pci_virtio_block.c revision 276349 
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 276349 2014-12-28 21:27:13Z neel $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 276349 2014-12-28 21:27:13Z neel $");
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_FLUSH		4
98#define	VBH_OP_FLUSH_OUT	5
99#define	VBH_OP_IDENT		8
100#define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
101	uint32_t       	vbh_type;
102	uint32_t	vbh_ioprio;
103	uint64_t	vbh_sector;
104} __packed;
105
106/*
107 * Debug printf
108 */
109static int pci_vtblk_debug;
110#define DPRINTF(params) if (pci_vtblk_debug) printf params
111#define WPRINTF(params) printf params
112
113/*
114 * Per-device softc
115 */
116struct pci_vtblk_softc {
117	struct virtio_softc vbsc_vs;
118	pthread_mutex_t vsc_mtx;
119	struct vqueue_info vbsc_vq;
120	int		vbsc_fd;
121	struct vtblk_config vbsc_cfg;
122	char vbsc_ident[VTBLK_BLK_ID_BYTES];
123};
124
125static void pci_vtblk_reset(void *);
126static void pci_vtblk_notify(void *, struct vqueue_info *);
127static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
128static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
129
130static struct virtio_consts vtblk_vi_consts = {
131	"vtblk",		/* our name */
132	1,			/* we support 1 virtqueue */
133	sizeof(struct vtblk_config), /* config reg size */
134	pci_vtblk_reset,	/* reset */
135	pci_vtblk_notify,	/* device-wide qnotify */
136	pci_vtblk_cfgread,	/* read PCI config */
137	pci_vtblk_cfgwrite,	/* write PCI config */
138	NULL,			/* apply negotiated features */
139	VTBLK_S_HOSTCAPS,	/* our capabilities */
140};
141
142static void
143pci_vtblk_reset(void *vsc)
144{
145	struct pci_vtblk_softc *sc = vsc;
146
147	DPRINTF(("vtblk: device reset requested !\n"));
148	vi_reset_dev(&sc->vbsc_vs);
149}
150
151static void
152pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
153{
154	struct virtio_blk_hdr *vbh;
155	uint8_t *status;
156	int i, n;
157	int err;
158	int iolen;
159	int writeop, type;
160	off_t offset;
161	struct iovec iov[VTBLK_MAXSEGS + 2];
162	uint16_t flags[VTBLK_MAXSEGS + 2];
163
164	n = vq_getchain(vq, iov, VTBLK_MAXSEGS + 2, flags);
165
166	/*
167	 * The first descriptor will be the read-only fixed header,
168	 * and the last is for status (hence +2 above and below).
169	 * The remaining iov's are the actual data I/O vectors.
170	 *
171	 * XXX - note - this fails on crash dump, which does a
172	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
173	 */
174	assert(n >= 2 && n <= VTBLK_MAXSEGS + 2);
175
176	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
177	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
178	vbh = iov[0].iov_base;
179
180	status = iov[--n].iov_base;
181	assert(iov[n].iov_len == 1);
182	assert(flags[n] & VRING_DESC_F_WRITE);
183
184	/*
185	 * XXX
186	 * The guest should not be setting the BARRIER flag because
187	 * we don't advertise the capability.
188	 */
189	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
190	writeop = (type == VBH_OP_WRITE);
191
192	offset = vbh->vbh_sector * DEV_BSIZE;
193
194	iolen = 0;
195	for (i = 1; i < n; i++) {
196		/*
197		 * - write op implies read-only descriptor,
198		 * - read/ident op implies write-only descriptor,
199		 * therefore test the inverse of the descriptor bit
200		 * to the op.
201		 */
202		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
203		iolen += iov[i].iov_len;
204	}
205
206	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
207		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
208
209	switch (type) {
210	case VBH_OP_WRITE:
211		err = pwritev(sc->vbsc_fd, iov + 1, i - 1, offset);
212		break;
213	case VBH_OP_READ:
214		err = preadv(sc->vbsc_fd, iov + 1, i - 1, offset);
215		break;
216	case VBH_OP_IDENT:
217		/* Assume a single buffer */
218		strlcpy(iov[1].iov_base, sc->vbsc_ident,
219		    MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
220		err = 0;
221		break;
222	case VBH_OP_FLUSH:
223	case VBH_OP_FLUSH_OUT:
224		err = fsync(sc->vbsc_fd);
225		break;
226	default:
227		err = -ENOSYS;
228		break;
229	}
230
231	/* convert errno into a virtio block error return */
232	if (err < 0) {
233		if (err == -ENOSYS)
234			*status = VTBLK_S_UNSUPP;
235		else
236			*status = VTBLK_S_IOERR;
237	} else
238		*status = VTBLK_S_OK;
239
240	/*
241	 * Return the descriptor back to the host.
242	 * We wrote 1 byte (our status) to host.
243	 */
244	vq_relchain(vq, 1);
245}
246
247static void
248pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
249{
250	struct pci_vtblk_softc *sc = vsc;
251
252	vq_startchains(vq);
253	while (vq_has_descs(vq))
254		pci_vtblk_proc(sc, vq);
255	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
256}
257
258static int
259pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
260{
261	struct stat sbuf;
262	MD5_CTX mdctx;
263	u_char digest[16];
264	struct pci_vtblk_softc *sc;
265	off_t size;
266	int fd;
267	int sectsz;
268
269	if (opts == NULL) {
270		printf("virtio-block: backing device required\n");
271		return (1);
272	}
273
274	/*
275	 * The supplied backing file has to exist
276	 */
277	fd = open(opts, O_RDWR);
278	if (fd < 0) {
279		perror("Could not open backing file");
280		return (1);
281	}
282
283	if (fstat(fd, &sbuf) < 0) {
284		perror("Could not stat backing file");
285		close(fd);
286		return (1);
287	}
288
289	/*
290	 * Deal with raw devices
291	 */
292	size = sbuf.st_size;
293	sectsz = DEV_BSIZE;
294	if (S_ISCHR(sbuf.st_mode)) {
295		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
296		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
297			perror("Could not fetch dev blk/sector size");
298			close(fd);
299			return (1);
300		}
301		assert(size != 0);
302		assert(sectsz != 0);
303	}
304
305	sc = calloc(1, sizeof(struct pci_vtblk_softc));
306
307	/* record fd of storage device/file */
308	sc->vbsc_fd = fd;
309
310	pthread_mutex_init(&sc->vsc_mtx, NULL);
311
312	/* init virtio softc and virtqueues */
313	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
314	sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
315
316	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
317	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
318
319	/*
320	 * Create an identifier for the backing file. Use parts of the
321	 * md5 sum of the filename
322	 */
323	MD5Init(&mdctx);
324	MD5Update(&mdctx, opts, strlen(opts));
325	MD5Final(digest, &mdctx);
326	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
327	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
328
329	/* setup virtio block config space */
330	sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */
331	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
332	sc->vbsc_cfg.vbc_blk_size = sectsz;
333	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
334	sc->vbsc_cfg.vbc_geom_c = 0;	/* no geometry */
335	sc->vbsc_cfg.vbc_geom_h = 0;
336	sc->vbsc_cfg.vbc_geom_s = 0;
337	sc->vbsc_cfg.vbc_sectors_max = 0;
338
339	/*
340	 * Should we move some of this into virtio.c?  Could
341	 * have the device, class, and subdev_0 as fields in
342	 * the virtio constants structure.
343	 */
344	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
345	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
346	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
347	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
348
349	pci_lintr_request(pi);
350
351	if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix()))
352		return (1);
353	vi_set_io_bar(&sc->vbsc_vs, 0);
354	return (0);
355}
356
357static int
358pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
359{
360
361	DPRINTF(("vtblk: write to readonly reg %d\n\r", offset));
362	return (1);
363}
364
365static int
366pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
367{
368	struct pci_vtblk_softc *sc = vsc;
369	void *ptr;
370
371	/* our caller has already verified offset and size */
372	ptr = (uint8_t *)&sc->vbsc_cfg + offset;
373	memcpy(retval, ptr, size);
374	return (0);
375}
376
377struct pci_devemu pci_de_vblk = {
378	.pe_emu =	"virtio-blk",
379	.pe_init =	pci_vtblk_init,
380	.pe_barwrite =	vi_pci_write,
381	.pe_barread =	vi_pci_read
382};
383PCI_EMUL_SET(pci_de_vblk);
384