pci_virtio_block.c revision 280749
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 280749 2015-03-27 08:58:30Z mav $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_block.c 280749 2015-03-27 08:58:30Z 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_S_OK	0
59#define VTBLK_S_IOERR	1
60#define	VTBLK_S_UNSUPP	2
61
62#define	VTBLK_BLK_ID_BYTES	20
63
64/* Capability bits */
65#define	VTBLK_F_SEG_MAX		(1 << 2)	/* Maximum request segments */
66#define	VTBLK_F_BLK_SIZE	(1 << 6)	/* cfg block size valid */
67#define	VTBLK_F_TOPOLOGY	(1 << 10)	/* Optimal I/O alignment */
68
69/*
70 * Host capabilities
71 */
72#define VTBLK_S_HOSTCAPS      \
73  ( VTBLK_F_SEG_MAX  |						    \
74    VTBLK_F_BLK_SIZE |						    \
75    VTBLK_F_TOPOLOGY |						    \
76    VIRTIO_RING_F_INDIRECT_DESC )	/* indirect descriptors */
77
78/*
79 * Config space "registers"
80 */
81struct vtblk_config {
82	uint64_t	vbc_capacity;
83	uint32_t	vbc_size_max;
84	uint32_t	vbc_seg_max;
85	struct {
86		uint16_t cylinders;
87		uint8_t heads;
88		uint8_t sectors;
89	} vbc_geometry;
90	uint32_t	vbc_blk_size;
91	struct {
92		uint8_t physical_block_exp;
93		uint8_t alignment_offset;
94		uint16_t min_io_size;
95		uint32_t opt_io_size;
96	} vbc_topology;
97	uint8_t		vbc_writeback;
98} __packed;
99
100/*
101 * Fixed-size block header
102 */
103struct virtio_blk_hdr {
104#define	VBH_OP_READ		0
105#define	VBH_OP_WRITE		1
106#define	VBH_OP_FLUSH		4
107#define	VBH_OP_FLUSH_OUT	5
108#define	VBH_OP_IDENT		8
109#define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
110	uint32_t       	vbh_type;
111	uint32_t	vbh_ioprio;
112	uint64_t	vbh_sector;
113} __packed;
114
115/*
116 * Debug printf
117 */
118static int pci_vtblk_debug;
119#define DPRINTF(params) if (pci_vtblk_debug) printf params
120#define WPRINTF(params) printf params
121
122struct pci_vtblk_ioreq {
123	struct blockif_req		io_req;
124	struct pci_vtblk_softc	       *io_sc;
125	uint8_t			       *io_status;
126	uint16_t			io_idx;
127};
128
129/*
130 * Per-device softc
131 */
132struct pci_vtblk_softc {
133	struct virtio_softc vbsc_vs;
134	pthread_mutex_t vsc_mtx;
135	struct vqueue_info vbsc_vq;
136	struct vtblk_config vbsc_cfg;
137	struct blockif_ctxt *bc;
138	char vbsc_ident[VTBLK_BLK_ID_BYTES];
139	struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ];
140};
141
142static void pci_vtblk_reset(void *);
143static void pci_vtblk_notify(void *, struct vqueue_info *);
144static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
145static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
146
147static struct virtio_consts vtblk_vi_consts = {
148	"vtblk",		/* our name */
149	1,			/* we support 1 virtqueue */
150	sizeof(struct vtblk_config), /* config reg size */
151	pci_vtblk_reset,	/* reset */
152	pci_vtblk_notify,	/* device-wide qnotify */
153	pci_vtblk_cfgread,	/* read PCI config */
154	pci_vtblk_cfgwrite,	/* write PCI config */
155	NULL,			/* apply negotiated features */
156	VTBLK_S_HOSTCAPS,	/* our capabilities */
157};
158
159static void
160pci_vtblk_reset(void *vsc)
161{
162	struct pci_vtblk_softc *sc = vsc;
163
164	DPRINTF(("vtblk: device reset requested !\n"));
165	vi_reset_dev(&sc->vbsc_vs);
166}
167
168static void
169pci_vtblk_done(struct blockif_req *br, int err)
170{
171	struct pci_vtblk_ioreq *io = br->br_param;
172	struct pci_vtblk_softc *sc = io->io_sc;
173
174	/* convert errno into a virtio block error return */
175	if (err == EOPNOTSUPP || err == ENOSYS)
176		*io->io_status = VTBLK_S_UNSUPP;
177	else if (err != 0)
178		*io->io_status = VTBLK_S_IOERR;
179	else
180		*io->io_status = VTBLK_S_OK;
181
182	/*
183	 * Return the descriptor back to the host.
184	 * We wrote 1 byte (our status) to host.
185	 */
186	pthread_mutex_lock(&sc->vsc_mtx);
187	vq_relchain(&sc->vbsc_vq, io->io_idx, 1);
188	vq_endchains(&sc->vbsc_vq, 0);
189	pthread_mutex_unlock(&sc->vsc_mtx);
190}
191
192static void
193pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
194{
195	struct virtio_blk_hdr *vbh;
196	struct pci_vtblk_ioreq *io;
197	int i, n;
198	int err;
199	int iolen;
200	int writeop, type;
201	off_t offset;
202	struct iovec iov[BLOCKIF_IOV_MAX + 2];
203	uint16_t idx, flags[BLOCKIF_IOV_MAX + 2];
204
205	n = vq_getchain(vq, &idx, iov, BLOCKIF_IOV_MAX + 2, flags);
206
207	/*
208	 * The first descriptor will be the read-only fixed header,
209	 * and the last is for status (hence +2 above and below).
210	 * The remaining iov's are the actual data I/O vectors.
211	 *
212	 * XXX - note - this fails on crash dump, which does a
213	 * VIRTIO_BLK_T_FLUSH with a zero transfer length
214	 */
215	assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2);
216
217	io = &sc->vbsc_ios[idx];
218	assert((flags[0] & VRING_DESC_F_WRITE) == 0);
219	assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
220	vbh = iov[0].iov_base;
221	memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2));
222	io->io_req.br_iovcnt = n - 2;
223	io->io_req.br_offset = vbh->vbh_sector * DEV_BSIZE;
224	io->io_status = iov[--n].iov_base;
225	assert(iov[n].iov_len == 1);
226	assert(flags[n] & VRING_DESC_F_WRITE);
227
228	/*
229	 * XXX
230	 * The guest should not be setting the BARRIER flag because
231	 * we don't advertise the capability.
232	 */
233	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
234	writeop = (type == VBH_OP_WRITE);
235
236	iolen = 0;
237	for (i = 1; i < n; i++) {
238		/*
239		 * - write op implies read-only descriptor,
240		 * - read/ident op implies write-only descriptor,
241		 * therefore test the inverse of the descriptor bit
242		 * to the op.
243		 */
244		assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
245		iolen += iov[i].iov_len;
246	}
247
248	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
249		 writeop ? "write" : "read/ident", iolen, i - 1, offset));
250
251	switch (type) {
252	case VBH_OP_READ:
253		err = blockif_read(sc->bc, &io->io_req);
254		break;
255	case VBH_OP_WRITE:
256		err = blockif_write(sc->bc, &io->io_req);
257		break;
258	case VBH_OP_FLUSH:
259	case VBH_OP_FLUSH_OUT:
260		err = blockif_flush(sc->bc, &io->io_req);
261		break;
262	case VBH_OP_IDENT:
263		/* Assume a single buffer */
264		/* S/n equal to buffer is not zero-terminated. */
265		memset(iov[1].iov_base, 0, iov[1].iov_len);
266		strncpy(iov[1].iov_base, sc->vbsc_ident,
267		    MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
268		pci_vtblk_done(&io->io_req, 0);
269		return;
270	default:
271		pci_vtblk_done(&io->io_req, EOPNOTSUPP);
272		return;
273	}
274	assert(err == 0);
275}
276
277static void
278pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
279{
280	struct pci_vtblk_softc *sc = vsc;
281
282	while (vq_has_descs(vq))
283		pci_vtblk_proc(sc, vq);
284}
285
286static int
287pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
288{
289	char bident[sizeof("XX:X:X")];
290	struct blockif_ctxt *bctxt;
291	MD5_CTX mdctx;
292	u_char digest[16];
293	struct pci_vtblk_softc *sc;
294	off_t size;
295	int i, sectsz, sts, sto;
296
297	if (opts == NULL) {
298		printf("virtio-block: backing device required\n");
299		return (1);
300	}
301
302	/*
303	 * The supplied backing file has to exist
304	 */
305	snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func);
306	bctxt = blockif_open(opts, bident);
307	if (bctxt == NULL) {
308		perror("Could not open backing file");
309		return (1);
310	}
311
312	size = blockif_size(bctxt);
313	sectsz = blockif_sectsz(bctxt);
314	blockif_psectsz(bctxt, &sts, &sto);
315
316	sc = calloc(1, sizeof(struct pci_vtblk_softc));
317	sc->bc = bctxt;
318	for (i = 0; i < VTBLK_RINGSZ; i++) {
319		struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i];
320		io->io_req.br_callback = pci_vtblk_done;
321		io->io_req.br_param = io;
322		io->io_sc = sc;
323		io->io_idx = i;
324	}
325
326	pthread_mutex_init(&sc->vsc_mtx, NULL);
327
328	/* init virtio softc and virtqueues */
329	vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq);
330	sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
331
332	sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
333	/* sc->vbsc_vq.vq_notify = we have no per-queue notify */
334
335	/*
336	 * Create an identifier for the backing file. Use parts of the
337	 * md5 sum of the filename
338	 */
339	MD5Init(&mdctx);
340	MD5Update(&mdctx, opts, strlen(opts));
341	MD5Final(digest, &mdctx);
342	sprintf(sc->vbsc_ident, "BHYVE-%02X%02X-%02X%02X-%02X%02X",
343	    digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
344
345	/* setup virtio block config space */
346	sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */
347	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
348	sc->vbsc_cfg.vbc_seg_max = BLOCKIF_IOV_MAX;
349	sc->vbsc_cfg.vbc_geometry.cylinders = 0;	/* no geometry */
350	sc->vbsc_cfg.vbc_geometry.heads = 0;
351	sc->vbsc_cfg.vbc_geometry.sectors = 0;
352	sc->vbsc_cfg.vbc_blk_size = sectsz;
353	sc->vbsc_cfg.vbc_topology.physical_block_exp =
354	    (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
355	sc->vbsc_cfg.vbc_topology.alignment_offset =
356	    (sto != 0) ? ((sts - sto) / sectsz) : 0;
357	sc->vbsc_cfg.vbc_topology.min_io_size = 0;
358	sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
359	sc->vbsc_cfg.vbc_writeback = 0;
360
361	/*
362	 * Should we move some of this into virtio.c?  Could
363	 * have the device, class, and subdev_0 as fields in
364	 * the virtio constants structure.
365	 */
366	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
367	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
368	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
369	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
370
371	pci_lintr_request(pi);
372
373	if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) {
374		blockif_close(sc->bc);
375		free(sc);
376		return (1);
377	}
378	vi_set_io_bar(&sc->vbsc_vs, 0);
379	return (0);
380}
381
382static int
383pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
384{
385
386	DPRINTF(("vtblk: write to readonly reg %d\n\r", offset));
387	return (1);
388}
389
390static int
391pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
392{
393	struct pci_vtblk_softc *sc = vsc;
394	void *ptr;
395
396	/* our caller has already verified offset and size */
397	ptr = (uint8_t *)&sc->vbsc_cfg + offset;
398	memcpy(retval, ptr, size);
399	return (0);
400}
401
402struct pci_devemu pci_de_vblk = {
403	.pe_emu =	"virtio-blk",
404	.pe_init =	pci_vtblk_init,
405	.pe_barwrite =	vi_pci_write,
406	.pe_barread =	vi_pci_read
407};
408PCI_EMUL_SET(pci_de_vblk);
409