xref: /freebsd/sys/dev/virtio/block/virtio_blk.c (revision 5ca8e32633c4ffbbcd6762e5888b6a4ba0708c6c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/bio.h>
35 #include <sys/malloc.h>
36 #include <sys/module.h>
37 #include <sys/msan.h>
38 #include <sys/sglist.h>
39 #include <sys/sysctl.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/queue.h>
43 
44 #include <geom/geom.h>
45 #include <geom/geom_disk.h>
46 
47 #include <machine/bus.h>
48 #include <machine/resource.h>
49 #include <sys/bus.h>
50 #include <sys/rman.h>
51 
52 #include <dev/virtio/virtio.h>
53 #include <dev/virtio/virtqueue.h>
54 #include <dev/virtio/block/virtio_blk.h>
55 
56 #include "virtio_if.h"
57 
58 struct vtblk_request {
59 	struct vtblk_softc		*vbr_sc;
60 	bus_dmamap_t			 vbr_mapp;
61 
62 	/* Fields after this point are zeroed for each request. */
63 	struct virtio_blk_outhdr	 vbr_hdr;
64 	struct bio			*vbr_bp;
65 	uint8_t				 vbr_ack;
66 	uint8_t				 vbr_requeue_on_error;
67 	uint8_t				 vbr_busdma_wait;
68 	int				 vbr_error;
69 	TAILQ_ENTRY(vtblk_request)	 vbr_link;
70 };
71 
72 enum vtblk_cache_mode {
73 	VTBLK_CACHE_WRITETHROUGH,
74 	VTBLK_CACHE_WRITEBACK,
75 	VTBLK_CACHE_MAX
76 };
77 
78 struct vtblk_softc {
79 	device_t		 vtblk_dev;
80 	struct mtx		 vtblk_mtx;
81 	uint64_t		 vtblk_features;
82 	uint32_t		 vtblk_flags;
83 #define VTBLK_FLAG_INDIRECT	0x0001
84 #define VTBLK_FLAG_DETACH	0x0002
85 #define VTBLK_FLAG_SUSPEND	0x0004
86 #define VTBLK_FLAG_BARRIER	0x0008
87 #define VTBLK_FLAG_WCE_CONFIG	0x0010
88 #define VTBLK_FLAG_BUSDMA_WAIT	0x0020
89 #define VTBLK_FLAG_BUSDMA_ALIGN	0x0040
90 
91 	struct virtqueue	*vtblk_vq;
92 	struct sglist		*vtblk_sglist;
93 	bus_dma_tag_t		 vtblk_dmat;
94 	struct disk		*vtblk_disk;
95 
96 	struct bio_queue_head	 vtblk_bioq;
97 	TAILQ_HEAD(, vtblk_request)
98 				 vtblk_req_free;
99 	TAILQ_HEAD(, vtblk_request)
100 				 vtblk_req_ready;
101 	struct vtblk_request	*vtblk_req_ordered;
102 
103 	int			 vtblk_max_nsegs;
104 	int			 vtblk_request_count;
105 	enum vtblk_cache_mode	 vtblk_write_cache;
106 
107 	struct bio_queue	 vtblk_dump_queue;
108 	struct vtblk_request	 vtblk_dump_request;
109 };
110 
111 static struct virtio_feature_desc vtblk_feature_desc[] = {
112 	{ VIRTIO_BLK_F_BARRIER,		"HostBarrier"	},
113 	{ VIRTIO_BLK_F_SIZE_MAX,	"MaxSegSize"	},
114 	{ VIRTIO_BLK_F_SEG_MAX,		"MaxNumSegs"	},
115 	{ VIRTIO_BLK_F_GEOMETRY,	"DiskGeometry"	},
116 	{ VIRTIO_BLK_F_RO,		"ReadOnly"	},
117 	{ VIRTIO_BLK_F_BLK_SIZE,	"BlockSize"	},
118 	{ VIRTIO_BLK_F_SCSI,		"SCSICmds"	},
119 	{ VIRTIO_BLK_F_FLUSH,		"FlushCmd"	},
120 	{ VIRTIO_BLK_F_TOPOLOGY,	"Topology"	},
121 	{ VIRTIO_BLK_F_CONFIG_WCE,	"ConfigWCE"	},
122 	{ VIRTIO_BLK_F_MQ,		"Multiqueue"	},
123 	{ VIRTIO_BLK_F_DISCARD,		"Discard"	},
124 	{ VIRTIO_BLK_F_WRITE_ZEROES,	"WriteZeros"	},
125 
126 	{ 0, NULL }
127 };
128 
129 static int	vtblk_modevent(module_t, int, void *);
130 
131 static int	vtblk_probe(device_t);
132 static int	vtblk_attach(device_t);
133 static int	vtblk_detach(device_t);
134 static int	vtblk_suspend(device_t);
135 static int	vtblk_resume(device_t);
136 static int	vtblk_shutdown(device_t);
137 static int	vtblk_attach_completed(device_t);
138 static int	vtblk_config_change(device_t);
139 
140 static int	vtblk_open(struct disk *);
141 static int	vtblk_close(struct disk *);
142 static int	vtblk_ioctl(struct disk *, u_long, void *, int,
143 		    struct thread *);
144 static int	vtblk_dump(void *, void *, off_t, size_t);
145 static void	vtblk_strategy(struct bio *);
146 
147 static int	vtblk_negotiate_features(struct vtblk_softc *);
148 static int	vtblk_setup_features(struct vtblk_softc *);
149 static int	vtblk_maximum_segments(struct vtblk_softc *,
150 		    struct virtio_blk_config *);
151 static int	vtblk_alloc_virtqueue(struct vtblk_softc *);
152 static void	vtblk_resize_disk(struct vtblk_softc *, uint64_t);
153 static void	vtblk_alloc_disk(struct vtblk_softc *,
154 		    struct virtio_blk_config *);
155 static void	vtblk_create_disk(struct vtblk_softc *);
156 
157 static int	vtblk_request_prealloc(struct vtblk_softc *);
158 static void	vtblk_request_free(struct vtblk_softc *);
159 static struct vtblk_request *
160 		vtblk_request_dequeue(struct vtblk_softc *);
161 static void	vtblk_request_enqueue(struct vtblk_softc *,
162 		    struct vtblk_request *);
163 static struct vtblk_request *
164 		vtblk_request_next_ready(struct vtblk_softc *);
165 static void	vtblk_request_requeue_ready(struct vtblk_softc *,
166 		    struct vtblk_request *);
167 static struct vtblk_request *
168 		vtblk_request_next(struct vtblk_softc *);
169 static struct vtblk_request *
170 		vtblk_request_bio(struct vtblk_softc *);
171 static int	vtblk_request_execute(struct vtblk_request *, int);
172 static void	vtblk_request_execute_cb(void *,
173 		    bus_dma_segment_t *, int, int);
174 static int	vtblk_request_error(struct vtblk_request *);
175 
176 static void	vtblk_queue_completed(struct vtblk_softc *,
177 		    struct bio_queue *);
178 static void	vtblk_done_completed(struct vtblk_softc *,
179 		    struct bio_queue *);
180 static void	vtblk_drain_vq(struct vtblk_softc *);
181 static void	vtblk_drain(struct vtblk_softc *);
182 
183 static void	vtblk_startio(struct vtblk_softc *);
184 static void	vtblk_bio_done(struct vtblk_softc *, struct bio *, int);
185 
186 static void	vtblk_read_config(struct vtblk_softc *,
187 		    struct virtio_blk_config *);
188 static void	vtblk_ident(struct vtblk_softc *);
189 static int	vtblk_poll_request(struct vtblk_softc *,
190 		    struct vtblk_request *);
191 static int	vtblk_quiesce(struct vtblk_softc *);
192 static void	vtblk_vq_intr(void *);
193 static void	vtblk_stop(struct vtblk_softc *);
194 
195 static void	vtblk_dump_quiesce(struct vtblk_softc *);
196 static int	vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t);
197 static int	vtblk_dump_flush(struct vtblk_softc *);
198 static void	vtblk_dump_complete(struct vtblk_softc *);
199 
200 static void	vtblk_set_write_cache(struct vtblk_softc *, int);
201 static int	vtblk_write_cache_enabled(struct vtblk_softc *sc,
202 		    struct virtio_blk_config *);
203 static int	vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS);
204 
205 static void	vtblk_setup_sysctl(struct vtblk_softc *);
206 static int	vtblk_tunable_int(struct vtblk_softc *, const char *, int);
207 
208 #define vtblk_modern(_sc) (((_sc)->vtblk_features & VIRTIO_F_VERSION_1) != 0)
209 #define vtblk_htog16(_sc, _val)	virtio_htog16(vtblk_modern(_sc), _val)
210 #define vtblk_htog32(_sc, _val)	virtio_htog32(vtblk_modern(_sc), _val)
211 #define vtblk_htog64(_sc, _val)	virtio_htog64(vtblk_modern(_sc), _val)
212 #define vtblk_gtoh16(_sc, _val)	virtio_gtoh16(vtblk_modern(_sc), _val)
213 #define vtblk_gtoh32(_sc, _val)	virtio_gtoh32(vtblk_modern(_sc), _val)
214 #define vtblk_gtoh64(_sc, _val)	virtio_gtoh64(vtblk_modern(_sc), _val)
215 
216 /* Tunables. */
217 static int vtblk_no_ident = 0;
218 TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident);
219 static int vtblk_writecache_mode = -1;
220 TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode);
221 
222 #define VTBLK_COMMON_FEATURES \
223     (VIRTIO_BLK_F_SIZE_MAX		| \
224      VIRTIO_BLK_F_SEG_MAX		| \
225      VIRTIO_BLK_F_GEOMETRY		| \
226      VIRTIO_BLK_F_RO			| \
227      VIRTIO_BLK_F_BLK_SIZE		| \
228      VIRTIO_BLK_F_FLUSH			| \
229      VIRTIO_BLK_F_TOPOLOGY		| \
230      VIRTIO_BLK_F_CONFIG_WCE		| \
231      VIRTIO_BLK_F_DISCARD		| \
232      VIRTIO_RING_F_INDIRECT_DESC)
233 
234 #define VTBLK_MODERN_FEATURES	(VTBLK_COMMON_FEATURES)
235 #define VTBLK_LEGACY_FEATURES	(VIRTIO_BLK_F_BARRIER | VTBLK_COMMON_FEATURES)
236 
237 #define VTBLK_MTX(_sc)		&(_sc)->vtblk_mtx
238 #define VTBLK_LOCK_INIT(_sc, _name) \
239 				mtx_init(VTBLK_MTX((_sc)), (_name), \
240 				    "VirtIO Block Lock", MTX_DEF)
241 #define VTBLK_LOCK(_sc)		mtx_lock(VTBLK_MTX((_sc)))
242 #define VTBLK_UNLOCK(_sc)	mtx_unlock(VTBLK_MTX((_sc)))
243 #define VTBLK_LOCK_DESTROY(_sc)	mtx_destroy(VTBLK_MTX((_sc)))
244 #define VTBLK_LOCK_ASSERT(_sc)	mtx_assert(VTBLK_MTX((_sc)), MA_OWNED)
245 #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \
246 				mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED)
247 
248 #define VTBLK_DISK_NAME		"vtbd"
249 #define VTBLK_QUIESCE_TIMEOUT	(30 * hz)
250 #define VTBLK_BSIZE		512
251 
252 /*
253  * Each block request uses at least two segments - one for the header
254  * and one for the status.
255  */
256 #define VTBLK_MIN_SEGMENTS	2
257 
258 static device_method_t vtblk_methods[] = {
259 	/* Device methods. */
260 	DEVMETHOD(device_probe,		vtblk_probe),
261 	DEVMETHOD(device_attach,	vtblk_attach),
262 	DEVMETHOD(device_detach,	vtblk_detach),
263 	DEVMETHOD(device_suspend,	vtblk_suspend),
264 	DEVMETHOD(device_resume,	vtblk_resume),
265 	DEVMETHOD(device_shutdown,	vtblk_shutdown),
266 
267 	/* VirtIO methods. */
268 	DEVMETHOD(virtio_attach_completed, vtblk_attach_completed),
269 	DEVMETHOD(virtio_config_change,	vtblk_config_change),
270 
271 	DEVMETHOD_END
272 };
273 
274 static driver_t vtblk_driver = {
275 	"vtblk",
276 	vtblk_methods,
277 	sizeof(struct vtblk_softc)
278 };
279 
280 VIRTIO_DRIVER_MODULE(virtio_blk, vtblk_driver, vtblk_modevent, NULL);
281 MODULE_VERSION(virtio_blk, 1);
282 MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1);
283 
284 VIRTIO_SIMPLE_PNPINFO(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter");
285 
286 static int
287 vtblk_modevent(module_t mod, int type, void *unused)
288 {
289 	int error;
290 
291 	error = 0;
292 
293 	switch (type) {
294 	case MOD_LOAD:
295 	case MOD_QUIESCE:
296 	case MOD_UNLOAD:
297 	case MOD_SHUTDOWN:
298 		break;
299 	default:
300 		error = EOPNOTSUPP;
301 		break;
302 	}
303 
304 	return (error);
305 }
306 
307 static int
308 vtblk_probe(device_t dev)
309 {
310 	return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk));
311 }
312 
313 static int
314 vtblk_attach(device_t dev)
315 {
316 	struct vtblk_softc *sc;
317 	struct virtio_blk_config blkcfg;
318 	int error;
319 
320 	sc = device_get_softc(dev);
321 	sc->vtblk_dev = dev;
322 	virtio_set_feature_desc(dev, vtblk_feature_desc);
323 
324 	VTBLK_LOCK_INIT(sc, device_get_nameunit(dev));
325 	bioq_init(&sc->vtblk_bioq);
326 	TAILQ_INIT(&sc->vtblk_dump_queue);
327 	TAILQ_INIT(&sc->vtblk_req_free);
328 	TAILQ_INIT(&sc->vtblk_req_ready);
329 
330 	vtblk_setup_sysctl(sc);
331 
332 	error = vtblk_setup_features(sc);
333 	if (error) {
334 		device_printf(dev, "cannot setup features\n");
335 		goto fail;
336 	}
337 
338 	vtblk_read_config(sc, &blkcfg);
339 
340 	/*
341 	 * With the current sglist(9) implementation, it is not easy
342 	 * for us to support a maximum segment size as adjacent
343 	 * segments are coalesced. For now, just make sure it's larger
344 	 * than the maximum supported transfer size.
345 	 */
346 	if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) {
347 		if (blkcfg.size_max < maxphys) {
348 			error = ENOTSUP;
349 			device_printf(dev, "host requires unsupported "
350 			    "maximum segment size feature\n");
351 			goto fail;
352 		}
353 	}
354 
355 	sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg);
356 	if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) {
357 		error = EINVAL;
358 		device_printf(dev, "fewer than minimum number of segments "
359 		    "allowed: %d\n", sc->vtblk_max_nsegs);
360 		goto fail;
361 	}
362 
363 	sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT);
364 	if (sc->vtblk_sglist == NULL) {
365 		error = ENOMEM;
366 		device_printf(dev, "cannot allocate sglist\n");
367 		goto fail;
368 	}
369 
370 	/*
371 	 * If vtblk_max_nsegs == VTBLK_MIN_SEGMENTS + 1, the device only
372 	 * supports a single data segment; in that case we need busdma to
373 	 * align to a page boundary so we can send a *contiguous* page size
374 	 * request to the host.
375 	 */
376 	if (sc->vtblk_max_nsegs == VTBLK_MIN_SEGMENTS + 1)
377 		sc->vtblk_flags |= VTBLK_FLAG_BUSDMA_ALIGN;
378 	error = bus_dma_tag_create(
379 	    bus_get_dma_tag(dev),			/* parent */
380 	    (sc->vtblk_flags & VTBLK_FLAG_BUSDMA_ALIGN) ? PAGE_SIZE : 1,
381 	    0,						/* boundary */
382 	    BUS_SPACE_MAXADDR,				/* lowaddr */
383 	    BUS_SPACE_MAXADDR,				/* highaddr */
384 	    NULL, NULL,					/* filter, filterarg */
385 	    maxphys,					/* max request size */
386 	    sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS,	/* max # segments */
387 	    maxphys,					/* maxsegsize */
388 	    0,						/* flags */
389 	    busdma_lock_mutex,				/* lockfunc */
390 	    &sc->vtblk_mtx,				/* lockarg */
391 	    &sc->vtblk_dmat);
392 	if (error) {
393 		device_printf(dev, "cannot create bus dma tag\n");
394 		goto fail;
395 	}
396 
397 #ifdef __powerpc__
398 	/*
399 	 * Virtio uses physical addresses rather than bus addresses, so we
400 	 * need to ask busdma to skip the iommu physical->bus mapping.  At
401 	 * present, this is only a thing on the powerpc architectures.
402 	 */
403 	bus_dma_tag_set_iommu(sc->vtblk_dmat, NULL, NULL);
404 #endif
405 
406 	error = vtblk_alloc_virtqueue(sc);
407 	if (error) {
408 		device_printf(dev, "cannot allocate virtqueue\n");
409 		goto fail;
410 	}
411 
412 	error = vtblk_request_prealloc(sc);
413 	if (error) {
414 		device_printf(dev, "cannot preallocate requests\n");
415 		goto fail;
416 	}
417 
418 	vtblk_alloc_disk(sc, &blkcfg);
419 
420 	error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY);
421 	if (error) {
422 		device_printf(dev, "cannot setup virtqueue interrupt\n");
423 		goto fail;
424 	}
425 
426 	virtqueue_enable_intr(sc->vtblk_vq);
427 
428 fail:
429 	if (error)
430 		vtblk_detach(dev);
431 
432 	return (error);
433 }
434 
435 static int
436 vtblk_detach(device_t dev)
437 {
438 	struct vtblk_softc *sc;
439 
440 	sc = device_get_softc(dev);
441 
442 	VTBLK_LOCK(sc);
443 	sc->vtblk_flags |= VTBLK_FLAG_DETACH;
444 	if (device_is_attached(dev))
445 		vtblk_stop(sc);
446 	VTBLK_UNLOCK(sc);
447 
448 	vtblk_drain(sc);
449 
450 	if (sc->vtblk_disk != NULL) {
451 		disk_destroy(sc->vtblk_disk);
452 		sc->vtblk_disk = NULL;
453 	}
454 
455 	if (sc->vtblk_dmat != NULL) {
456 		bus_dma_tag_destroy(sc->vtblk_dmat);
457 		sc->vtblk_dmat = NULL;
458 	}
459 
460 	if (sc->vtblk_sglist != NULL) {
461 		sglist_free(sc->vtblk_sglist);
462 		sc->vtblk_sglist = NULL;
463 	}
464 
465 	VTBLK_LOCK_DESTROY(sc);
466 
467 	return (0);
468 }
469 
470 static int
471 vtblk_suspend(device_t dev)
472 {
473 	struct vtblk_softc *sc;
474 	int error;
475 
476 	sc = device_get_softc(dev);
477 
478 	VTBLK_LOCK(sc);
479 	sc->vtblk_flags |= VTBLK_FLAG_SUSPEND;
480 	/* XXX BMV: virtio_stop(), etc needed here? */
481 	error = vtblk_quiesce(sc);
482 	if (error)
483 		sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
484 	VTBLK_UNLOCK(sc);
485 
486 	return (error);
487 }
488 
489 static int
490 vtblk_resume(device_t dev)
491 {
492 	struct vtblk_softc *sc;
493 
494 	sc = device_get_softc(dev);
495 
496 	VTBLK_LOCK(sc);
497 	/* XXX BMV: virtio_reinit(), etc needed here? */
498 	sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND;
499 	vtblk_startio(sc);
500 	VTBLK_UNLOCK(sc);
501 
502 	return (0);
503 }
504 
505 static int
506 vtblk_shutdown(device_t dev)
507 {
508 
509 	return (0);
510 }
511 
512 static int
513 vtblk_attach_completed(device_t dev)
514 {
515 	struct vtblk_softc *sc;
516 
517 	sc = device_get_softc(dev);
518 
519 	/*
520 	 * Create disk after attach as VIRTIO_BLK_T_GET_ID can only be
521 	 * processed after the device acknowledged
522 	 * VIRTIO_CONFIG_STATUS_DRIVER_OK.
523 	 */
524 	vtblk_create_disk(sc);
525 	return (0);
526 }
527 
528 static int
529 vtblk_config_change(device_t dev)
530 {
531 	struct vtblk_softc *sc;
532 	struct virtio_blk_config blkcfg;
533 	uint64_t capacity;
534 
535 	sc = device_get_softc(dev);
536 
537 	vtblk_read_config(sc, &blkcfg);
538 
539 	/* Capacity is always in 512-byte units. */
540 	capacity = blkcfg.capacity * VTBLK_BSIZE;
541 
542 	if (sc->vtblk_disk->d_mediasize != capacity)
543 		vtblk_resize_disk(sc, capacity);
544 
545 	return (0);
546 }
547 
548 static int
549 vtblk_open(struct disk *dp)
550 {
551 	struct vtblk_softc *sc;
552 
553 	if ((sc = dp->d_drv1) == NULL)
554 		return (ENXIO);
555 
556 	return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0);
557 }
558 
559 static int
560 vtblk_close(struct disk *dp)
561 {
562 	struct vtblk_softc *sc;
563 
564 	if ((sc = dp->d_drv1) == NULL)
565 		return (ENXIO);
566 
567 	return (0);
568 }
569 
570 static int
571 vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag,
572     struct thread *td)
573 {
574 	struct vtblk_softc *sc;
575 
576 	if ((sc = dp->d_drv1) == NULL)
577 		return (ENXIO);
578 
579 	return (ENOTTY);
580 }
581 
582 static int
583 vtblk_dump(void *arg, void *virtual, off_t offset, size_t length)
584 {
585 	struct disk *dp;
586 	struct vtblk_softc *sc;
587 	int error;
588 
589 	dp = arg;
590 	error = 0;
591 
592 	if ((sc = dp->d_drv1) == NULL)
593 		return (ENXIO);
594 
595 	VTBLK_LOCK(sc);
596 
597 	vtblk_dump_quiesce(sc);
598 
599 	if (length > 0)
600 		error = vtblk_dump_write(sc, virtual, offset, length);
601 	if (error || (virtual == NULL && offset == 0))
602 		vtblk_dump_complete(sc);
603 
604 	VTBLK_UNLOCK(sc);
605 
606 	return (error);
607 }
608 
609 static void
610 vtblk_strategy(struct bio *bp)
611 {
612 	struct vtblk_softc *sc;
613 
614 	if ((sc = bp->bio_disk->d_drv1) == NULL) {
615 		vtblk_bio_done(NULL, bp, EINVAL);
616 		return;
617 	}
618 
619 	if ((bp->bio_cmd != BIO_READ) && (bp->bio_cmd != BIO_WRITE) &&
620 	    (bp->bio_cmd != BIO_FLUSH) && (bp->bio_cmd != BIO_DELETE)) {
621 		vtblk_bio_done(sc, bp, EOPNOTSUPP);
622 		return;
623 	}
624 
625 	VTBLK_LOCK(sc);
626 
627 	if (sc->vtblk_flags & VTBLK_FLAG_DETACH) {
628 		VTBLK_UNLOCK(sc);
629 		vtblk_bio_done(sc, bp, ENXIO);
630 		return;
631 	}
632 
633 	bioq_insert_tail(&sc->vtblk_bioq, bp);
634 	vtblk_startio(sc);
635 
636 	VTBLK_UNLOCK(sc);
637 }
638 
639 static int
640 vtblk_negotiate_features(struct vtblk_softc *sc)
641 {
642 	device_t dev;
643 	uint64_t features;
644 
645 	dev = sc->vtblk_dev;
646 	features = virtio_bus_is_modern(dev) ? VTBLK_MODERN_FEATURES :
647 	    VTBLK_LEGACY_FEATURES;
648 
649 	sc->vtblk_features = virtio_negotiate_features(dev, features);
650 	return (virtio_finalize_features(dev));
651 }
652 
653 static int
654 vtblk_setup_features(struct vtblk_softc *sc)
655 {
656 	device_t dev;
657 	int error;
658 
659 	dev = sc->vtblk_dev;
660 
661 	error = vtblk_negotiate_features(sc);
662 	if (error)
663 		return (error);
664 
665 	if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
666 		sc->vtblk_flags |= VTBLK_FLAG_INDIRECT;
667 	if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE))
668 		sc->vtblk_flags |= VTBLK_FLAG_WCE_CONFIG;
669 
670 	/* Legacy. */
671 	if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER))
672 		sc->vtblk_flags |= VTBLK_FLAG_BARRIER;
673 
674 	return (0);
675 }
676 
677 static int
678 vtblk_maximum_segments(struct vtblk_softc *sc,
679     struct virtio_blk_config *blkcfg)
680 {
681 	device_t dev;
682 	int nsegs;
683 
684 	dev = sc->vtblk_dev;
685 	nsegs = VTBLK_MIN_SEGMENTS;
686 
687 	if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) {
688 		nsegs += MIN(blkcfg->seg_max, maxphys / PAGE_SIZE + 1);
689 		if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT)
690 			nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT);
691 	} else
692 		nsegs += 1;
693 
694 	return (nsegs);
695 }
696 
697 static int
698 vtblk_alloc_virtqueue(struct vtblk_softc *sc)
699 {
700 	device_t dev;
701 	struct vq_alloc_info vq_info;
702 
703 	dev = sc->vtblk_dev;
704 
705 	VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs,
706 	    vtblk_vq_intr, sc, &sc->vtblk_vq,
707 	    "%s request", device_get_nameunit(dev));
708 
709 	return (virtio_alloc_virtqueues(dev, 1, &vq_info));
710 }
711 
712 static void
713 vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity)
714 {
715 	device_t dev;
716 	struct disk *dp;
717 	int error;
718 
719 	dev = sc->vtblk_dev;
720 	dp = sc->vtblk_disk;
721 
722 	dp->d_mediasize = new_capacity;
723 	if (bootverbose) {
724 		device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n",
725 		    (uintmax_t) dp->d_mediasize >> 20,
726 		    (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
727 		    dp->d_sectorsize);
728 	}
729 
730 	error = disk_resize(dp, M_NOWAIT);
731 	if (error) {
732 		device_printf(dev,
733 		    "disk_resize(9) failed, error: %d\n", error);
734 	}
735 }
736 
737 static void
738 vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
739 {
740 	device_t dev;
741 	struct disk *dp;
742 
743 	dev = sc->vtblk_dev;
744 
745 	sc->vtblk_disk = dp = disk_alloc();
746 	dp->d_open = vtblk_open;
747 	dp->d_close = vtblk_close;
748 	dp->d_ioctl = vtblk_ioctl;
749 	dp->d_strategy = vtblk_strategy;
750 	dp->d_name = VTBLK_DISK_NAME;
751 	dp->d_unit = device_get_unit(dev);
752 	dp->d_drv1 = sc;
753 	dp->d_flags = DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION;
754 	dp->d_hba_vendor = virtio_get_vendor(dev);
755 	dp->d_hba_device = virtio_get_device(dev);
756 	dp->d_hba_subvendor = virtio_get_subvendor(dev);
757 	dp->d_hba_subdevice = virtio_get_subdevice(dev);
758 
759 	if (virtio_with_feature(dev, VIRTIO_BLK_F_RO))
760 		dp->d_flags |= DISKFLAG_WRITE_PROTECT;
761 	else {
762 		if (virtio_with_feature(dev, VIRTIO_BLK_F_FLUSH))
763 			dp->d_flags |= DISKFLAG_CANFLUSHCACHE;
764 		dp->d_dump = vtblk_dump;
765 	}
766 
767 	/* Capacity is always in 512-byte units. */
768 	dp->d_mediasize = blkcfg->capacity * VTBLK_BSIZE;
769 
770 	if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE))
771 		dp->d_sectorsize = blkcfg->blk_size;
772 	else
773 		dp->d_sectorsize = VTBLK_BSIZE;
774 
775 	/*
776 	 * The VirtIO maximum I/O size is given in terms of segments.
777 	 * However, FreeBSD limits I/O size by logical buffer size, not
778 	 * by physically contiguous pages. Therefore, we have to assume
779 	 * no pages are contiguous. This may impose an artificially low
780 	 * maximum I/O size. But in practice, since QEMU advertises 128
781 	 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE,
782 	 * which is typically greater than maxphys. Eventually we should
783 	 * just advertise maxphys and split buffers that are too big.
784 	 *
785 	 * If we're not asking busdma to align data to page boundaries, the
786 	 * maximum I/O size is reduced by PAGE_SIZE in order to accommodate
787 	 * unaligned I/Os.
788 	 */
789 	dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS) *
790 	    PAGE_SIZE;
791 	if ((sc->vtblk_flags & VTBLK_FLAG_BUSDMA_ALIGN) == 0)
792 		dp->d_maxsize -= PAGE_SIZE;
793 
794 	if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) {
795 		dp->d_fwsectors = blkcfg->geometry.sectors;
796 		dp->d_fwheads = blkcfg->geometry.heads;
797 	}
798 
799 	if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) &&
800 	    blkcfg->topology.physical_block_exp > 0) {
801 		dp->d_stripesize = dp->d_sectorsize *
802 		    (1 << blkcfg->topology.physical_block_exp);
803 		dp->d_stripeoffset = (dp->d_stripesize -
804 		    blkcfg->topology.alignment_offset * dp->d_sectorsize) %
805 		    dp->d_stripesize;
806 	}
807 
808 	if (virtio_with_feature(dev, VIRTIO_BLK_F_DISCARD)) {
809 		dp->d_flags |= DISKFLAG_CANDELETE;
810 		dp->d_delmaxsize = blkcfg->max_discard_sectors * VTBLK_BSIZE;
811 	}
812 
813 	if (vtblk_write_cache_enabled(sc, blkcfg) != 0)
814 		sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK;
815 	else
816 		sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH;
817 }
818 
819 static void
820 vtblk_create_disk(struct vtblk_softc *sc)
821 {
822 	struct disk *dp;
823 
824 	dp = sc->vtblk_disk;
825 
826 	vtblk_ident(sc);
827 
828 	device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n",
829 	    (uintmax_t) dp->d_mediasize >> 20,
830 	    (uintmax_t) dp->d_mediasize / dp->d_sectorsize,
831 	    dp->d_sectorsize);
832 
833 	disk_create(dp, DISK_VERSION);
834 }
835 
836 static int
837 vtblk_request_prealloc(struct vtblk_softc *sc)
838 {
839 	struct vtblk_request *req;
840 	int i, nreqs;
841 
842 	nreqs = virtqueue_size(sc->vtblk_vq);
843 
844 	/*
845 	 * Preallocate sufficient requests to keep the virtqueue full. Each
846 	 * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce
847 	 * the number allocated when indirect descriptors are not available.
848 	 */
849 	if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0)
850 		nreqs /= VTBLK_MIN_SEGMENTS;
851 
852 	for (i = 0; i < nreqs; i++) {
853 		req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT);
854 		if (req == NULL)
855 			return (ENOMEM);
856 
857 		req->vbr_sc = sc;
858 		if (bus_dmamap_create(sc->vtblk_dmat, 0, &req->vbr_mapp)) {
859 			free(req, M_DEVBUF);
860 			return (ENOMEM);
861 		}
862 
863 		MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1);
864 		MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1);
865 
866 		sc->vtblk_request_count++;
867 		vtblk_request_enqueue(sc, req);
868 	}
869 
870 	return (0);
871 }
872 
873 static void
874 vtblk_request_free(struct vtblk_softc *sc)
875 {
876 	struct vtblk_request *req;
877 
878 	MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready));
879 
880 	while ((req = vtblk_request_dequeue(sc)) != NULL) {
881 		sc->vtblk_request_count--;
882 		bus_dmamap_destroy(sc->vtblk_dmat, req->vbr_mapp);
883 		free(req, M_DEVBUF);
884 	}
885 
886 	KASSERT(sc->vtblk_request_count == 0,
887 	    ("%s: leaked %d requests", __func__, sc->vtblk_request_count));
888 }
889 
890 static struct vtblk_request *
891 vtblk_request_dequeue(struct vtblk_softc *sc)
892 {
893 	struct vtblk_request *req;
894 
895 	req = TAILQ_FIRST(&sc->vtblk_req_free);
896 	if (req != NULL) {
897 		TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link);
898 		bzero(&req->vbr_hdr, sizeof(struct vtblk_request) -
899 		    offsetof(struct vtblk_request, vbr_hdr));
900 	}
901 
902 	return (req);
903 }
904 
905 static void
906 vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req)
907 {
908 
909 	TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link);
910 }
911 
912 static struct vtblk_request *
913 vtblk_request_next_ready(struct vtblk_softc *sc)
914 {
915 	struct vtblk_request *req;
916 
917 	req = TAILQ_FIRST(&sc->vtblk_req_ready);
918 	if (req != NULL)
919 		TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link);
920 
921 	return (req);
922 }
923 
924 static void
925 vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req)
926 {
927 
928 	/* NOTE: Currently, there will be at most one request in the queue. */
929 	TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link);
930 }
931 
932 static struct vtblk_request *
933 vtblk_request_next(struct vtblk_softc *sc)
934 {
935 	struct vtblk_request *req;
936 
937 	req = vtblk_request_next_ready(sc);
938 	if (req != NULL)
939 		return (req);
940 
941 	return (vtblk_request_bio(sc));
942 }
943 
944 static struct vtblk_request *
945 vtblk_request_bio(struct vtblk_softc *sc)
946 {
947 	struct bio_queue_head *bioq;
948 	struct vtblk_request *req;
949 	struct bio *bp;
950 
951 	bioq = &sc->vtblk_bioq;
952 
953 	if (bioq_first(bioq) == NULL)
954 		return (NULL);
955 
956 	req = vtblk_request_dequeue(sc);
957 	if (req == NULL)
958 		return (NULL);
959 
960 	bp = bioq_takefirst(bioq);
961 	req->vbr_bp = bp;
962 	req->vbr_ack = -1;
963 	req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
964 
965 	switch (bp->bio_cmd) {
966 	case BIO_FLUSH:
967 		req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH);
968 		req->vbr_hdr.sector = 0;
969 		break;
970 	case BIO_READ:
971 		req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_IN);
972 		req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
973 		break;
974 	case BIO_WRITE:
975 		req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT);
976 		req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
977 		break;
978 	case BIO_DELETE:
979 		req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_DISCARD);
980 		req->vbr_hdr.sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
981 		break;
982 	default:
983 		panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd);
984 	}
985 
986 	if (bp->bio_flags & BIO_ORDERED)
987 		req->vbr_hdr.type |= vtblk_gtoh32(sc, VIRTIO_BLK_T_BARRIER);
988 
989 	return (req);
990 }
991 
992 static int
993 vtblk_request_execute(struct vtblk_request *req, int flags)
994 {
995 	struct vtblk_softc *sc = req->vbr_sc;
996 	struct bio *bp = req->vbr_bp;
997 	int error = 0;
998 
999 	/*
1000 	 * Call via bus_dmamap_load_bio or directly depending on whether we
1001 	 * have a buffer we need to map.  If we don't have a busdma map,
1002 	 * try to perform the I/O directly and hope that it works (this will
1003 	 * happen when dumping).
1004 	 */
1005 	if ((req->vbr_mapp != NULL) &&
1006 	    (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) {
1007 		error = bus_dmamap_load_bio(sc->vtblk_dmat, req->vbr_mapp,
1008 		    req->vbr_bp, vtblk_request_execute_cb, req, flags);
1009 		if (error == EINPROGRESS) {
1010 			req->vbr_busdma_wait = 1;
1011 			sc->vtblk_flags |= VTBLK_FLAG_BUSDMA_WAIT;
1012 		}
1013 	} else {
1014 		vtblk_request_execute_cb(req, NULL, 0, 0);
1015 	}
1016 
1017 	return (error ? error : req->vbr_error);
1018 }
1019 
1020 static void
1021 vtblk_request_execute_cb(void * callback_arg, bus_dma_segment_t * segs,
1022     int nseg, int error)
1023 {
1024 	struct vtblk_request *req;
1025 	struct vtblk_softc *sc;
1026 	struct virtqueue *vq;
1027 	struct sglist *sg;
1028 	struct bio *bp;
1029 	int ordered, readable, writable, i;
1030 
1031 	req = (struct vtblk_request *)callback_arg;
1032 	sc = req->vbr_sc;
1033 	vq = sc->vtblk_vq;
1034 	sg = sc->vtblk_sglist;
1035 	bp = req->vbr_bp;
1036 	ordered = 0;
1037 	writable = 0;
1038 
1039 	/*
1040 	 * If we paused request queueing while we waited for busdma to call us
1041 	 * asynchronously, unpause it now; this request made it through so we
1042 	 * don't need to worry about others getting ahead of us.  (Note that we
1043 	 * hold the device mutex so nothing will happen until after we return
1044 	 * anyway.)
1045 	 */
1046 	if (req->vbr_busdma_wait)
1047 		sc->vtblk_flags &= ~VTBLK_FLAG_BUSDMA_WAIT;
1048 
1049 	/* Fail on errors from busdma. */
1050 	if (error)
1051 		goto out1;
1052 
1053 	/*
1054 	 * Some hosts (such as bhyve) do not implement the barrier feature,
1055 	 * so we emulate it in the driver by allowing the barrier request
1056 	 * to be the only one in flight.
1057 	 */
1058 	if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) {
1059 		if (sc->vtblk_req_ordered != NULL) {
1060 			error = EBUSY;
1061 			goto out;
1062 		}
1063 		if (bp->bio_flags & BIO_ORDERED) {
1064 			if (!virtqueue_empty(vq)) {
1065 				error = EBUSY;
1066 				goto out;
1067 			}
1068 			ordered = 1;
1069 			req->vbr_hdr.type &= vtblk_gtoh32(sc,
1070 				~VIRTIO_BLK_T_BARRIER);
1071 		}
1072 	}
1073 
1074 	sglist_reset(sg);
1075 	sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr));
1076 
1077 	if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1078 		/*
1079 		 * We cast bus_addr_t to vm_paddr_t here; since we skip the
1080 		 * iommu mapping (see vtblk_attach) this should be safe.
1081 		 */
1082 		for (i = 0; i < nseg; i++) {
1083 			error = sglist_append_phys(sg,
1084 			    (vm_paddr_t)segs[i].ds_addr, segs[i].ds_len);
1085 			if (error || sg->sg_nseg == sg->sg_maxseg) {
1086 				panic("%s: bio %p data buffer too big %d",
1087 				    __func__, bp, error);
1088 			}
1089 		}
1090 
1091 		/* Special handling for dump, which bypasses busdma. */
1092 		if (req->vbr_mapp == NULL) {
1093 			error = sglist_append_bio(sg, bp);
1094 			if (error || sg->sg_nseg == sg->sg_maxseg) {
1095 				panic("%s: bio %p data buffer too big %d",
1096 				    __func__, bp, error);
1097 			}
1098 		}
1099 
1100 		/* BIO_READ means the host writes into our buffer. */
1101 		if (bp->bio_cmd == BIO_READ)
1102 			writable = sg->sg_nseg - 1;
1103 	} else if (bp->bio_cmd == BIO_DELETE) {
1104 		struct virtio_blk_discard_write_zeroes *discard;
1105 
1106 		discard = malloc(sizeof(*discard), M_DEVBUF, M_NOWAIT | M_ZERO);
1107 		if (discard == NULL) {
1108 			error = ENOMEM;
1109 			goto out;
1110 		}
1111 
1112 		bp->bio_driver1 = discard;
1113 		discard->sector = vtblk_gtoh64(sc, bp->bio_offset / VTBLK_BSIZE);
1114 		discard->num_sectors = vtblk_gtoh32(sc, bp->bio_bcount / VTBLK_BSIZE);
1115 		error = sglist_append(sg, discard, sizeof(*discard));
1116 		if (error || sg->sg_nseg == sg->sg_maxseg) {
1117 			panic("%s: bio %p data buffer too big %d",
1118 			    __func__, bp, error);
1119 		}
1120 	}
1121 
1122 	writable++;
1123 	sglist_append(sg, &req->vbr_ack, sizeof(uint8_t));
1124 	readable = sg->sg_nseg - writable;
1125 
1126 	if (req->vbr_mapp != NULL) {
1127 		switch (bp->bio_cmd) {
1128 		case BIO_READ:
1129 			bus_dmamap_sync(sc->vtblk_dmat, req->vbr_mapp,
1130 			    BUS_DMASYNC_PREREAD);
1131 			break;
1132 		case BIO_WRITE:
1133 			bus_dmamap_sync(sc->vtblk_dmat, req->vbr_mapp,
1134 			    BUS_DMASYNC_PREWRITE);
1135 			break;
1136 		}
1137 	}
1138 
1139 	error = virtqueue_enqueue(vq, req, sg, readable, writable);
1140 	if (error == 0 && ordered)
1141 		sc->vtblk_req_ordered = req;
1142 
1143 	/*
1144 	 * If we were called asynchronously, we need to notify the queue that
1145 	 * we've added a new request, since the notification from startio was
1146 	 * performed already.
1147 	 */
1148 	if (error == 0 && req->vbr_busdma_wait)
1149 		virtqueue_notify(vq);
1150 
1151 out:
1152 	if (error && (req->vbr_mapp != NULL))
1153 		bus_dmamap_unload(sc->vtblk_dmat, req->vbr_mapp);
1154 out1:
1155 	if (error && req->vbr_requeue_on_error)
1156 		vtblk_request_requeue_ready(sc, req);
1157 	req->vbr_error = error;
1158 }
1159 
1160 static int
1161 vtblk_request_error(struct vtblk_request *req)
1162 {
1163 	int error;
1164 
1165 	switch (req->vbr_ack) {
1166 	case VIRTIO_BLK_S_OK:
1167 		error = 0;
1168 		break;
1169 	case VIRTIO_BLK_S_UNSUPP:
1170 		error = ENOTSUP;
1171 		break;
1172 	default:
1173 		error = EIO;
1174 		break;
1175 	}
1176 
1177 	return (error);
1178 }
1179 
1180 static void
1181 vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue)
1182 {
1183 	struct vtblk_request *req;
1184 	struct bio *bp;
1185 
1186 	while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) {
1187 		if (sc->vtblk_req_ordered != NULL) {
1188 			MPASS(sc->vtblk_req_ordered == req);
1189 			sc->vtblk_req_ordered = NULL;
1190 		}
1191 
1192 		bp = req->vbr_bp;
1193 		if (req->vbr_mapp != NULL) {
1194 			switch (bp->bio_cmd) {
1195 			case BIO_READ:
1196 				bus_dmamap_sync(sc->vtblk_dmat, req->vbr_mapp,
1197 				    BUS_DMASYNC_POSTREAD);
1198 				bus_dmamap_unload(sc->vtblk_dmat,
1199 				    req->vbr_mapp);
1200 				break;
1201 			case BIO_WRITE:
1202 				bus_dmamap_sync(sc->vtblk_dmat, req->vbr_mapp,
1203 				    BUS_DMASYNC_POSTWRITE);
1204 				bus_dmamap_unload(sc->vtblk_dmat,
1205 				    req->vbr_mapp);
1206 				break;
1207 			}
1208 		}
1209 		bp->bio_error = vtblk_request_error(req);
1210 		TAILQ_INSERT_TAIL(queue, bp, bio_queue);
1211 
1212 		vtblk_request_enqueue(sc, req);
1213 	}
1214 }
1215 
1216 static void
1217 vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue)
1218 {
1219 	struct bio *bp, *tmp;
1220 
1221 	TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) {
1222 		if (bp->bio_error != 0)
1223 			disk_err(bp, "hard error", -1, 1);
1224 		vtblk_bio_done(sc, bp, bp->bio_error);
1225 	}
1226 }
1227 
1228 static void
1229 vtblk_drain_vq(struct vtblk_softc *sc)
1230 {
1231 	struct virtqueue *vq;
1232 	struct vtblk_request *req;
1233 	int last;
1234 
1235 	vq = sc->vtblk_vq;
1236 	last = 0;
1237 
1238 	while ((req = virtqueue_drain(vq, &last)) != NULL) {
1239 		vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1240 		vtblk_request_enqueue(sc, req);
1241 	}
1242 
1243 	sc->vtblk_req_ordered = NULL;
1244 	KASSERT(virtqueue_empty(vq), ("virtqueue not empty"));
1245 }
1246 
1247 static void
1248 vtblk_drain(struct vtblk_softc *sc)
1249 {
1250 	struct bio_queue_head *bioq;
1251 	struct vtblk_request *req;
1252 	struct bio *bp;
1253 
1254 	bioq = &sc->vtblk_bioq;
1255 
1256 	if (sc->vtblk_vq != NULL) {
1257 		struct bio_queue queue;
1258 
1259 		TAILQ_INIT(&queue);
1260 		vtblk_queue_completed(sc, &queue);
1261 		vtblk_done_completed(sc, &queue);
1262 
1263 		vtblk_drain_vq(sc);
1264 	}
1265 
1266 	while ((req = vtblk_request_next_ready(sc)) != NULL) {
1267 		vtblk_bio_done(sc, req->vbr_bp, ENXIO);
1268 		vtblk_request_enqueue(sc, req);
1269 	}
1270 
1271 	while (bioq_first(bioq) != NULL) {
1272 		bp = bioq_takefirst(bioq);
1273 		vtblk_bio_done(sc, bp, ENXIO);
1274 	}
1275 
1276 	vtblk_request_free(sc);
1277 }
1278 
1279 static void
1280 vtblk_startio(struct vtblk_softc *sc)
1281 {
1282 	struct virtqueue *vq;
1283 	struct vtblk_request *req;
1284 	int enq;
1285 
1286 	VTBLK_LOCK_ASSERT(sc);
1287 	vq = sc->vtblk_vq;
1288 	enq = 0;
1289 
1290 	if (sc->vtblk_flags & (VTBLK_FLAG_SUSPEND | VTBLK_FLAG_BUSDMA_WAIT))
1291 		return;
1292 
1293 	while (!virtqueue_full(vq)) {
1294 		req = vtblk_request_next(sc);
1295 		if (req == NULL)
1296 			break;
1297 
1298 		req->vbr_requeue_on_error = 1;
1299 		if (vtblk_request_execute(req, BUS_DMA_WAITOK))
1300 			break;
1301 
1302 		enq++;
1303 	}
1304 
1305 	if (enq > 0)
1306 		virtqueue_notify(vq);
1307 }
1308 
1309 static void
1310 vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error)
1311 {
1312 
1313 	/* Because of GEOM direct dispatch, we cannot hold any locks. */
1314 	if (sc != NULL)
1315 		VTBLK_LOCK_ASSERT_NOTOWNED(sc);
1316 
1317 	if (error) {
1318 		bp->bio_resid = bp->bio_bcount;
1319 		bp->bio_error = error;
1320 		bp->bio_flags |= BIO_ERROR;
1321 	} else {
1322 		kmsan_mark_bio(bp, KMSAN_STATE_INITED);
1323 	}
1324 
1325 	if (bp->bio_driver1 != NULL) {
1326 		free(bp->bio_driver1, M_DEVBUF);
1327 		bp->bio_driver1 = NULL;
1328 	}
1329 
1330 	biodone(bp);
1331 }
1332 
1333 #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg)			\
1334 	if (virtio_with_feature(_dev, _feature)) {			\
1335 		virtio_read_device_config(_dev,				\
1336 		    offsetof(struct virtio_blk_config, _field),		\
1337 		    &(_cfg)->_field, sizeof((_cfg)->_field));		\
1338 	}
1339 
1340 static void
1341 vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg)
1342 {
1343 	device_t dev;
1344 
1345 	dev = sc->vtblk_dev;
1346 
1347 	bzero(blkcfg, sizeof(struct virtio_blk_config));
1348 
1349 	/* The capacity is always available. */
1350 	virtio_read_device_config(dev, offsetof(struct virtio_blk_config,
1351 	    capacity), &blkcfg->capacity, sizeof(blkcfg->capacity));
1352 
1353 	/* Read the configuration if the feature was negotiated. */
1354 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg);
1355 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg);
1356 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY,
1357 	    geometry.cylinders, blkcfg);
1358 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY,
1359 	    geometry.heads, blkcfg);
1360 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY,
1361 	    geometry.sectors, blkcfg);
1362 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg);
1363 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1364 	    topology.physical_block_exp, blkcfg);
1365 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1366 	    topology.alignment_offset, blkcfg);
1367 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1368 	    topology.min_io_size, blkcfg);
1369 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY,
1370 	    topology.opt_io_size, blkcfg);
1371 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, wce, blkcfg);
1372 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_sectors,
1373 	    blkcfg);
1374 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, max_discard_seg, blkcfg);
1375 	VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_DISCARD, discard_sector_alignment,
1376 	    blkcfg);
1377 }
1378 
1379 #undef VTBLK_GET_CONFIG
1380 
1381 static void
1382 vtblk_ident(struct vtblk_softc *sc)
1383 {
1384 	struct bio buf;
1385 	struct disk *dp;
1386 	struct vtblk_request *req;
1387 	int len, error;
1388 
1389 	dp = sc->vtblk_disk;
1390 	len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE);
1391 
1392 	if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0)
1393 		return;
1394 
1395 	req = vtblk_request_dequeue(sc);
1396 	if (req == NULL)
1397 		return;
1398 
1399 	req->vbr_ack = -1;
1400 	req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_GET_ID);
1401 	req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
1402 	req->vbr_hdr.sector = 0;
1403 
1404 	req->vbr_bp = &buf;
1405 	g_reset_bio(&buf);
1406 
1407 	buf.bio_cmd = BIO_READ;
1408 	buf.bio_data = dp->d_ident;
1409 	buf.bio_bcount = len;
1410 
1411 	VTBLK_LOCK(sc);
1412 	error = vtblk_poll_request(sc, req);
1413 	VTBLK_UNLOCK(sc);
1414 
1415 	vtblk_request_enqueue(sc, req);
1416 
1417 	if (error) {
1418 		device_printf(sc->vtblk_dev,
1419 		    "error getting device identifier: %d\n", error);
1420 	}
1421 }
1422 
1423 static int
1424 vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req)
1425 {
1426 	struct virtqueue *vq;
1427 	int error;
1428 
1429 	vq = sc->vtblk_vq;
1430 
1431 	if (!virtqueue_empty(vq))
1432 		return (EBUSY);
1433 
1434 	error = vtblk_request_execute(req, BUS_DMA_NOWAIT);
1435 	if (error)
1436 		return (error);
1437 
1438 	virtqueue_notify(vq);
1439 	virtqueue_poll(vq, NULL);
1440 
1441 	error = vtblk_request_error(req);
1442 	if (error && bootverbose) {
1443 		device_printf(sc->vtblk_dev,
1444 		    "%s: IO error: %d\n", __func__, error);
1445 	}
1446 
1447 	return (error);
1448 }
1449 
1450 static int
1451 vtblk_quiesce(struct vtblk_softc *sc)
1452 {
1453 	int error;
1454 
1455 	VTBLK_LOCK_ASSERT(sc);
1456 	error = 0;
1457 
1458 	while (!virtqueue_empty(sc->vtblk_vq)) {
1459 		if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq",
1460 		    VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) {
1461 			error = EBUSY;
1462 			break;
1463 		}
1464 	}
1465 
1466 	return (error);
1467 }
1468 
1469 static void
1470 vtblk_vq_intr(void *xsc)
1471 {
1472 	struct vtblk_softc *sc;
1473 	struct virtqueue *vq;
1474 	struct bio_queue queue;
1475 
1476 	sc = xsc;
1477 	vq = sc->vtblk_vq;
1478 	TAILQ_INIT(&queue);
1479 
1480 	VTBLK_LOCK(sc);
1481 
1482 again:
1483 	if (sc->vtblk_flags & VTBLK_FLAG_DETACH)
1484 		goto out;
1485 
1486 	vtblk_queue_completed(sc, &queue);
1487 	vtblk_startio(sc);
1488 
1489 	if (virtqueue_enable_intr(vq) != 0) {
1490 		virtqueue_disable_intr(vq);
1491 		goto again;
1492 	}
1493 
1494 	if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND)
1495 		wakeup(&sc->vtblk_vq);
1496 
1497 out:
1498 	VTBLK_UNLOCK(sc);
1499 	vtblk_done_completed(sc, &queue);
1500 }
1501 
1502 static void
1503 vtblk_stop(struct vtblk_softc *sc)
1504 {
1505 
1506 	virtqueue_disable_intr(sc->vtblk_vq);
1507 	virtio_stop(sc->vtblk_dev);
1508 }
1509 
1510 static void
1511 vtblk_dump_quiesce(struct vtblk_softc *sc)
1512 {
1513 
1514 	/*
1515 	 * Spin here until all the requests in-flight at the time of the
1516 	 * dump are completed and queued. The queued requests will be
1517 	 * biodone'd once the dump is finished.
1518 	 */
1519 	while (!virtqueue_empty(sc->vtblk_vq))
1520 		vtblk_queue_completed(sc, &sc->vtblk_dump_queue);
1521 }
1522 
1523 static int
1524 vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset,
1525     size_t length)
1526 {
1527 	struct bio buf;
1528 	struct vtblk_request *req;
1529 
1530 	req = &sc->vtblk_dump_request;
1531 	req->vbr_sc = sc;
1532 	req->vbr_ack = -1;
1533 	req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_OUT);
1534 	req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
1535 	req->vbr_hdr.sector = vtblk_gtoh64(sc, offset / VTBLK_BSIZE);
1536 
1537 	req->vbr_bp = &buf;
1538 	g_reset_bio(&buf);
1539 
1540 	buf.bio_cmd = BIO_WRITE;
1541 	buf.bio_data = virtual;
1542 	buf.bio_bcount = length;
1543 
1544 	return (vtblk_poll_request(sc, req));
1545 }
1546 
1547 static int
1548 vtblk_dump_flush(struct vtblk_softc *sc)
1549 {
1550 	struct bio buf;
1551 	struct vtblk_request *req;
1552 
1553 	req = &sc->vtblk_dump_request;
1554 	req->vbr_sc = sc;
1555 	req->vbr_ack = -1;
1556 	req->vbr_hdr.type = vtblk_gtoh32(sc, VIRTIO_BLK_T_FLUSH);
1557 	req->vbr_hdr.ioprio = vtblk_gtoh32(sc, 1);
1558 	req->vbr_hdr.sector = 0;
1559 
1560 	req->vbr_bp = &buf;
1561 	g_reset_bio(&buf);
1562 
1563 	buf.bio_cmd = BIO_FLUSH;
1564 
1565 	return (vtblk_poll_request(sc, req));
1566 }
1567 
1568 static void
1569 vtblk_dump_complete(struct vtblk_softc *sc)
1570 {
1571 
1572 	vtblk_dump_flush(sc);
1573 
1574 	VTBLK_UNLOCK(sc);
1575 	vtblk_done_completed(sc, &sc->vtblk_dump_queue);
1576 	VTBLK_LOCK(sc);
1577 }
1578 
1579 static void
1580 vtblk_set_write_cache(struct vtblk_softc *sc, int wc)
1581 {
1582 
1583 	/* Set either writeback (1) or writethrough (0) mode. */
1584 	virtio_write_dev_config_1(sc->vtblk_dev,
1585 	    offsetof(struct virtio_blk_config, wce), wc);
1586 }
1587 
1588 static int
1589 vtblk_write_cache_enabled(struct vtblk_softc *sc,
1590     struct virtio_blk_config *blkcfg)
1591 {
1592 	int wc;
1593 
1594 	if (sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) {
1595 		wc = vtblk_tunable_int(sc, "writecache_mode",
1596 		    vtblk_writecache_mode);
1597 		if (wc >= 0 && wc < VTBLK_CACHE_MAX)
1598 			vtblk_set_write_cache(sc, wc);
1599 		else
1600 			wc = blkcfg->wce;
1601 	} else
1602 		wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_FLUSH);
1603 
1604 	return (wc);
1605 }
1606 
1607 static int
1608 vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS)
1609 {
1610 	struct vtblk_softc *sc;
1611 	int wc, error;
1612 
1613 	sc = oidp->oid_arg1;
1614 	wc = sc->vtblk_write_cache;
1615 
1616 	error = sysctl_handle_int(oidp, &wc, 0, req);
1617 	if (error || req->newptr == NULL)
1618 		return (error);
1619 	if ((sc->vtblk_flags & VTBLK_FLAG_WCE_CONFIG) == 0)
1620 		return (EPERM);
1621 	if (wc < 0 || wc >= VTBLK_CACHE_MAX)
1622 		return (EINVAL);
1623 
1624 	VTBLK_LOCK(sc);
1625 	sc->vtblk_write_cache = wc;
1626 	vtblk_set_write_cache(sc, sc->vtblk_write_cache);
1627 	VTBLK_UNLOCK(sc);
1628 
1629 	return (0);
1630 }
1631 
1632 static void
1633 vtblk_setup_sysctl(struct vtblk_softc *sc)
1634 {
1635 	device_t dev;
1636 	struct sysctl_ctx_list *ctx;
1637 	struct sysctl_oid *tree;
1638 	struct sysctl_oid_list *child;
1639 
1640 	dev = sc->vtblk_dev;
1641 	ctx = device_get_sysctl_ctx(dev);
1642 	tree = device_get_sysctl_tree(dev);
1643 	child = SYSCTL_CHILDREN(tree);
1644 
1645 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode",
1646 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
1647 	    vtblk_write_cache_sysctl, "I",
1648 	    "Write cache mode (writethrough (0) or writeback (1))");
1649 }
1650 
1651 static int
1652 vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def)
1653 {
1654 	char path[64];
1655 
1656 	snprintf(path, sizeof(path),
1657 	    "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob);
1658 	TUNABLE_INT_FETCH(path, &def);
1659 
1660 	return (def);
1661 }
1662