pci_virtio_block.c revision 280243 
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 280243 2015-03-19 09:53:00Z mav $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 280243 2015-03-19 09:53:00Z mav $");
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#define	VTBLK_F_TOPOLOGY	(1 << 10)	/* Optimal I/O alignment */
69
70/*
71 * Host capabilities
72 */
73#define VTBLK_S_HOSTCAPS      \
74  ( VTBLK_F_SEG_MAX  |						    \
75    VTBLK_F_BLK_SIZE |						    \
76    VTBLK_F_TOPOLOGY |						    \
77    VIRTIO_RING_F_INDIRECT_DESC )	/* indirect descriptors */
78
79/*
80 * Config space "registers"
81 */
82struct vtblk_config {
83	uint64_t	vbc_capacity;
84	uint32_t	vbc_size_max;
85	uint32_t	vbc_seg_max;
86	struct {
87		uint16_t cylinders;
88		uint8_t heads;
89		uint8_t sectors;
90	} vbc_geometry;
91	uint32_t	vbc_blk_size;
92	struct {
93		uint8_t physical_block_exp;
94		uint8_t alignment_offset;
95		uint16_t min_io_size;
96		uint32_t opt_io_size;
97	} vbc_topology;
98	uint8_t		vbc_writeback;
99} __packed;
100
101/*
102 * Fixed-size block header
103 */
104struct virtio_blk_hdr {
105#define	VBH_OP_READ		0
106#define	VBH_OP_WRITE		1
107#define	VBH_OP_FLUSH		4
108#define	VBH_OP_FLUSH_OUT	5
109#define	VBH_OP_IDENT		8
110#define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
111	uint32_t       	vbh_type;
112	uint32_t	vbh_ioprio;
113	uint64_t	vbh_sector;
114} __packed;
115
116/*
117 * Debug printf
118 */
119static int pci_vtblk_debug;
120#define DPRINTF(params) if (pci_vtblk_debug) printf params
121#define WPRINTF(params) printf params
122
123/*
124 * Per-device softc
125 */
126struct pci_vtblk_softc {
127	struct virtio_softc vbsc_vs;
128	pthread_mutex_t vsc_mtx;
129	struct vqueue_info vbsc_vq;
130	int		vbsc_fd;
131	struct vtblk_config vbsc_cfg;
132	char vbsc_ident[VTBLK_BLK_ID_BYTES];
133};
134
135static void pci_vtblk_reset(void *);
136static void pci_vtblk_notify(void *, struct vqueue_info *);
137static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
138static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
139
140static struct virtio_consts vtblk_vi_consts = {
141	"vtblk",		/* our name */
142	1,			/* we support 1 virtqueue */
143	sizeof(struct vtblk_config), /* config reg size */
144	pci_vtblk_reset,	/* reset */
145	pci_vtblk_notify,	/* device-wide qnotify */
146	pci_vtblk_cfgread,	/* read PCI config */
147	pci_vtblk_cfgwrite,	/* write PCI config */
148	NULL,			/* apply negotiated features */
149	VTBLK_S_HOSTCAPS,	/* our capabilities */
150};
151
152static void
153pci_vtblk_reset(void *vsc)
154{
155	struct pci_vtblk_softc *sc = vsc;
156
157	DPRINTF(("vtblk: device reset requested !\n"));
158	vi_reset_dev(&sc->vbsc_vs);
159}
160
161static void
162pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
163{
164	struct virtio_blk_hdr *vbh;
165	uint8_t *status;
166	int i, n;
167	int err;
168	int iolen;
169	int writeop, type;
170	off_t offset;
171	struct iovec iov[VTBLK_MAXSEGS + 2];
172	uint16_t flags[VTBLK_MAXSEGS + 2];
173
174	n = vq_getchain(vq, iov, VTBLK_MAXSEGS + 2, flags);
175
176	/*
177	 * The first descriptor will be the read-only fixed header,
178	 * and the last is for status (hence +2 above and below).
179	 * The remaining iov's are the actual data I/O vectors.
180	 *
181	 * XXX - note - this fails on crash dump, which does a
182	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
183	 */
184	assert(n >= 2 && n <= VTBLK_MAXSEGS + 2);
185
186	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
187	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
188	vbh = iov[0].iov_base;
189
190	status = iov[--n].iov_base;
191	assert(iov[n].iov_len == 1);
192	assert(flags[n] & VRING_DESC_F_WRITE);
193
194	/*
195	 * XXX
196	 * The guest should not be setting the BARRIER flag because
197	 * we don't advertise the capability.
198	 */
199	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
200	writeop = (type == VBH_OP_WRITE);
201
202	offset = vbh->vbh_sector * DEV_BSIZE;
203
204	iolen = 0;
205	for (i = 1; i < n; i++) {
206		/*
207		 * - write op implies read-only descriptor,
208		 * - read/ident op implies write-only descriptor,
209		 * therefore test the inverse of the descriptor bit
210		 * to the op.
211		 */
212		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
213		iolen += iov[i].iov_len;
214	}
215
216	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
217		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
218
219	switch (type) {
220	case VBH_OP_WRITE:
221		err = pwritev(sc->vbsc_fd, iov + 1, i - 1, offset);
222		break;
223	case VBH_OP_READ:
224		err = preadv(sc->vbsc_fd, iov + 1, i - 1, offset);
225		break;
226	case VBH_OP_IDENT:
227		/* Assume a single buffer */
228		strlcpy(iov[1].iov_base, sc->vbsc_ident,
229		    MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
230		err = 0;
231		break;
232	case VBH_OP_FLUSH:
233	case VBH_OP_FLUSH_OUT:
234		err = fsync(sc->vbsc_fd);
235		break;
236	default:
237		err = -ENOSYS;
238		break;
239	}
240
241	/* convert errno into a virtio block error return */
242	if (err < 0) {
243		if (err == -ENOSYS)
244			*status = VTBLK_S_UNSUPP;
245		else
246			*status = VTBLK_S_IOERR;
247	} else
248		*status = VTBLK_S_OK;
249
250	/*
251	 * Return the descriptor back to the host.
252	 * We wrote 1 byte (our status) to host.
253	 */
254	vq_relchain(vq, 1);
255}
256
257static void
258pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
259{
260	struct pci_vtblk_softc *sc = vsc;
261
262	vq_startchains(vq);
263	while (vq_has_descs(vq))
264		pci_vtblk_proc(sc, vq);
265	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
266}
267
268static int
269pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
270{
271	struct stat sbuf;
272	MD5_CTX mdctx;
273	u_char digest[16];
274	struct pci_vtblk_softc *sc;
275	off_t size, sts, sto;
276	int fd;
277	int sectsz;
278
279	if (opts == NULL) {
280		printf("virtio-block: backing device required\n");
281		return (1);
282	}
283
284	/*
285	 * The supplied backing file has to exist
286	 */
287	fd = open(opts, O_RDWR);
288	if (fd < 0) {
289		perror("Could not open backing file");
290		return (1);
291	}
292
293	if (fstat(fd, &sbuf) < 0) {
294		perror("Could not stat backing file");
295		close(fd);
296		return (1);
297	}
298
299	/*
300	 * Deal with raw devices
301	 */
302	size = sbuf.st_size;
303	sectsz = DEV_BSIZE;
304	sts = sto = 0;
305	if (S_ISCHR(sbuf.st_mode)) {
306		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
307		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
308			perror("Could not fetch dev blk/sector size");
309			close(fd);
310			return (1);
311		}
312		assert(size != 0);
313		assert(sectsz != 0);
314		if (ioctl(fd, DIOCGSTRIPESIZE, &sts) == 0 && sts > 0)
315			ioctl(fd, DIOCGSTRIPEOFFSET, &sto);
316	} else
317		sts = sbuf.st_blksize;
318
319	sc = calloc(1, sizeof(struct pci_vtblk_softc));
320
321	/* record fd of storage device/file */
322	sc->vbsc_fd = fd;
323
324	pthread_mutex_init(&sc->vsc_mtx, NULL);
325
326	/* init virtio softc and virtqueues */
327	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
328	sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
329
330	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
331	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
332
333	/*
334	 * Create an identifier for the backing file. Use parts of the
335	 * md5 sum of the filename
336	 */
337	MD5Init(&mdctx);
338	MD5Update(&mdctx, opts, strlen(opts));
339	MD5Final(digest, &mdctx);
340	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
341	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
342
343	/* setup virtio block config space */
344	sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */
345	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
346	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
347	sc->vbsc_cfg.vbc_geometry.cylinders = 0;	/* no geometry */
348	sc->vbsc_cfg.vbc_geometry.heads = 0;
349	sc->vbsc_cfg.vbc_geometry.sectors = 0;
350	sc->vbsc_cfg.vbc_blk_size = sectsz;
351	sc->vbsc_cfg.vbc_topology.physical_block_exp =
352	    (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
353	sc->vbsc_cfg.vbc_topology.alignment_offset =
354	    (sto != 0) ? ((sts - sto) / sectsz) : 0;
355	sc->vbsc_cfg.vbc_topology.min_io_size = 0;
356	sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
357	sc->vbsc_cfg.vbc_writeback = 0;
358
359	/*
360	 * Should we move some of this into virtio.c?  Could
361	 * have the device, class, and subdev_0 as fields in
362	 * the virtio constants structure.
363	 */
364	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
365	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
366	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
367	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
368
369	pci_lintr_request(pi);
370
371	if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix()))
372		return (1);
373	vi_set_io_bar(&sc->vbsc_vs, 0);
374	return (0);
375}
376
377static int
378pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
379{
380
381	DPRINTF(("vtblk: write to readonly reg %d\n\r", offset));
382	return (1);
383}
384
385static int
386pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
387{
388	struct pci_vtblk_softc *sc = vsc;
389	void *ptr;
390
391	/* our caller has already verified offset and size */
392	ptr = (uint8_t *)&sc->vbsc_cfg + offset;
393	memcpy(retval, ptr, size);
394	return (0);
395}
396
397struct pci_devemu pci_de_vblk = {
398	.pe_emu =	"virtio-blk",
399	.pe_init =	pci_vtblk_init,
400	.pe_barwrite =	vi_pci_write,
401	.pe_barread =	vi_pci_read
402};
403PCI_EMUL_SET(pci_de_vblk);
404