtblk_req_ordered;
95 
96 	int			 vtblk_max_nsegs;
97 	int			 vtblk_request_count;
98 	enum vtblk_cache_mode	 vtblk_write_cache;
99 
100 	struct bio_queue	 vtblk_dump_queue;
101 	struct vtblk_request	 vtblk_dump_request;
102 };
103 
104 static struct virtio_feature_desc vtblk_feature_desc[] = {
105 	{ VIRTIO_BLK_F_BARRIER,		"HostBarrier"	},
106 	{ VIRTIO_BLK_F_SIZE_MAX,	"MaxSegSize"	},
107 	{ VIRTIO_BLK_F_SEG_MAX,		"MaxNumSegs"	},
108 	{ VIRTIO_BLK_F_GEOMETRY,	"DiskGeometry"	},
109 	{ VIRTIO_BLK_F_RO,		"ReadOnly"	},
110 	{ VIRTIO_BLK_F_BLK_SIZE,	"BlockSize"	},
111 	{ VIRTIO_BLK_F_SCSI,		"SCSICmds"	},
112 	{ VIRTIO_BLK_F_WCE,		"WriteCache"	},
113 	{ VIRTIO_BLK_F_TOPOLOGY,	"Topology"	},
114 	{ VIRTIO_BLK_F_CONFIG_WCE,	"ConfigWCE"	},
115 
116 	{ 0, NULL }
117 };
118 
119 static int	vtblk_modevent(module_t, int, void *);
120 
121 static int	vtblk_probe(device_t);
122 static int	vtblk_attach(device_t);
123 static int	vtblk_detach(device_t);
124 static int	vtblk_suspend(device_t);
125 static int	vtblk_resume(device_t);
126 static int	vtblk_shutdown(device_t);
127 static int	vtblk_config_change(device_t);
128 
129 static int	vtblk_open(struct disk *);
130 static int	vtblk_close(struct disk *);
131 static int	vtblk_ioctl(struct disk *, u_long, void *, int,
132 		    struct thread *);
133 static int	vtblk_dump(void *, void *, vm_offset_t, off_t, size_t);
134 static void	vtblk_strategy(struct bio *);
135 
136 static void	vtblk_negotiate_features(struct vtblk_softc *);
137 static void	vtblk_setup_features(struct vtblk_softc *);
138 static int	vtblk_maximum_segments(struct vtblk_softc *,
139 		    struct virtio_blk_config *);
140 static int	vtblk_alloc_virtqueue(struct vtblk_softc *);
141 static void	vtblk_resize_disk(struct vtblk_softc *, uint64_t);
142 static void	vtblk_alloc_disk(struct vtblk_softc *,
143 		    struct virtio_blk_config *);
144 static void	vtblk_create_disk(struct vtblk_softc *);
145 
146 static int	vtblk_request_prealloc(struct vtblk_softc *);
147 static void	vtblk_request_free(struct vtblk_softc *);
148 static struct vtblk_request *
149 		vtblk_request_dequeue(struct vtblk_softc *);
150 static void	vtblk_request_enqueue(struct vtblk_softc *,
151 		    struct vtblk_request *);
152 static struct vtblk_request *
153 		vtblk_request_next_ready(struct vtblk_softc *);
154 static void	vtblk_request_requeue_ready(struct vtblk_softc *,
155 		    struct vtblk_request *);
156 static struct vtblk_request *
157 		vtblk_request_next(struct vtblk_softc *);
158 static struct vtblk_request *
159 		vtblk_request_bio(struct vtblk_softc *);
160 static int	vtblk_request_execute(struct vtblk_softc *,
161 		    struct vtblk_request *);
162 static int	vtblk_request_error(struct vtblk_request *);
163 
164 static void	vtblk_queue_completed(struct vtblk_softc *,
165 		    struct bio_queue *);
166 static void	vtblk_done_completed(struct vtblk_softc *,
167 		    struct bio_queue *);
168 static void	vtblk_drain_vq(struct vtblk_softc *);
169 static void	vtblk_drain(struct vtblk_softc *);
170 
171 static void	vtblk_startio(struct vtblk_softc *);
172 static void	vtblk_bio_done(struct vtblk_softc *, struct bio *, int);
173 
174 static void	vtblk_read_config(struct vtblk_softc *,
175 		    struct virtio_blk_config *);
176 static void	vtblk_ident(struct vtblk_softc *);
177 static int	vtblk_poll_request(struct vtblk_softc *,
178 		    struct vtblk_request *);
179 static int	vtblk_quiesce(struct vtblk_softc *);
180 static void	vtblk_vq_intr(void *);
181 static void	vtblk_stop(struct vtblk_softc *);
182 
183 static void	vtblk_dump_quiesce(struct vtblk_softc *);
184 static int	vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t);
185 static int	vtblk_dump_flush(struct vtblk_softc *);
186 static void	vtblk_dump_complete(struct vtblk_softc *);
187 
188 static void	vtblk_set_write_cache(struct vtblk_softc *, int);
189 static int	vtblk_write_cache_enabled(struct vtblk_softc *sc,
190 		    struct virtio_blk_config *);
191 static int	vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS);
192 
193 static void	vtblk_setup_sysctl(struct vtblk_softc *);
194 static int	vtblk_tunable_int(struct vtblk_softc *, const char *, int);
195 
196 /* Tunables. */
197 static int vtblk_no_ident = 0;
198 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident);
199 static int vtblk_writecache_mode = -1;
200 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode);
201 
202 /* Features desired/implemented by this driver. */
203 #define VTBLK_FEATURES \
204     (VIRTIO_BLK_F_BARRIER		| \
205      VIRTIO_BLK_F_SIZE_MAX		| \
206      VIRTIO_BLK_F_SEG_MAX		| \
207      VIRTIO_BLK_F_GEOMETRY		| \
208      VIRTIO_BLK_F_RO			| \
209      VIRTIO_BLK_F_BLK_SIZE		| \
210      VIRTIO_BLK_F_WCE			| \
211      VIRTIO_BLK_F_TOPOLOGY		| \
212      VIRTIO_BLK_F_CONFIG_WCE		| \
213      VIRTIO_RING_F_INDIRECT_DESC)
214 
215 #define VTBLK_MTX(_sc)		&(_sc)->vtblk_mtx
216 #define VTBLK_LOCK_INIT(_sc, _name) \
217 				mtx_init(VTBLK_MTX((_sc)), (_name), \
218 				    "VirtIO Block Lock", MTX_DEF)
219 #define VTBLK_LOCK(_sc)		mtx_lock(VTBLK_MTX((_sc)))
220 #define VTBLK_UNLOCK(_sc)	mtx_unlock(VTBLK_MTX((_sc)))
221 #define VTBLK_LOCK_DESTROY(_sc)	mtx_destroy(VTBLK_MTX((_sc)))
222 #define VTBLK_LOCK_ASSERT(_sc)	mtx_assert(VTBLK_MTX((_sc)), MA_OWNED)
223 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \
224 				mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED)
225 
226 #define VTBLK_DISK_NAME		"vtbd"
227 #define VTBLK_QUIESCE_TIMEOUT	(30 * hz)
228 
229 /*
230  * Each block request uses at least two segments - one for the header
231  * and one for the status.
232  */
233 #define VTBLK_MIN_SEGMENTS	2
234 
235 static device_method_t vtblk_methods[] = {
236 	/* Device methods. */
237 	DEVMETHOD(device_probe,		vtblk_probe),
238 	DEVMETHOD(device_attach,	vtblk_attach),
239 	DEVMETHOD(device_detach,	vtblk_detach),
240 	DEVMETHOD(device_suspend,	vtblk_suspend),
241 	DEVMETHOD(device_resume,	vtblk_resume),
242 	DEVMETHOD(device_shutdown,	vtblk_shutdown),
243 
244 	/* VirtIO methods. */
245 	DEVMETHOD(virtio_config_change,	vtblk_config_change),
246 
247 	DEVMETHOD_END
248 };
249 
250 static driver_t vtblk_driver = {
251 	"vtblk",
252 	vtblk_methods,
253 	sizeof(struct vtblk_softc)
254 };
255 static devclass_t vtblk_devclass;
256 
257 DRIVER_MODULE(virtio_blk, virtio_mmio, vtblk_driver, vtblk_devclass,
258     vtblk_modevent, 0);
259 DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass,
260     vtblk_modevent, 0);
261 MODULE_VERSION(virtio_blk, 1);
262 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1);
263 
264 VIRTIO_SIMPLE_PNPTABLE(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter");
265 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, virtio_blk);
266 VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_blk);
267 
268 static int
269 vtblk_modevent(module_t mod, int type, void *unused)
270 {
271 	int error;
272 
273 	error = 0;
274 
275 	switch (type) {
276 	case MOD_LOAD:
277 	case MOD_QUIESCE:
278 	case MOD_UNLOAD:
279 	case MOD_SHUTDOWN:
280 		break;
281 	default:
282 		error = EOPNOTSUPP;
283 		break;
284 	}
285 
286 	return (error);
287 }
288 
289 static int
290 vtblk_probe(device_t dev)
291 {
292 	return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk));
293 }
294 
295 static int
296 vtblk_attach(device_t dev)
297 {
298 	struct vtblk_softc *sc;
299 	struct virtio_blk_config blkcfg;
300 	int error;
301 
302 	virtio_set_feature_desc(dev, vtblk_feature_desc);
303 
304 	sc = device_get_softc(dev);
305 	sc->vtblk_dev = dev;
306 	VTBLK_LOCK_INIT(sc, device_get_nameunit(dev));
307 	bioq_init(&sc->vtblk_bioq);
308 	TAILQ_INIT(&sc->vtblk_dump_queue);
309 	TAILQ_INIT(&sc->vtblk_req_free);
310 	TAILQ_INIT(&sc->vtblk_req_ready);
311 
312 	vtblk_setup_sysctl(sc);
313 	vtblk_setup_features(sc);
314 
315 	vtblk_read_config(sc, &blkcfg);
316 
317 	/*
318 	 * With the current sglist(9) implementation, it is not easy
319 	 * for us to support a maximum segment size as adjacent
320 	 * segments are coalesced. For now, just make sure it's larger
321 	 * than the maximum supported transfer size.
322 	 */
323 	if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) {
324 		if (blkcfg.size_max < MAXPHYS) {
325 			error = ENOTSUP;
326 			device_printf(dev, "host requires unsupported "
327 			    "maximum segment size feature\n");
328 			goto fail;
329 		}
330 	}
331 
332 	sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg);
333 	if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) {
334 		error = EINVAL;
335 		device_printf(dev, "fewer than minimum number of segments "
336 		    "allowed: %d\n", sc->vtblk_max_nsegs);
337 		goto fail;
338 	}
339 
340 	sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT);
341 	if (sc->vtblk_sglist == NULL) {
342 		error = ENOMEM;
343 		device_printf(dev, "cannot allocate sglist\n");
344 		goto fail;
345 	}
346 
347 	error = vtblk_alloc_virtqueue(sc);
348 	if (error) {
349 		device_printf(dev, "cannot allocate virtqueue\n");
350 		goto fail;
351 	}
352 
353 	error = vtblk_request_prealloc(sc);
354 	if (error) {
355 		device_printf(dev, "cannot preallocate requests\n");
356 		goto fail;
357 	}
358 
359 	vtblk_alloc_disk(sc, &blkcfg);
360 
361 	error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY);
362 	if (error) {
363 		device_printf(dev, "cannot setup virtqueue interrupt\n");
364 		goto fail;
365 	}
366 
367 	vtblk_create_disk(sc);
368 
369 	virtqueue_enable_intr(sc->vtblk_vq);
370 
371 fail:
372 	if (error)
373 		vtblk_detach(dev);
374 
375 	return (error);
376 }
377 
378 static int
379 vtblk_detach(device_t dev)
380 {
381 	struct vtblk_softc *sc;
382 
383 	sc = device_get_softc(dev);
384 
385 	VTBLK_LOCK(sc);
386 	sc->vtblk_flags |= VTBLK_FLAG_DETACH;
387 	if (device_is_attached(dev))
388 		vtblk_stop(sc);
389 	VTBLK_UNLOCK(sc);
390 
391 	vtblk_drain(sc);
392 
393 	if (sc->vtblk_disk != NULL) {
394 		disk_destroy(sc->vtblk_disk);
395 		sc->vtblk_disk = NULL;
396 	}
397 
398 	if (sc->vtblk_sglist != NULL) {
399 		sglist_free(sc->vtblk_sglist);
400 		sc->vtblk_sglist = NULL;
401 	}
402 
403 	VTBLK_LOCK_DESTROY(sc);
404 
405 	return (0);
406 }
407 
408 static int
409 vtblk_suspend(device_t dev)
410 {
411 	struct vtblk_softc *sc;
412 	int error;
413 
414 	sc = device_get_softc(dev);
415 
416 	VTBLK_LOCK(sc);
417 	sc->vtblk_flags |= VTBLK_FLAG_SUSPEND;
418 	/* XXX BMV: virtio_stop(), etc needed here? */
419 	error = vtblk_quiesce(sc);
420 	if (error)
421 		sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
422 	VTBLK_UNLOCK(sc);
423 
424 	return (error);
425 }
426 
427 static int
428 vtblk_resume(device_t dev)
429 {
430 	struct vtblk_softc *sc;
431 
432 	sc = device_get_softc(dev);
433 
434 	VTBLK_LOCK(sc);
435 	/* XXX BMV: virtio_reinit(), etc needed here? */
436 	sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
437 	vtblk_startio(sc);
438 	VTBLK_UNLOCK(sc);
439 
440 	return (0);
441 }
442 
443 static int
444 vtblk_shutdown(device_t dev)
445 {
446 
447 	return (0);
448 }
449 
450 static int
451 vtblk_config_change(device_t dev)
452 {
453 	struct vtblk_softc *sc;
454 	struct virtio_blk_config blkcfg;
455 	uint64_t capacity;
456 
457 	sc = device_get_softc(dev);
458 
459 	vtblk_read_config(sc, &blkcfg);
460 
461 	/* Capacity is always in 512-byte units. */
462 	capacity = blkcfg.capacity * 512;
463 
464 	if (sc->vtblk_disk->d_mediasize != capacity)
465 		vtblk_resize_disk(sc, capacity);
466 
467 	return (0);
468 }
469 
470 static int
471 vtblk_open(struct disk *dp)
472 {
473 	struct vtblk_softc *sc;
474 
475 	if ((sc = dp->d_drv1) == NULL)
476 		return (ENXIO);
477 
478 	return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0);
479 }
480 
481 static int
482 vtblk_close(struct disk *dp)
483 {
484 	struct vtblk_softc *sc;
485 
486 	if ((sc = dp->d_drv1) == NULL)
487 		return (ENXIO);
488 
489 	return (0);
490 }
491 
492 static int
493 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag,
494     struct thread *td)
495 {
496 	struct vtblk_softc *sc;
497 
498 	if ((sc = dp->d_drv1) == NULL)
499 		return (ENXIO);
500 
501 	return (ENOTTY);
502 }
503 
504 static int
505 vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
506     size_t length)
507 {
508 	struct disk *dp;
509 	struct vtblk_softc *sc;
510 	int error;
511 
512 	dp = arg;
513 	error = 0;
514 
515 	if ((sc = dp->d_drv1) == NULL)
516 		return (ENXIO);
517 
518 	VTBLK_LOCK(sc);
519 
520 	vtblk_dump_quiesce(sc);
521 
522 	if (length > 0)
523 		error = vtblk_dump_write(sc, virtual, offset, length);
524 	if (error || (virtual == NULL && offset == 0))
525 		vtblk_dump_complete(sc);
526 
527 	VTBLK_UNLOCK(sc);
528 
529 	return (error);
530 }
531 
532 static void
533 vtblk_strategy(struct bio *bp)
534 {
535 	struct vtblk_softc *sc;
536 
537 	if ((sc = bp->bio_disk->d_drv1) == NULL) {
538 		vtblk_bio_done(NULL, bp, EINVAL);
539 		return;
540 	}
541 
542 	/*
543 	 * Fail any write if RO. Unfortunately, there does not seem to
544 	 * be a better way to report our readonly'ness to GEOM above.
545 	 */
546 	if (sc->vtblk_flags & VTBLK_FLAG_READONLY &&
547 	    (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_FLUSH)) {
548 		vtblk_bio_done(sc, bp, EROFS);
549 		return;
550 	}
551 
552 	if ((bp->bio_cmd != BIO_READ) && (bp->bio_cmd != BIO_WRITE) &&
553 	    (bp->bio_cmd != BIO_FLUSH)) {
554 		vtblk_bio_done(sc, bp, EOPNOTSUPP);
555 		return;
556 	}
557 
558 	VTBLK_LOCK(sc);
559 
560 	if (sc->vtblk_flags & VTBLK_FLAG_DETACH) {
561 		VTBLK_UNLOCK(sc);
562 		vtblk_bio_done(sc, bp, ENXIO);
563 		return;
564 	}
565 
566 	bioq_insert_tail(&sc->vtblk_bioq, bp);
567 	vtblk_startio(sc);
568 
569 	VTBLK_UNLOCK(sc);
570 }
571 
572 static void
573 vtblk_negotiate_features(struct vtblk_softc *sc)
574 {
575 	device_t dev;
576 	uint64_t features;
577 
578 	dev = sc->vtblk_dev;
579 	features = VTBLK_FEATURES;
580 
581 	sc->vtblk_features = virtio_negotiate_features(dev, features);
582 }
583 
584 static void
585 vtblk_setup_features(struct vtblk_softc *sc)
586 {
587 	device_t dev;
588 
589 	dev = sc->vtblk_dev;
590 
591 	vtblk_negotiate_features(sc);
592 
593 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
594 		sc->vtblk_flags |= VTBLK_FLAG_INDIRECT;
595 	if (virtio_with_feature(dev, VIRTIO_BLK_F_RO))
596 		sc->vtblk_flags |= VTBLK_FLAG_READONLY;
597 	if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER))
598 		sc->vtblk_flags |= VTBLK_FLAG_BARRIER;
599 	if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE))
600 		sc->vtblk_flags |= VTBLK_FLAG_WC_CONFIG;
601 }
602 
603 static int
604 vtblk_maximum_segments(struct vtblk_softc *sc,
605     struct virtio_blk_config *blkcfg)
606 {
607 	device_t dev;
608 	int nsegs;
609 
610 	dev = sc->vtblk_dev;
611 	nsegs = VTBLK_MIN_SEGMENTS;
612 
613 	if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) {
614 		nsegs += MIN(blkcfg->seg_max, MAXPHYS / PAGE_SIZE + 1);
615 		if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT)
616 			nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT);
617 	} else
618 		nsegs += 1;
619 
620 	return (nsegs);
621 }
622 
623 static int
624 vtblk_alloc_virtqueue(struct vtblk_softc *sc)
625 {
626 	device_t dev;
627 	struct vq_alloc_info vq_info;
628 
629 	dev = sc->vtblk_dev;
630 
631 	VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs,
632 	    vtblk_vq_intr, sc, &sc->vtblk_vq,
633 	    "%s request", device_get_nameunit(dev));
634 
635 	return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info));
636 }
637 
638 static void
639 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity)
640 {
641 	device_t dev;
642 	struct disk *dp;
643 	int error;
644 
645 	dev = sc->vtblk_dev;
646 	dp = sc->vtblk_disk;
647 
648 	dp->d_mediasize = new_capacity;
649 	if (bootverbose) {
650 		device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n",
651 		    (uintmax_t) dp->d_mediasize >> 20,
652 		    (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
653 		    dp->d_sectorsize);
654 	}
655 
656 	error = disk_resize(dp, M_NOWAIT);
657 	if (error) {
658 		device_printf(dev,
659 		    "disk_resize(9) failed, error: %d\n", error);
660 	}
661 }
662 
663 static void
664 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
665 {
666 	device_t dev;
667 	struct disk *dp;
668 
669 	dev = sc->vtblk_dev;
670 
671 	sc->vtblk_disk = dp = disk_alloc();
672 	dp->d_open = vtblk_open;
673 	dp->d_close = vtblk_close;
674 	dp->d_ioctl = vtblk_ioctl;
675 	dp->d_strategy = vtblk_strategy;
676 	dp->d_name = VTBLK_DISK_NAME;
677 	dp->d_unit = device_get_unit(dev);
678 	dp->d_drv1 = sc;
679 	dp->d_flags = DISKFLAG_CANFLUSHCACHE | DISKFLAG_UNMAPPED_BIO |
680 	    DISKFLAG_DIRECT_COMPLETION;
681 	dp->d_hba_vendor = virtio_get_vendor(dev);
682 	dp->d_hba_device = virtio_get_device(dev);
683 	dp->d_hba_subvendor = virtio_get_subvendor(dev);
684 	dp->d_hba_subdevice = virtio_get_subdevice(dev);
685 
686 	if ((sc->vtblk_flags & VTBLK_FLAG_READONLY) == 0)
687 		dp->d_dump = vtblk_dump;
688 
689 	/* Capacity is always in 512-byte units. */
690 	dp->d_mediasize = blkcfg->capacity * 512;
691 
692 	if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE))
693 		dp->d_sectorsize = blkcfg->blk_size;
694 	else
695 		dp->d_sectorsize = 512;
696 
697 	/*
698 	 * The VirtIO maximum I/O size is given in terms of segments.
699 	 * However, FreeBSD limits I/O size by logical buffer size, not
700 	 * by physically contiguous pages. Therefore, we have to assume
701 	 * no pages are contiguous. This may impose an artificially low
702 	 * maximum I/O size. But in practice, since QEMU advertises 128
703 	 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE,
704 	 * which is typically greater than MAXPHYS. Eventually we should
705 	 * just advertise MAXPHYS and split buffers that are too big.
706 	 *
707 	 * Note we must subtract one additional segment in case of non
708 	 * page aligned buffers.
709 	 */
710 	dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) *
711 	    PAGE_SIZE;
712 	if (dp->d_maxsize < PAGE_SIZE)
713 		dp->d_maxsize = PAGE_SIZE; /* XXX */
714 
715 	if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) {
716 		dp->d_fwsectors = blkcfg->geometry.sectors;
717 		dp->d_fwheads = blkcfg->geometry.heads;
718 	}
719 
720 	if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) &&
721 	    blkcfg->topology.physical_block_exp > 0) {
722 		dp->d_stripesize = dp->d_sectorsize *
723 		    (1 << blkcfg->topology.physical_block_exp);
724 		dp->d_stripeoffset = (dp->d_stripesize -
725 		    blkcfg->topology.alignment_offset * dp->d_sectorsize) %
726 		    dp->d_stripesize;
727 	}
728 
729 	if (vtblk_write_cache_enabled(sc, blkcfg) != 0)
730 		sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK;
731 	else
732 		sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH;
733 }
734 
735 static void
736 vtblk_create_disk(struct vtblk_softc *sc)
737 {
738 	struct disk *dp;
739 
740 	dp = sc->vtblk_disk;
741 
742 	vtblk_ident(sc);
743 
744 	device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n",
745 	    (uintmax_t) dp->d_mediasize >> 20,
746 	    (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
747 	    dp->d_sectorsize);
748 
749 	disk_create(dp, DISK_VERSION);
750 }
751 
752 static int
753 vtblk_request_prealloc(struct vtblk_softc *sc)
754 {
755 	struct vtblk_request *req;
756 	int i, nreqs;
757 
758 	nreqs = virtqueue_size(sc->vtblk_vq);
759 
760 	/*
761 	 * Preallocate sufficient requests to keep the virtqueue full. Each
762 	 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce
763 	 * the number allocated when indirect descriptors are not available.
764 	 */
765 	if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0)
766 		nreqs /= VTBLK_MIN_SEGMENTS;
767 
768 	for (i = 0; i < nreqs; i++) {
769 		req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT);
770 		if (req == NULL)
771 			return (ENOMEM);
772 
773 		MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1);
774 		MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1);
775 
776 		sc->vtblk_request_count++;
777 		vtblk_request_enqueue(sc, req);
778 	}
779 
780 	return (0);
781 }
782 
783 static void
784 vtblk_request_free(struct vtblk_softc *sc)
785 {
786 	struct vtblk_request *req;
787 
788 	MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready));
789 
790 	while ((req = vtblk_request_dequeue(sc)) != NULL) {
791 		sc->vtblk_request_count--;
792 		free(req, M_DEVBUF);
793 	}
794 
795 	KASSERT(sc->vtblk_request_count == 0,
796 	    ("%s: leaked %d requests", __func__, sc->vtblk_request_count));
797 }
798 
799 static struct vtblk_request *
800 vtblk_request_dequeue(struct vtblk_softc *sc)
801 {
802 	struct vtblk_request *req;
803 
804 	req = TAILQ_FIRST(&sc->vtblk_req_free);
805 	if (req != NULL) {
806 		TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link);
807 		bzero(req, sizeof(struct vtblk_request));
808 	}
809 
810 	return (req);
811 }
812 
813 static void
814 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req)
815 {
816 
817 	TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link);
818 }
819 
820 static struct vtblk_request *
821 vtblk_request_next_ready(struct vtblk_softc *sc)
822 {
823 	struct vtblk_request *req;
824 
825 	req = TAILQ_FIRST(&sc->vtblk_req_ready);
826 	if (req != NULL)
827 		TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link);
828 
829 	return (req);
830 }
831 
832 static void
833 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req)
834 {
835 
836 	/* NOTE: Currently, there will be at most one request in the queue. */
837 	TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link);
838 }
839 
840 static struct vtblk_request *
841 vtblk_request_next(struct vtblk_softc *sc)
842 {
843 	struct vtblk_request *req;
844 
845 	req = vtblk_request_next_ready(sc);
846 	if (req != NULL)
847 		return (req);
848 
849 	return (vtblk_request_bio(sc));
850 }
851 
852 static struct vtblk_request *
853 vtblk_request_bio(struct vtblk_softc *sc)
854 {
855 	struct bio_queue_head *bioq;
856 	struct vtblk_request *req;
857 	struct bio *bp;
858 
859 	bioq = &sc->vtblk_bioq;
860 
861 	if (bioq_first(bioq) == NULL)
862 		return (NULL);
863 
864 	req = vtblk_request_dequeue(sc);
865 	if (req == NULL)
866 		return (NULL);
867 
868 	bp = bioq_takefirst(bioq);
869 	req->vbr_bp = bp;
870 	req->vbr_ack = -1;
871 	req->vbr_hdr.ioprio = 1;
872 
873 	switch (bp->bio_cmd) {
874 	case BIO_FLUSH:
875 		req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
876 		break;
877 	case BIO_READ:
878 		req->vbr_hdr.type = VIRTIO_BLK_T_IN;
879 		req->vbr_hdr.sector = bp->bio_offset / 512;
880 		break;
881 	case BIO_WRITE:
882 		req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
883 		req->vbr_hdr.sector = bp->bio_offset / 512;
884 		break;
885 	default:
886 		panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd);
887 	}
888 
889 	if (bp->bio_flags & BIO_ORDERED)
890 		req->vbr_hdr.type |= VIRTIO_BLK_T_BARRIER;
891 
892 	return (req);
893 }
894 
895 static int
896 vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req)
897 {
898 	struct virtqueue *vq;
899 	struct sglist *sg;
900 	struct bio *bp;
901 	int ordered, readable, writable, error;
902 
903 	vq = sc->vtblk_vq;
904 	sg = sc->vtblk_sglist;
905 	bp = req->vbr_bp;
906 	ordered = 0;
907 	writable = 0;
908 
909 	/*
910 	 * Some hosts (such as bhyve) do not implement the barrier feature,
911 	 * so we emulate it in the driver by allowing the barrier request
912 	 * to be the only one in flight.
913 	 */
914 	if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) {
915 		if (sc->vtblk_req_ordered != NULL)
916 			return (EBUSY);
917 		if (bp->bio_flags & BIO_ORDERED) {
918 			if (!virtqueue_empty(vq))
919 				return (EBUSY);
920 			ordered = 1;
921 			req->vbr_hdr.type &= ~VIRTIO_BLK_T_BARRIER;
922 		}
923 	}
924 
925 	sglist_reset(sg);
926 	sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr));
927 
928 	if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
929 		error = sglist_append_bio(sg, bp);
930 		if (error || sg->sg_nseg == sg->sg_maxseg) {
931 			panic("%s: bio %p data buffer too big %d",
932 			    __func__, bp, error);
933 		}
934 
935 		/* BIO_READ means the host writes into our buffer. */
936 		if (bp->bio_cmd == BIO_READ)
937 			writable = sg->sg_nseg - 1;
938 	}
939 
940 	writable++;
941 	sglist_append(sg, &req->vbr_ack, sizeof(uint8_t));
942 	readable = sg->sg_nseg - writable;
943 
944 	error = virtqueue_enqueue(vq, req, sg, readable, writable);
945 	if (error == 0 && ordered)
946 		sc->vtblk_req_ordered = req;
947 
948 	return (error);
949 }
950 
951 static int
952 vtblk_request_error(struct vtblk_request *req)
953 {
954 	int error;
955 
956 	switch (req->vbr_ack) {
957 	case VIRTIO_BLK_S_OK:
958 		error = 0;
959 		break;
960 	case VIRTIO_BLK_S_UNSUPP:
961 		error = ENOTSUP;
962 		break;
963 	default:
964 		error = EIO;
965 		break;
966 	}
967 
968 	return (error);
969 }
970 
971 static void
972 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue)
973 {
974 	struct vtblk_request *req;
975 	struct bio *bp;
976 
977 	while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) {
978 		if (sc->vtblk_req_ordered != NULL) {
979 			MPASS(sc->vtblk_req_ordered == req);
980 			sc->vtblk_req_ordered = NULL;
981 		}
982 
983 		bp = req->vbr_bp;
984 		bp->bio_error = vtblk_request_error(req);
985 		TAILQ_INSERT_TAIL(queue, bp, bio_queue);
986 
987 		vtblk_request_enqueue(sc, req);
988 	}
989 }
990 
991 static void
992 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue)
993 {
994 	struct bio *bp, *tmp;
995 
996 	TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) {
997 		if (bp->bio_error != 0)
998 			disk_err(bp, "hard error", -1, 1);
999 		vtblk_bio_done(sc, bp, bp->bio_error);
1000 	}
1001 }
1002 
1003 static void
1004 vtblk_drain_vq(struct vtblk_softc *sc)
1005 {
1006 	struct virtqueue *vq;
1007 	struct vtblk_request *req;
1008 	int last;
1009 
1010 	vq = sc->vtblk_vq;
1011 	last = 0;
1012 
1013 	while ((req = virtqueue_drain(vq, &last)) != NULL) {
1014 		vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1015 		vtblk_request_enqueue(sc, req);
1016 	}
1017 
1018 	sc->vtblk_req_ordered = NULL;
1019 	KASSERT(virtqueue_empty(vq), ("virtqueue not empty"));
1020 }
1021 
1022 static void
1023 vtblk_drain(struct vtblk_softc *sc)
1024 {
1025 	struct bio_queue queue;
1026 	struct bio_queue_head *bioq;
1027 	struct vtblk_request *req;
1028 	struct bio *bp;
1029 
1030 	bioq = &sc->vtblk_bioq;
1031 	TAILQ_INIT(&queue);
1032 
1033 	if (sc->vtblk_vq != NULL) {
1034 		vtblk_queue_completed(sc, &queue);
1035 		vtblk_done_completed(sc, &queue);
1036 
1037 		vtblk_drain_vq(sc);
1038 	}
1039 
1040 	while ((req = vtblk_request_next_ready(sc)) != NULL) {
1041 		vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1042 		vtblk_request_enqueue(sc, req);
1043 	}
1044 
1045 	while (bioq_first(bioq) != NULL) {
1046 		bp = bioq_takefirst(bioq);
1047 		vtblk_bio_done(sc, bp, ENXIO);
1048 	}
1049 
1050 	vtblk_request_free(sc);
1051 }
1052 
1053 static void
1054 vtblk_startio(struct vtblk_softc *sc)
1055 {
1056 	struct virtqueue *vq;
1057 	struct vtblk_request *req;
1058 	int enq;
1059 
1060 	VTBLK_LOCK_ASSERT(sc);
1061 	vq = sc->vtblk_vq;
1062 	enq = 0;
1063 
1064 	if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1065 		return;
1066 
1067 	while (!virtqueue_full(vq)) {
1068 		req = vtblk_request_next(sc);
1069 		if (req == NULL)
1070 			break;
1071 
1072 		if (vtblk_request_execute(sc, req) != 0) {
1073 			vtblk_request_requeue_ready(sc, req);
1074 			break;
1075 		}
1076 
1077 		enq++;
1078 	}
1079 
1080 	if (enq > 0)
1081 		virtqueue_notify(vq);
1082 }
1083 
1084 static void
1085 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error)
1086 {
1087 
1088 	/* Because of GEOM direct dispatch, we cannot hold any locks. */
1089 	if (sc != NULL)
1090 		VTBLK_LOCK_ASSERT_NOTOWNED(sc);
1091 
1092 	if (error) {
1093 		bp->bio_resid = bp->bio_bcount;
1094 		bp->bio_error = error;
1095 		bp->bio_flags |= BIO_ERROR;
1096 	}
1097 
1098 	biodone(bp);
1099 }
1100 
1101 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg)			\
1102 	if (virtio_with_feature(_dev, _feature)) {			\
1103 		virtio_read_device_config(_dev,				\
1104 		    offsetof(struct virtio_blk_config, _field),		\
1105 		    &(_cfg)->_field, sizeof((_cfg)->_field));		\
1106 	}
1107 
1108 static void
1109 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
1110 {
1111 	device_t dev;
1112 
1113 	dev = sc->vtblk_dev;
1114 
1115 	bzero(blkcfg, sizeof(struct virtio_blk_config));
1116 
1117 	/* The capacity is always available. */
1118 	virtio_read_device_config(dev, offsetof(struct virtio_blk_config,
1119 	    capacity), &blkcfg->capacity, sizeof(blkcfg->capacity));
1120 
1121 	/* Read the configuration if the feature was negotiated. */
1122 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg);
1123 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg);
1124 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, geometry, blkcfg);
1125 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg);
1126 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, topology, blkcfg);
1127 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, writeback, blkcfg);
1128 }
1129 
1130 #undef VTBLK_GET_CONFIG
1131 
1132 static void
1133 vtblk_ident(struct vtblk_softc *sc)
1134 {
1135 	struct bio buf;
1136 	struct disk *dp;
1137 	struct vtblk_request *req;
1138 	int len, error;
1139 
1140 	dp = sc->vtblk_disk;
1141 	len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE);
1142 
1143 	if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0)
1144 		return;
1145 
1146 	req = vtblk_request_dequeue(sc);
1147 	if (req == NULL)
1148 		return;
1149 
1150 	req->vbr_ack = -1;
1151 	req->vbr_hdr.type = VIRTIO_BLK_T_GET_ID;
1152 	req->vbr_hdr.ioprio = 1;
1153 	req->vbr_hdr.sector = 0;
1154 
1155 	req->vbr_bp = &buf;
1156 	g_reset_bio(&buf);
1157 
1158 	buf.bio_cmd = BIO_READ;
1159 	buf.bio_data = dp->d_ident;
1160 	buf.bio_bcount = len;
1161 
1162 	VTBLK_LOCK(sc);
1163 	error = vtblk_poll_request(sc, req);
1164 	VTBLK_UNLOCK(sc);
1165 
1166 	vtblk_request_enqueue(sc, req);
1167 
1168 	if (error) {
1169 		device_printf(sc->vtblk_dev,
1170 		    "error getting device identifier: %d\n", error);
1171 	}
1172 }
1173 
1174 static int
1175 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req)
1176 {
1177 	struct virtqueue *vq;
1178 	int error;
1179 
1180 	vq = sc->vtblk_vq;
1181 
1182 	if (!virtqueue_empty(vq))
1183 		return (EBUSY);
1184 
1185 	error = vtblk_request_execute(sc, req);
1186 	if (error)
1187 		return (error);
1188 
1189 	virtqueue_notify(vq);
1190 	virtqueue_poll(vq, NULL);
1191 
1192 	error = vtblk_request_error(req);
1193 	if (error && bootverbose) {
1194 		device_printf(sc->vtblk_dev,
1195 		    "%s: IO error: %d\n", __func__, error);
1196 	}
1197 
1198 	return (error);
1199 }
1200 
1201 static int
1202 vtblk_quiesce(struct vtblk_softc *sc)
1203 {
1204 	int error;
1205 
1206 	VTBLK_LOCK_ASSERT(sc);
1207 	error = 0;
1208 
1209 	while (!virtqueue_empty(sc->vtblk_vq)) {
1210 		if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq",
1211 		    VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) {
1212 			error = EBUSY;
1213 			break;
1214 		}
1215 	}
1216 
1217 	return (error);
1218 }
1219 
1220 static void
1221 vtblk_vq_intr(void *xsc)
1222 {
1223 	struct vtblk_softc *sc;
1224 	struct virtqueue *vq;
1225 	struct bio_queue queue;
1226 
1227 	sc = xsc;
1228 	vq = sc->vtblk_vq;
1229 	TAILQ_INIT(&queue);
1230 
1231 	VTBLK_LOCK(sc);
1232 
1233 again:
1234 	if (sc->vtblk_flags & VTBLK_FLAG_DETACH)
1235 		goto out;
1236 
1237 	vtblk_queue_completed(sc, &queue);
1238 	vtblk_startio(sc);
1239 
1240 	if (virtqueue_enable_intr(vq) != 0) {
1241 		virtqueue_disable_intr(vq);
1242 		goto again;
1243 	}
1244 
1245 	if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1246 		wakeup(&sc->vtblk_vq);
1247 
1248 out:
1249 	VTBLK_UNLOCK(sc);
1250 	vtblk_done_completed(sc, &queue);
1251 }
1252 
1253 static void
1254 vtblk_stop(struct vtblk_softc *sc)
1255 {
1256 
1257 	virtqueue_disable_intr(sc->vtblk_vq);
1258 	virtio_stop(sc->vtblk_dev);
1259 }
1260 
1261 static void
1262 vtblk_dump_quiesce(struct vtblk_softc *sc)
1263 {
1264 
1265 	/*
1266 	 * Spin here until all the requests in-flight at the time of the
1267 	 * dump are completed and queued. The queued requests will be
1268 	 * biodone'd once the dump is finished.
1269 	 */
1270 	while (!virtqueue_empty(sc->vtblk_vq))
1271 		vtblk_queue_completed(sc, &sc->vtblk_dump_queue);
1272 }
1273 
1274 static int
1275 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset,
1276     size_t length)
1277 {
1278 	struct bio buf;
1279 	struct vtblk_request *req;
1280 
1281 	req = &sc->vtblk_dump_request;
1282 	req->vbr_ack = -1;
1283 	req->vbr_hdr.type = VIRTIO_BLK_T_OUT;
1284 	req->vbr_hdr.ioprio = 1;
1285 	req->vbr_hdr.sector = offset / 512;
1286 
1287 	req->vbr_bp = &buf;
1288 	g_reset_bio(&buf);
1289 
1290 	buf.bio_cmd = BIO_WRITE;
1291 	buf.bio_data = virtual;
1292 	buf.bio_bcount = length;
1293 
1294 	return (vtblk_poll_request(sc, req));
1295 }
1296 
1297 static int
1298 vtblk_dump_flush(struct vtblk_softc *sc)
1299 {
1300 	struct bio buf;
1301 	struct vtblk_request *req;
1302 
1303 	req = &sc->vtblk_dump_request;
1304 	req->vbr_ack = -1;
1305 	req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH;
1306 	req->vbr_hdr.ioprio = 1;
1307 	req->vbr_hdr.sector = 0;
1308 
1309 	req->vbr_bp = &buf;
1310 	g_reset_bio(&buf);
1311 
1312 	buf.bio_cmd = BIO_FLUSH;
1313 
1314 	return (vtblk_poll_request(sc, req));
1315 }
1316 
1317 static void
1318 vtblk_dump_complete(struct vtblk_softc *sc)
1319 {
1320 
1321 	vtblk_dump_flush(sc);
1322 
1323 	VTBLK_UNLOCK(sc);
1324 	vtblk_done_completed(sc, &sc->vtblk_dump_queue);
1325 	VTBLK_LOCK(sc);
1326 }
1327 
1328 static void
1329 vtblk_set_write_cache(struct vtblk_softc *sc, int wc)
1330 {
1331 
1332 	/* Set either writeback (1) or writethrough (0) mode. */
1333 	virtio_write_dev_config_1(sc->vtblk_dev,
1334 	    offsetof(struct virtio_blk_config, writeback), wc);
1335 }
1336 
1337 static int
1338 vtblk_write_cache_enabled(struct vtblk_softc *sc,
1339     struct virtio_blk_config *blkcfg)
1340 {
1341 	int wc;
1342 
1343 	if (sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) {
1344 		wc = vtblk_tunable_int(sc, "writecache_mode",
1345 		    vtblk_writecache_mode);
1346 		if (wc >= 0 && wc < VTBLK_CACHE_MAX)
1347 			vtblk_set_write_cache(sc, wc);
1348 		else
1349 			wc = blkcfg->writeback;
1350 	} else
1351 		wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_WCE);
1352 
1353 	return (wc);
1354 }
1355 
1356 static int
1357 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS)
1358 {
1359 	struct vtblk_softc *sc;
1360 	int wc, error;
1361 
1362 	sc = oidp->oid_arg1;
1363 	wc = sc->vtblk_write_cache;
1364 
1365 	error = sysctl_handle_int(oidp, &wc, 0, req);
1366 	if (error || req->newptr == NULL)
1367 		return (error);
1368 	if ((sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) == 0)
1369 		return (EPERM);
1370 	if (wc < 0 || wc >= VTBLK_CACHE_MAX)
1371 		return (EINVAL);
1372 
1373 	VTBLK_LOCK(sc);
1374 	sc->vtblk_write_cache = wc;
1375 	vtblk_set_write_cache(sc, sc->vtblk_write_cache);
1376 	VTBLK_UNLOCK(sc);
1377 
1378 	return (0);
1379 }
1380 
1381 static void
1382 vtblk_setup_sysctl(struct vtblk_softc *sc)
1383 {
1384 	device_t dev;
1385 	struct sysctl_ctx_list *ctx;
1386 	struct sysctl_oid *tree;
1387 	struct sysctl_oid_list *child;
1388 
1389 	dev = sc->vtblk_dev;
1390 	ctx = device_get_sysctl_ctx(dev);
1391 	tree = device_get_sysctl_tree(dev);
1392 	child = SYSCTL_CHILDREN(tree);
1393 
1394 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode",
1395 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, vtblk_write_cache_sysctl,
1396 	    "I", "Write cache mode (writethrough (0) or writeback (1))");
1397 }
1398 
1399 static int
1400 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def)
1401 {
1402 	char path[64];
1403 
1404 	snprintf(path, sizeof(path),
1405 	    "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob);
1406 	TUNABLE_INT_FETCH(path, &def);
1407 
1408 	return (def);
1409 }
1410