pci_virtio_block.c revision 268976 
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 268976 2014-07-22 04:39:16Z jhb $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 268976 2014-07-22 04:39:16Z jhb $");
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#define VTBLK_RINGSZ	64
56
57#define VTBLK_MAXSEGS	32
58
59#define VTBLK_S_OK	0
60#define VTBLK_S_IOERR	1
61#define	VTBLK_S_UNSUPP	2
62
63#define	VTBLK_BLK_ID_BYTES	20
64
65/* Capability bits */
66#define	VTBLK_F_SEG_MAX		(1 << 2)	/* Maximum request segments */
67#define	VTBLK_F_BLK_SIZE       	(1 << 6)	/* cfg block size valid */
68
69/*
70 * Host capabilities
71 */
72#define VTBLK_S_HOSTCAPS      \
73  ( VTBLK_F_SEG_MAX  |						    \
74    VTBLK_F_BLK_SIZE |						    \
75    VIRTIO_RING_F_INDIRECT_DESC )	/* indirect descriptors */
76
77/*
78 * Config space "registers"
79 */
80struct vtblk_config {
81	uint64_t	vbc_capacity;
82	uint32_t	vbc_size_max;
83	uint32_t	vbc_seg_max;
84	uint16_t	vbc_geom_c;
85	uint8_t		vbc_geom_h;
86	uint8_t		vbc_geom_s;
87	uint32_t	vbc_blk_size;
88	uint32_t	vbc_sectors_max;
89} __packed;
90
91/*
92 * Fixed-size block header
93 */
94struct virtio_blk_hdr {
95#define	VBH_OP_READ		0
96#define	VBH_OP_WRITE		1
97#define	VBH_OP_IDENT		8
98#define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
99	uint32_t       	vbh_type;
100	uint32_t	vbh_ioprio;
101	uint64_t	vbh_sector;
102} __packed;
103
104/*
105 * Debug printf
106 */
107static int pci_vtblk_debug;
108#define DPRINTF(params) if (pci_vtblk_debug) printf params
109#define WPRINTF(params) printf params
110
111/*
112 * Per-device softc
113 */
114struct pci_vtblk_softc {
115	struct virtio_softc vbsc_vs;
116	pthread_mutex_t vsc_mtx;
117	struct vqueue_info vbsc_vq;
118	int		vbsc_fd;
119	struct vtblk_config vbsc_cfg;
120	char vbsc_ident[VTBLK_BLK_ID_BYTES];
121};
122
123static void pci_vtblk_reset(void *);
124static void pci_vtblk_notify(void *, struct vqueue_info *);
125static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
126static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
127
128static struct virtio_consts vtblk_vi_consts = {
129	"vtblk",		/* our name */
130	1,			/* we support 1 virtqueue */
131	sizeof(struct vtblk_config), /* config reg size */
132	pci_vtblk_reset,	/* reset */
133	pci_vtblk_notify,	/* device-wide qnotify */
134	pci_vtblk_cfgread,	/* read PCI config */
135	pci_vtblk_cfgwrite,	/* write PCI config */
136	VTBLK_S_HOSTCAPS,	/* our capabilities */
137};
138
139static void
140pci_vtblk_reset(void *vsc)
141{
142	struct pci_vtblk_softc *sc = vsc;
143
144	DPRINTF(("vtblk: device reset requested !\n"));
145	vi_reset_dev(&sc->vbsc_vs);
146}
147
148static void
149pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
150{
151	struct virtio_blk_hdr *vbh;
152	uint8_t *status;
153	int i, n;
154	int err;
155	int iolen;
156	int writeop, type;
157	off_t offset;
158	struct iovec iov[VTBLK_MAXSEGS + 2];
159	uint16_t flags[VTBLK_MAXSEGS + 2];
160
161	n = vq_getchain(vq, iov, VTBLK_MAXSEGS + 2, flags);
162
163	/*
164	 * The first descriptor will be the read-only fixed header,
165	 * and the last is for status (hence +2 above and below).
166	 * The remaining iov's are the actual data I/O vectors.
167	 *
168	 * XXX - note - this fails on crash dump, which does a
169	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
170	 */
171	assert(n >= 2 && n <= VTBLK_MAXSEGS + 2);
172
173	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
174	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
175	vbh = iov[0].iov_base;
176
177	status = iov[--n].iov_base;
178	assert(iov[n].iov_len == 1);
179	assert(flags[n] & VRING_DESC_F_WRITE);
180
181	/*
182	 * XXX
183	 * The guest should not be setting the BARRIER flag because
184	 * we don't advertise the capability.
185	 */
186	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
187	writeop = (type == VBH_OP_WRITE);
188
189	offset = vbh->vbh_sector * DEV_BSIZE;
190
191	iolen = 0;
192	for (i = 1; i < n; i++) {
193		/*
194		 * - write op implies read-only descriptor,
195		 * - read/ident op implies write-only descriptor,
196		 * therefore test the inverse of the descriptor bit
197		 * to the op.
198		 */
199		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
200		iolen += iov[i].iov_len;
201	}
202
203	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
204		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
205
206	switch (type) {
207	case VBH_OP_WRITE:
208		err = pwritev(sc->vbsc_fd, iov + 1, i - 1, offset);
209		break;
210	case VBH_OP_READ:
211		err = preadv(sc->vbsc_fd, iov + 1, i - 1, offset);
212		break;
213	case VBH_OP_IDENT:
214		/* Assume a single buffer */
215		strlcpy(iov[1].iov_base, sc->vbsc_ident,
216		    MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
217		err = 0;
218		break;
219	default:
220		err = -ENOSYS;
221		break;
222	}
223
224	/* convert errno into a virtio block error return */
225	if (err < 0) {
226		if (err == -ENOSYS)
227			*status = VTBLK_S_UNSUPP;
228		else
229			*status = VTBLK_S_IOERR;
230	} else
231		*status = VTBLK_S_OK;
232
233	/*
234	 * Return the descriptor back to the host.
235	 * We wrote 1 byte (our status) to host.
236	 */
237	vq_relchain(vq, 1);
238}
239
240static void
241pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
242{
243	struct pci_vtblk_softc *sc = vsc;
244
245	vq_startchains(vq);
246	while (vq_has_descs(vq))
247		pci_vtblk_proc(sc, vq);
248	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
249}
250
251static int
252pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
253{
254	struct stat sbuf;
255	MD5_CTX mdctx;
256	u_char digest[16];
257	struct pci_vtblk_softc *sc;
258	off_t size;
259	int fd;
260	int sectsz;
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 = calloc(1, sizeof(struct pci_vtblk_softc));
299
300	/* record fd of storage device/file */
301	sc->vbsc_fd = fd;
302
303	pthread_mutex_init(&sc->vsc_mtx, NULL);
304
305	/* init virtio softc and virtqueues */
306	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
307	sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
308
309	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
310	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
311
312	/*
313	 * Create an identifier for the backing file. Use parts of the
314	 * md5 sum of the filename
315	 */
316	MD5Init(&mdctx);
317	MD5Update(&mdctx, opts, strlen(opts));
318	MD5Final(digest, &mdctx);
319	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
320	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
321
322	/* setup virtio block config space */
323	sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */
324	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
325	sc->vbsc_cfg.vbc_blk_size = sectsz;
326	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
327	sc->vbsc_cfg.vbc_geom_c = 0;	/* no geometry */
328	sc->vbsc_cfg.vbc_geom_h = 0;
329	sc->vbsc_cfg.vbc_geom_s = 0;
330	sc->vbsc_cfg.vbc_sectors_max = 0;
331
332	/*
333	 * Should we move some of this into virtio.c?  Could
334	 * have the device, class, and subdev_0 as fields in
335	 * the virtio constants structure.
336	 */
337	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
338	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
339	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
340	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
341
342	pci_lintr_request(pi);
343
344	if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_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