pci_virtio_block.c revision 256390 
1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 256390 2013-10-12 19:41:35Z grehan $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 256390 2013-10-12 19:41:35Z grehan $");
31
32#include <sys/param.h>
33#include <sys/linker_set.h>
34#include <sys/stat.h>
35#include <sys/uio.h>
36#include <sys/ioctl.h>
37#include <sys/disk.h>
38
39#include <errno.h>
40#include <fcntl.h>
41#include <stdio.h>
42#include <stdlib.h>
43#include <stdint.h>
44#include <string.h>
45#include <strings.h>
46#include <unistd.h>
47#include <assert.h>
48#include <pthread.h>
49#include <md5.h>
50
51#include "bhyverun.h"
52#include "pci_emul.h"
53#include "virtio.h"
54
55#ifndef min
56#define	min(a, b)	((a) < (b) ? (a) : (b))
57#endif
58
59#define VTBLK_RINGSZ	64
60
61#define VTBLK_MAXSEGS	32
62
63#define VTBLK_S_OK	0
64#define VTBLK_S_IOERR	1
65#define	VTBLK_S_UNSUPP	2
66
67#define	VTBLK_BLK_ID_BYTES	20
68
69/*
70 * Host capabilities
71 */
72#define VTBLK_S_HOSTCAPS      \
73  ( 0x00000004 |	/* host maximum request segments */ \
74    VIRTIO_RING_F_INDIRECT_DESC )	/* indirect descriptors */
75
76/*
77 * Config space "registers"
78 */
79struct vtblk_config {
80	uint64_t	vbc_capacity;
81	uint32_t	vbc_size_max;
82	uint32_t	vbc_seg_max;
83	uint16_t	vbc_geom_c;
84	uint8_t		vbc_geom_h;
85	uint8_t		vbc_geom_s;
86	uint32_t	vbc_blk_size;
87	uint32_t	vbc_sectors_max;
88} __packed;
89
90/*
91 * Fixed-size block header
92 */
93struct virtio_blk_hdr {
94#define	VBH_OP_READ		0
95#define	VBH_OP_WRITE		1
96#define	VBH_OP_IDENT		8
97#define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
98	uint32_t       	vbh_type;
99	uint32_t	vbh_ioprio;
100	uint64_t	vbh_sector;
101} __packed;
102
103/*
104 * Debug printf
105 */
106static int pci_vtblk_debug;
107#define DPRINTF(params) if (pci_vtblk_debug) printf params
108#define WPRINTF(params) printf params
109
110/*
111 * Per-device softc
112 */
113struct pci_vtblk_softc {
114	struct virtio_softc vbsc_vs;
115	struct vqueue_info vbsc_vq;
116	int		vbsc_fd;
117	struct vtblk_config vbsc_cfg;
118	char vbsc_ident[VTBLK_BLK_ID_BYTES];
119};
120
121static void pci_vtblk_reset(void *);
122static void pci_vtblk_notify(void *, struct vqueue_info *);
123static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
124static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
125
126static struct virtio_consts vtblk_vi_consts = {
127	"vtblk",		/* our name */
128	1,			/* we support 1 virtqueue */
129	sizeof(struct vtblk_config), /* config reg size */
130	pci_vtblk_reset,	/* reset */
131	pci_vtblk_notify,	/* device-wide qnotify */
132	pci_vtblk_cfgread,	/* read PCI config */
133	pci_vtblk_cfgwrite,	/* write PCI config */
134	VTBLK_S_HOSTCAPS,	/* our capabilities */
135};
136
137static void
138pci_vtblk_reset(void *vsc)
139{
140	struct pci_vtblk_softc *sc = vsc;
141
142	DPRINTF(("vtblk: device reset requested !\n"));
143	vi_reset_dev(&sc->vbsc_vs);
144}
145
146static void
147pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
148{
149	struct virtio_blk_hdr *vbh;
150	uint8_t *status;
151	int i, n;
152	int err;
153	int iolen;
154	int writeop, type;
155	off_t offset;
156	struct iovec iov[VTBLK_MAXSEGS + 2];
157	uint16_t flags[VTBLK_MAXSEGS + 2];
158
159	n = vq_getchain(vq, iov, VTBLK_MAXSEGS + 2, flags);
160
161	/*
162	 * The first descriptor will be the read-only fixed header,
163	 * and the last is for status (hence +2 above and below).
164	 * The remaining iov's are the actual data I/O vectors.
165	 *
166	 * XXX - note - this fails on crash dump, which does a
167	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
168	 */
169	assert(n >= 2 && n <= VTBLK_MAXSEGS + 2);
170
171	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
172	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
173	vbh = iov[0].iov_base;
174
175	status = iov[--n].iov_base;
176	assert(iov[n].iov_len == 1);
177	assert(flags[n] & VRING_DESC_F_WRITE);
178
179	/*
180	 * XXX
181	 * The guest should not be setting the BARRIER flag because
182	 * we don't advertise the capability.
183	 */
184	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
185	writeop = (type == VBH_OP_WRITE);
186
187	offset = vbh->vbh_sector * DEV_BSIZE;
188
189	iolen = 0;
190	for (i = 1; i < n; i++) {
191		/*
192		 * - write op implies read-only descriptor,
193		 * - read/ident op implies write-only descriptor,
194		 * therefore test the inverse of the descriptor bit
195		 * to the op.
196		 */
197		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
198		iolen += iov[i].iov_len;
199	}
200
201	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
202		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
203
204	switch (type) {
205	case VBH_OP_WRITE:
206		err = pwritev(sc->vbsc_fd, iov + 1, i - 1, offset);
207		break;
208	case VBH_OP_READ:
209		err = preadv(sc->vbsc_fd, iov + 1, i - 1, offset);
210		break;
211	case VBH_OP_IDENT:
212		/* Assume a single buffer */
213		strlcpy(iov[1].iov_base, sc->vbsc_ident,
214		    min(iov[1].iov_len, sizeof(sc->vbsc_ident)));
215		err = 0;
216		break;
217	default:
218		err = -ENOSYS;
219		break;
220	}
221
222	/* convert errno into a virtio block error return */
223	if (err < 0) {
224		if (err == -ENOSYS)
225			*status = VTBLK_S_UNSUPP;
226		else
227			*status = VTBLK_S_IOERR;
228	} else
229		*status = VTBLK_S_OK;
230
231	/*
232	 * Return the descriptor back to the host.
233	 * We wrote 1 byte (our status) to host.
234	 */
235	vq_relchain(vq, 1);
236}
237
238static void
239pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
240{
241	struct pci_vtblk_softc *sc = vsc;
242
243	vq_startchains(vq);
244	while (vq_has_descs(vq))
245		pci_vtblk_proc(sc, vq);
246	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
247}
248
249static int
250pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
251{
252	struct stat sbuf;
253	MD5_CTX mdctx;
254	u_char digest[16];
255	struct pci_vtblk_softc *sc;
256	off_t size;
257	int fd;
258	int sectsz;
259	int use_msix;
260	const char *env_msi;
261
262	if (opts == NULL) {
263		printf("virtio-block: backing device required\n");
264		return (1);
265	}
266
267	/*
268	 * The supplied backing file has to exist
269	 */
270	fd = open(opts, O_RDWR);
271	if (fd < 0) {
272		perror("Could not open backing file");
273		return (1);
274	}
275
276	if (fstat(fd, &sbuf) < 0) {
277		perror("Could not stat backing file");
278		close(fd);
279		return (1);
280	}
281
282	/*
283	 * Deal with raw devices
284	 */
285	size = sbuf.st_size;
286	sectsz = DEV_BSIZE;
287	if (S_ISCHR(sbuf.st_mode)) {
288		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
289		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
290			perror("Could not fetch dev blk/sector size");
291			close(fd);
292			return (1);
293		}
294		assert(size != 0);
295		assert(sectsz != 0);
296	}
297
298	sc = malloc(sizeof(struct pci_vtblk_softc));
299	memset(sc, 0, sizeof(struct pci_vtblk_softc));
300
301	/* record fd of storage device/file */
302	sc->vbsc_fd = fd;
303
304	/* init virtio softc and virtqueues */
305	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
306	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
307	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
308
309	/*
310	 * Create an identifier for the backing file. Use parts of the
311	 * md5 sum of the filename
312	 */
313	MD5Init(&mdctx);
314	MD5Update(&mdctx, opts, strlen(opts));
315	MD5Final(digest, &mdctx);
316	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
317	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
318
319	/* setup virtio block config space */
320	sc->vbsc_cfg.vbc_capacity = size / sectsz;
321	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
322	sc->vbsc_cfg.vbc_blk_size = sectsz;
323	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
324	sc->vbsc_cfg.vbc_geom_c = 0;	/* no geometry */
325	sc->vbsc_cfg.vbc_geom_h = 0;
326	sc->vbsc_cfg.vbc_geom_s = 0;
327	sc->vbsc_cfg.vbc_sectors_max = 0;
328
329	/*
330	 * Should we move some of this into virtio.c?  Could
331	 * have the device, class, and subdev_0 as fields in
332	 * the virtio constants structure.
333	 */
334	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
335	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
336	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
337	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
338
339	use_msix = 1;
340	if ((env_msi = getenv("BHYVE_USE_MSI"))) {
341		if (strcasecmp(env_msi, "yes") == 0)
342			use_msix = 0;
343	}
344	if (vi_intr_init(&sc->vbsc_vs, 1, use_msix))
345		return (1);
346	vi_set_io_bar(&sc->vbsc_vs, 0);
347	return (0);
348}
349
350static int
351pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
352{
353
354	DPRINTF(("vtblk: write to readonly reg %d\n\r", offset));
355	return (1);
356}
357
358static int
359pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
360{
361	struct pci_vtblk_softc *sc = vsc;
362	void *ptr;
363
364	/* our caller has already verified offset and size */
365	ptr = (uint8_t *)&sc->vbsc_cfg + offset;
366	memcpy(retval, ptr, size);
367	return (0);
368}
369
370struct pci_devemu pci_de_vblk = {
371	.pe_emu =	"virtio-blk",
372	.pe_init =	pci_vtblk_init,
373	.pe_barwrite =	vi_pci_write,
374	.pe_barread =	vi_pci_read
375};
376PCI_EMUL_SET(pci_de_vblk);
377