pci_virtio_block.c revision 280743 
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 280743 2015-03-27 08:52:57Z mav $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 280743 2015-03-27 08:52:57Z 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	int		vbsc_ischr;
132	struct vtblk_config vbsc_cfg;
133	char vbsc_ident[VTBLK_BLK_ID_BYTES];
134};
135
136static void pci_vtblk_reset(void *);
137static void pci_vtblk_notify(void *, struct vqueue_info *);
138static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
139static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
140
141static struct virtio_consts vtblk_vi_consts = {
142	"vtblk",		/* our name */
143	1,			/* we support 1 virtqueue */
144	sizeof(struct vtblk_config), /* config reg size */
145	pci_vtblk_reset,	/* reset */
146	pci_vtblk_notify,	/* device-wide qnotify */
147	pci_vtblk_cfgread,	/* read PCI config */
148	pci_vtblk_cfgwrite,	/* write PCI config */
149	NULL,			/* apply negotiated features */
150	VTBLK_S_HOSTCAPS,	/* our capabilities */
151};
152
153static void
154pci_vtblk_reset(void *vsc)
155{
156	struct pci_vtblk_softc *sc = vsc;
157
158	DPRINTF(("vtblk: device reset requested !\n"));
159	vi_reset_dev(&sc->vbsc_vs);
160}
161
162static void
163pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
164{
165	struct virtio_blk_hdr *vbh;
166	uint8_t *status;
167	int i, n;
168	int err;
169	int iolen;
170	int writeop, type;
171	off_t offset;
172	struct iovec iov[VTBLK_MAXSEGS + 2];
173	uint16_t idx, flags[VTBLK_MAXSEGS + 2];
174
175	n = vq_getchain(vq, &idx, iov, VTBLK_MAXSEGS + 2, flags);
176
177	/*
178	 * The first descriptor will be the read-only fixed header,
179	 * and the last is for status (hence +2 above and below).
180	 * The remaining iov's are the actual data I/O vectors.
181	 *
182	 * XXX - note - this fails on crash dump, which does a
183	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
184	 */
185	assert(n >= 2 && n <= VTBLK_MAXSEGS + 2);
186
187	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
188	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
189	vbh = iov[0].iov_base;
190
191	status = iov[--n].iov_base;
192	assert(iov[n].iov_len == 1);
193	assert(flags[n] & VRING_DESC_F_WRITE);
194
195	/*
196	 * XXX
197	 * The guest should not be setting the BARRIER flag because
198	 * we don't advertise the capability.
199	 */
200	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
201	writeop = (type == VBH_OP_WRITE);
202
203	offset = vbh->vbh_sector * DEV_BSIZE;
204
205	iolen = 0;
206	for (i = 1; i < n; i++) {
207		/*
208		 * - write op implies read-only descriptor,
209		 * - read/ident op implies write-only descriptor,
210		 * therefore test the inverse of the descriptor bit
211		 * to the op.
212		 */
213		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
214		iolen += iov[i].iov_len;
215	}
216
217	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
218		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
219
220	err = 0;
221	switch (type) {
222	case VBH_OP_WRITE:
223		if (pwritev(sc->vbsc_fd, iov + 1, i - 1, offset) < 0)
224			err = errno;
225		break;
226	case VBH_OP_READ:
227		if (preadv(sc->vbsc_fd, iov + 1, i - 1, offset) < 0)
228			err = errno;
229		break;
230	case VBH_OP_IDENT:
231		/* Assume a single buffer */
232		strlcpy(iov[1].iov_base, sc->vbsc_ident,
233		    MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
234		err = 0;
235		break;
236	case VBH_OP_FLUSH:
237	case VBH_OP_FLUSH_OUT:
238		if (sc->vbsc_ischr) {
239			if (ioctl(sc->vbsc_fd, DIOCGFLUSH))
240				err = errno;
241		} else if (fsync(sc->vbsc_fd))
242			err = errno;
243		break;
244	default:
245		err = -ENOSYS;
246		break;
247	}
248
249	/* convert errno into a virtio block error return */
250	if (err == -ENOSYS)
251		*status = VTBLK_S_UNSUPP;
252	else if (err != 0)
253		*status = VTBLK_S_IOERR;
254	else
255		*status = VTBLK_S_OK;
256
257	/*
258	 * Return the descriptor back to the host.
259	 * We wrote 1 byte (our status) to host.
260	 */
261	vq_relchain(vq, idx, 1);
262}
263
264static void
265pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
266{
267	struct pci_vtblk_softc *sc = vsc;
268
269	while (vq_has_descs(vq))
270		pci_vtblk_proc(sc, vq);
271	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
272}
273
274static int
275pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
276{
277	struct stat sbuf;
278	MD5_CTX mdctx;
279	u_char digest[16];
280	struct pci_vtblk_softc *sc;
281	off_t size, sts, sto;
282	int fd;
283	int sectsz;
284
285	if (opts == NULL) {
286		printf("virtio-block: backing device required\n");
287		return (1);
288	}
289
290	/*
291	 * The supplied backing file has to exist
292	 */
293	fd = open(opts, O_RDWR);
294	if (fd < 0) {
295		perror("Could not open backing file");
296		return (1);
297	}
298
299	if (fstat(fd, &sbuf) < 0) {
300		perror("Could not stat backing file");
301		close(fd);
302		return (1);
303	}
304
305	/*
306	 * Deal with raw devices
307	 */
308	size = sbuf.st_size;
309	sectsz = DEV_BSIZE;
310	sts = sto = 0;
311	if (S_ISCHR(sbuf.st_mode)) {
312		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
313		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
314			perror("Could not fetch dev blk/sector size");
315			close(fd);
316			return (1);
317		}
318		assert(size != 0);
319		assert(sectsz != 0);
320		if (ioctl(fd, DIOCGSTRIPESIZE, &sts) == 0 && sts > 0)
321			ioctl(fd, DIOCGSTRIPEOFFSET, &sto);
322	} else
323		sts = sbuf.st_blksize;
324
325	sc = calloc(1, sizeof(struct pci_vtblk_softc));
326
327	/* record fd of storage device/file */
328	sc->vbsc_fd = fd;
329	sc->vbsc_ischr = S_ISCHR(sbuf.st_mode);
330
331	pthread_mutex_init(&sc->vsc_mtx, NULL);
332
333	/* init virtio softc and virtqueues */
334	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
335	sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
336
337	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
338	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
339
340	/*
341	 * Create an identifier for the backing file. Use parts of the
342	 * md5 sum of the filename
343	 */
344	MD5Init(&mdctx);
345	MD5Update(&mdctx, opts, strlen(opts));
346	MD5Final(digest, &mdctx);
347	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
348	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
349
350	/* setup virtio block config space */
351	sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */
352	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
353	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
354	sc->vbsc_cfg.vbc_geometry.cylinders = 0;	/* no geometry */
355	sc->vbsc_cfg.vbc_geometry.heads = 0;
356	sc->vbsc_cfg.vbc_geometry.sectors = 0;
357	sc->vbsc_cfg.vbc_blk_size = sectsz;
358	sc->vbsc_cfg.vbc_topology.physical_block_exp =
359	    (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
360	sc->vbsc_cfg.vbc_topology.alignment_offset =
361	    (sto != 0) ? ((sts - sto) / sectsz) : 0;
362	sc->vbsc_cfg.vbc_topology.min_io_size = 0;
363	sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
364	sc->vbsc_cfg.vbc_writeback = 0;
365
366	/*
367	 * Should we move some of this into virtio.c?  Could
368	 * have the device, class, and subdev_0 as fields in
369	 * the virtio constants structure.
370	 */
371	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
372	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
373	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
374	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
375
376	pci_lintr_request(pi);
377
378	if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix()))
379		return (1);
380	vi_set_io_bar(&sc->vbsc_vs, 0);
381	return (0);
382}
383
384static int
385pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
386{
387
388	DPRINTF(("vtblk: write to readonly reg %d\n\r", offset));
389	return (1);
390}
391
392static int
393pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
394{
395	struct pci_vtblk_softc *sc = vsc;
396	void *ptr;
397
398	/* our caller has already verified offset and size */
399	ptr = (uint8_t *)&sc->vbsc_cfg + offset;
400	memcpy(retval, ptr, size);
401	return (0);
402}
403
404struct pci_devemu pci_de_vblk = {
405	.pe_emu =	"virtio-blk",
406	.pe_init =	pci_vtblk_init,
407	.pe_barwrite =	vi_pci_write,
408	.pe_barread =	vi_pci_read
409};
410PCI_EMUL_SET(pci_de_vblk);
411