pci_virtio_block.c revision 280747 
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 280747 2015-03-27 08:56:44Z mav $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 280747 2015-03-27 08:56:44Z 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#include "block_if.h"
55
56#define VTBLK_RINGSZ	64
57
58#define VTBLK_MAXSEGS	32
59
60#define VTBLK_S_OK	0
61#define VTBLK_S_IOERR	1
62#define	VTBLK_S_UNSUPP	2
63
64#define	VTBLK_BLK_ID_BYTES	20
65
66/* Capability bits */
67#define	VTBLK_F_SEG_MAX		(1 << 2)	/* Maximum request segments */
68#define	VTBLK_F_BLK_SIZE	(1 << 6)	/* cfg block size valid */
69#define	VTBLK_F_TOPOLOGY	(1 << 10)	/* Optimal I/O alignment */
70
71/*
72 * Host capabilities
73 */
74#define VTBLK_S_HOSTCAPS      \
75  ( VTBLK_F_SEG_MAX  |						    \
76    VTBLK_F_BLK_SIZE |						    \
77    VTBLK_F_TOPOLOGY |						    \
78    VIRTIO_RING_F_INDIRECT_DESC )	/* indirect descriptors */
79
80/*
81 * Config space "registers"
82 */
83struct vtblk_config {
84	uint64_t	vbc_capacity;
85	uint32_t	vbc_size_max;
86	uint32_t	vbc_seg_max;
87	struct {
88		uint16_t cylinders;
89		uint8_t heads;
90		uint8_t sectors;
91	} vbc_geometry;
92	uint32_t	vbc_blk_size;
93	struct {
94		uint8_t physical_block_exp;
95		uint8_t alignment_offset;
96		uint16_t min_io_size;
97		uint32_t opt_io_size;
98	} vbc_topology;
99	uint8_t		vbc_writeback;
100} __packed;
101
102/*
103 * Fixed-size block header
104 */
105struct virtio_blk_hdr {
106#define	VBH_OP_READ		0
107#define	VBH_OP_WRITE		1
108#define	VBH_OP_FLUSH		4
109#define	VBH_OP_FLUSH_OUT	5
110#define	VBH_OP_IDENT		8
111#define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
112	uint32_t       	vbh_type;
113	uint32_t	vbh_ioprio;
114	uint64_t	vbh_sector;
115} __packed;
116
117/*
118 * Debug printf
119 */
120static int pci_vtblk_debug;
121#define DPRINTF(params) if (pci_vtblk_debug) printf params
122#define WPRINTF(params) printf params
123
124struct pci_vtblk_ioreq {
125	struct blockif_req		io_req;
126	struct pci_vtblk_softc	       *io_sc;
127	uint8_t			       *io_status;
128	uint16_t			io_idx;
129};
130
131/*
132 * Per-device softc
133 */
134struct pci_vtblk_softc {
135	struct virtio_softc vbsc_vs;
136	pthread_mutex_t vsc_mtx;
137	struct vqueue_info vbsc_vq;
138	struct vtblk_config vbsc_cfg;
139	struct blockif_ctxt *bc;
140	char vbsc_ident[VTBLK_BLK_ID_BYTES];
141	struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ];
142};
143
144static void pci_vtblk_reset(void *);
145static void pci_vtblk_notify(void *, struct vqueue_info *);
146static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
147static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
148
149static struct virtio_consts vtblk_vi_consts = {
150	"vtblk",		/* our name */
151	1,			/* we support 1 virtqueue */
152	sizeof(struct vtblk_config), /* config reg size */
153	pci_vtblk_reset,	/* reset */
154	pci_vtblk_notify,	/* device-wide qnotify */
155	pci_vtblk_cfgread,	/* read PCI config */
156	pci_vtblk_cfgwrite,	/* write PCI config */
157	NULL,			/* apply negotiated features */
158	VTBLK_S_HOSTCAPS,	/* our capabilities */
159};
160
161static void
162pci_vtblk_reset(void *vsc)
163{
164	struct pci_vtblk_softc *sc = vsc;
165
166	DPRINTF(("vtblk: device reset requested !\n"));
167	vi_reset_dev(&sc->vbsc_vs);
168}
169
170static void
171pci_vtblk_done(struct blockif_req *br, int err)
172{
173	struct pci_vtblk_ioreq *io = br->br_param;
174	struct pci_vtblk_softc *sc = io->io_sc;
175
176	/* convert errno into a virtio block error return */
177	if (err == EOPNOTSUPP || err == ENOSYS)
178		*io->io_status = VTBLK_S_UNSUPP;
179	else if (err != 0)
180		*io->io_status = VTBLK_S_IOERR;
181	else
182		*io->io_status = VTBLK_S_OK;
183
184	/*
185	 * Return the descriptor back to the host.
186	 * We wrote 1 byte (our status) to host.
187	 */
188	pthread_mutex_lock(&sc->vsc_mtx);
189	vq_relchain(&sc->vbsc_vq, io->io_idx, 1);
190	vq_endchains(&sc->vbsc_vq, 0);
191	pthread_mutex_unlock(&sc->vsc_mtx);
192}
193
194static void
195pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
196{
197	struct virtio_blk_hdr *vbh;
198	struct pci_vtblk_ioreq *io;
199	int i, n;
200	int err;
201	int iolen;
202	int writeop, type;
203	off_t offset;
204	struct iovec iov[VTBLK_MAXSEGS + 2];
205	uint16_t idx, flags[VTBLK_MAXSEGS + 2];
206
207	n = vq_getchain(vq, &idx, iov, VTBLK_MAXSEGS + 2, flags);
208
209	/*
210	 * The first descriptor will be the read-only fixed header,
211	 * and the last is for status (hence +2 above and below).
212	 * The remaining iov's are the actual data I/O vectors.
213	 *
214	 * XXX - note - this fails on crash dump, which does a
215	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
216	 */
217	assert(n >= 2 && n <= VTBLK_MAXSEGS + 2);
218
219	io = &sc->vbsc_ios[idx];
220	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
221	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
222	vbh = iov[0].iov_base;
223	memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2));
224	io->io_req.br_iovcnt = n - 2;
225	io->io_req.br_offset = vbh->vbh_sector * DEV_BSIZE;
226	io->io_status = iov[--n].iov_base;
227	assert(iov[n].iov_len == 1);
228	assert(flags[n] & VRING_DESC_F_WRITE);
229
230	/*
231	 * XXX
232	 * The guest should not be setting the BARRIER flag because
233	 * we don't advertise the capability.
234	 */
235	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
236	writeop = (type == VBH_OP_WRITE);
237
238	iolen = 0;
239	for (i = 1; i < n; i++) {
240		/*
241		 * - write op implies read-only descriptor,
242		 * - read/ident op implies write-only descriptor,
243		 * therefore test the inverse of the descriptor bit
244		 * to the op.
245		 */
246		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
247		iolen += iov[i].iov_len;
248	}
249
250	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
251		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
252
253	switch (type) {
254	case VBH_OP_READ:
255		err = blockif_read(sc->bc, &io->io_req);
256		break;
257	case VBH_OP_WRITE:
258		err = blockif_write(sc->bc, &io->io_req);
259		break;
260	case VBH_OP_FLUSH:
261	case VBH_OP_FLUSH_OUT:
262		err = blockif_flush(sc->bc, &io->io_req);
263		break;
264	case VBH_OP_IDENT:
265		/* Assume a single buffer */
266		/* S/n equal to buffer is not zero-terminated. */
267		memset(iov[1].iov_base, 0, iov[1].iov_len);
268		strncpy(iov[1].iov_base, sc->vbsc_ident,
269		    MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
270		pci_vtblk_done(&io->io_req, 0);
271		return;
272	default:
273		pci_vtblk_done(&io->io_req, EOPNOTSUPP);
274		return;
275	}
276	assert(err == 0);
277}
278
279static void
280pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
281{
282	struct pci_vtblk_softc *sc = vsc;
283
284	while (vq_has_descs(vq))
285		pci_vtblk_proc(sc, vq);
286}
287
288static int
289pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
290{
291	char bident[sizeof("XX:X:X")];
292	struct blockif_ctxt *bctxt;
293	MD5_CTX mdctx;
294	u_char digest[16];
295	struct pci_vtblk_softc *sc;
296	off_t size;
297	int i, sectsz, sts, sto;
298
299	if (opts == NULL) {
300		printf("virtio-block: backing device required\n");
301		return (1);
302	}
303
304	/*
305	 * The supplied backing file has to exist
306	 */
307	snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func);
308	bctxt = blockif_open(opts, bident);
309	if (bctxt == NULL) {
310		perror("Could not open backing file");
311		return (1);
312	}
313
314	size = blockif_size(bctxt);
315	sectsz = blockif_sectsz(bctxt);
316	blockif_psectsz(bctxt, &sts, &sto);
317
318	sc = calloc(1, sizeof(struct pci_vtblk_softc));
319	sc->bc = bctxt;
320	for (i = 0; i < VTBLK_RINGSZ; i++) {
321		struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i];
322		io->io_req.br_callback = pci_vtblk_done;
323		io->io_req.br_param = io;
324		io->io_sc = sc;
325		io->io_idx = i;
326	}
327
328	pthread_mutex_init(&sc->vsc_mtx, NULL);
329
330	/* init virtio softc and virtqueues */
331	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
332	sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
333
334	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
335	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
336
337	/*
338	 * Create an identifier for the backing file. Use parts of the
339	 * md5 sum of the filename
340	 */
341	MD5Init(&mdctx);
342	MD5Update(&mdctx, opts, strlen(opts));
343	MD5Final(digest, &mdctx);
344	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
345	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
346
347	/* setup virtio block config space */
348	sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */
349	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
350	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
351	sc->vbsc_cfg.vbc_geometry.cylinders = 0;	/* no geometry */
352	sc->vbsc_cfg.vbc_geometry.heads = 0;
353	sc->vbsc_cfg.vbc_geometry.sectors = 0;
354	sc->vbsc_cfg.vbc_blk_size = sectsz;
355	sc->vbsc_cfg.vbc_topology.physical_block_exp =
356	    (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
357	sc->vbsc_cfg.vbc_topology.alignment_offset =
358	    (sto != 0) ? ((sts - sto) / sectsz) : 0;
359	sc->vbsc_cfg.vbc_topology.min_io_size = 0;
360	sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
361	sc->vbsc_cfg.vbc_writeback = 0;
362
363	/*
364	 * Should we move some of this into virtio.c?  Could
365	 * have the device, class, and subdev_0 as fields in
366	 * the virtio constants structure.
367	 */
368	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
369	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
370	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
371	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
372
373	pci_lintr_request(pi);
374
375	if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) {
376		blockif_close(sc->bc);
377		free(sc);
378		return (1);
379	}
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