2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011 NetApp, Inc.
6 * Copyright 2020 Joyent, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
36 #include <sys/linker_set.h>
39 #include <sys/ioctl.h>
60 #define VTBLK_BSIZE 512
61 #define VTBLK_RINGSZ 128
63 _Static_assert(VTBLK_RINGSZ <= BLOCKIF_RING_MAX, "Each ring entry must be able to queue a request");
66 #define VTBLK_S_IOERR 1
67 #define VTBLK_S_UNSUPP 2
69 #define VTBLK_BLK_ID_BYTES 20 + 1
72 #define VTBLK_F_BARRIER (1 << 0) /* Does host support barriers? */
73 #define VTBLK_F_SIZE_MAX (1 << 1) /* Indicates maximum segment size */
74 #define VTBLK_F_SEG_MAX (1 << 2) /* Indicates maximum # of segments */
75 #define VTBLK_F_GEOMETRY (1 << 4) /* Legacy geometry available */
76 #define VTBLK_F_RO (1 << 5) /* Disk is read-only */
77 #define VTBLK_F_BLK_SIZE (1 << 6) /* Block size of disk is available*/
78 #define VTBLK_F_SCSI (1 << 7) /* Supports scsi command passthru */
79 #define VTBLK_F_FLUSH (1 << 9) /* Writeback mode enabled after reset */
80 #define VTBLK_F_WCE (1 << 9) /* Legacy alias for FLUSH */
81 #define VTBLK_F_TOPOLOGY (1 << 10) /* Topology information is available */
82 #define VTBLK_F_CONFIG_WCE (1 << 11) /* Writeback mode available in config */
83 #define VTBLK_F_MQ (1 << 12) /* Multi-Queue */
84 #define VTBLK_F_DISCARD (1 << 13) /* Trim blocks */
85 #define VTBLK_F_WRITE_ZEROES (1 << 14) /* Write zeros */
90 #define VTBLK_S_HOSTCAPS \
95 VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */
98 * The current blockif_delete() interface only allows a single delete
101 #define VTBLK_MAX_DISCARD_SEG 1
104 * An arbitrary limit to prevent excessive latency due to large
107 #define VTBLK_MAX_DISCARD_SECT ((16 << 20) / VTBLK_BSIZE) /* 16 MiB */
110 * Config space "registers"
112 struct vtblk_config {
113 uint64_t vbc_capacity;
114 uint32_t vbc_size_max;
115 uint32_t vbc_seg_max;
121 uint32_t vbc_blk_size;
123 uint8_t physical_block_exp;
124 uint8_t alignment_offset;
125 uint16_t min_io_size;
126 uint32_t opt_io_size;
128 uint8_t vbc_writeback;
131 uint32_t max_discard_sectors;
132 uint32_t max_discard_seg;
133 uint32_t discard_sector_alignment;
134 uint32_t max_write_zeroes_sectors;
135 uint32_t max_write_zeroes_seg;
136 uint8_t write_zeroes_may_unmap;
141 * Fixed-size block header
143 struct virtio_blk_hdr {
144 #define VBH_OP_READ 0
145 #define VBH_OP_WRITE 1
146 #define VBH_OP_SCSI_CMD 2
147 #define VBH_OP_SCSI_CMD_OUT 3
148 #define VBH_OP_FLUSH 4
149 #define VBH_OP_FLUSH_OUT 5
150 #define VBH_OP_IDENT 8
151 #define VBH_OP_DISCARD 11
152 #define VBH_OP_WRITE_ZEROES 13
154 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */
163 static int pci_vtblk_debug;
164 #define DPRINTF(params) if (pci_vtblk_debug) PRINTLN params
165 #define WPRINTF(params) PRINTLN params
167 struct pci_vtblk_ioreq {
168 struct blockif_req io_req;
169 struct pci_vtblk_softc *io_sc;
174 struct virtio_blk_discard_write_zeroes {
176 uint32_t num_sectors;
179 uint32_t reserved:31;
186 struct pci_vtblk_softc {
187 struct virtio_softc vbsc_vs;
188 pthread_mutex_t vsc_mtx;
189 struct vqueue_info vbsc_vq;
190 struct vtblk_config vbsc_cfg;
191 struct virtio_consts vbsc_consts;
192 struct blockif_ctxt *bc;
193 char vbsc_ident[VTBLK_BLK_ID_BYTES];
194 struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ];
197 static void pci_vtblk_reset(void *);
198 static void pci_vtblk_notify(void *, struct vqueue_info *);
199 static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
200 static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
202 static struct virtio_consts vtblk_vi_consts = {
203 "vtblk", /* our name */
204 1, /* we support 1 virtqueue */
205 sizeof(struct vtblk_config), /* config reg size */
206 pci_vtblk_reset, /* reset */
207 pci_vtblk_notify, /* device-wide qnotify */
208 pci_vtblk_cfgread, /* read PCI config */
209 pci_vtblk_cfgwrite, /* write PCI config */
210 NULL, /* apply negotiated features */
211 VTBLK_S_HOSTCAPS, /* our capabilities */
215 pci_vtblk_reset(void *vsc)
217 struct pci_vtblk_softc *sc = vsc;
219 DPRINTF(("vtblk: device reset requested !"));
220 vi_reset_dev(&sc->vbsc_vs);
224 pci_vtblk_done_locked(struct pci_vtblk_ioreq *io, int err)
226 struct pci_vtblk_softc *sc = io->io_sc;
228 /* convert errno into a virtio block error return */
229 if (err == EOPNOTSUPP || err == ENOSYS)
230 *io->io_status = VTBLK_S_UNSUPP;
232 *io->io_status = VTBLK_S_IOERR;
234 *io->io_status = VTBLK_S_OK;
237 * Return the descriptor back to the host.
238 * We wrote 1 byte (our status) to host.
240 vq_relchain(&sc->vbsc_vq, io->io_idx, 1);
241 vq_endchains(&sc->vbsc_vq, 0);
245 pci_vtblk_done(struct blockif_req *br, int err)
247 struct pci_vtblk_ioreq *io = br->br_param;
248 struct pci_vtblk_softc *sc = io->io_sc;
250 pthread_mutex_lock(&sc->vsc_mtx);
251 pci_vtblk_done_locked(io, err);
252 pthread_mutex_unlock(&sc->vsc_mtx);
256 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
258 struct virtio_blk_hdr *vbh;
259 struct pci_vtblk_ioreq *io;
264 struct iovec iov[BLOCKIF_IOV_MAX + 2];
265 uint16_t idx, flags[BLOCKIF_IOV_MAX + 2];
266 struct virtio_blk_discard_write_zeroes *discard;
268 n = vq_getchain(vq, &idx, iov, BLOCKIF_IOV_MAX + 2, flags);
271 * The first descriptor will be the read-only fixed header,
272 * and the last is for status (hence +2 above and below).
273 * The remaining iov's are the actual data I/O vectors.
275 * XXX - note - this fails on crash dump, which does a
276 * VIRTIO_BLK_T_FLUSH with a zero transfer length
278 assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2);
280 io = &sc->vbsc_ios[idx];
281 assert((flags[0] & VRING_DESC_F_WRITE) == 0);
282 assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
283 vbh = (struct virtio_blk_hdr *)iov[0].iov_base;
284 memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2));
285 io->io_req.br_iovcnt = n - 2;
286 io->io_req.br_offset = vbh->vbh_sector * VTBLK_BSIZE;
287 io->io_status = (uint8_t *)iov[--n].iov_base;
288 assert(iov[n].iov_len == 1);
289 assert(flags[n] & VRING_DESC_F_WRITE);
293 * The guest should not be setting the BARRIER flag because
294 * we don't advertise the capability.
296 type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
297 writeop = (type == VBH_OP_WRITE || type == VBH_OP_DISCARD);
300 for (i = 1; i < n; i++) {
302 * - write op implies read-only descriptor,
303 * - read/ident op implies write-only descriptor,
304 * therefore test the inverse of the descriptor bit
307 assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
308 iolen += iov[i].iov_len;
310 io->io_req.br_resid = iolen;
312 DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld",
313 writeop ? "write/discard" : "read/ident", iolen, i - 1,
314 io->io_req.br_offset));
318 err = blockif_read(sc->bc, &io->io_req);
321 err = blockif_write(sc->bc, &io->io_req);
325 * We currently only support a single request, if the guest
326 * has submitted a request that doesn't conform to the
327 * requirements, we return a error.
329 if (iov[1].iov_len != sizeof (*discard)) {
330 pci_vtblk_done_locked(io, EINVAL);
334 /* The segments to discard are provided rather than data */
335 discard = (struct virtio_blk_discard_write_zeroes *)
339 * virtio v1.1 5.2.6.2:
340 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP
341 * for discard and write zeroes commands if any unknown flag is
342 * set. Furthermore, the device MUST set the status byte to
343 * VIRTIO_BLK_S_UNSUPP for discard commands if the unmap flag
346 * Currently there are no known flags for a DISCARD request.
348 if (discard->flags.unmap != 0 || discard->flags.reserved != 0) {
349 pci_vtblk_done_locked(io, ENOTSUP);
353 /* Make sure the request doesn't exceed our size limit */
354 if (discard->num_sectors > VTBLK_MAX_DISCARD_SECT) {
355 pci_vtblk_done_locked(io, EINVAL);
359 io->io_req.br_offset = discard->sector * VTBLK_BSIZE;
360 io->io_req.br_resid = discard->num_sectors * VTBLK_BSIZE;
361 err = blockif_delete(sc->bc, &io->io_req);
364 case VBH_OP_FLUSH_OUT:
365 err = blockif_flush(sc->bc, &io->io_req);
368 /* Assume a single buffer */
369 /* S/n equal to buffer is not zero-terminated. */
370 memset(iov[1].iov_base, 0, iov[1].iov_len);
371 strncpy(iov[1].iov_base, sc->vbsc_ident,
372 MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
373 pci_vtblk_done_locked(io, 0);
376 pci_vtblk_done_locked(io, EOPNOTSUPP);
383 pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
385 struct pci_vtblk_softc *sc = vsc;
387 while (vq_has_descs(vq))
388 pci_vtblk_proc(sc, vq);
392 pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
394 char bident[sizeof("XX:X:X")];
395 struct blockif_ctxt *bctxt;
398 struct pci_vtblk_softc *sc;
400 int i, sectsz, sts, sto;
403 WPRINTF(("virtio-block: backing device required"));
408 * The supplied backing file has to exist
410 snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func);
411 bctxt = blockif_open(opts, bident);
413 perror("Could not open backing file");
417 size = blockif_size(bctxt);
418 sectsz = blockif_sectsz(bctxt);
419 blockif_psectsz(bctxt, &sts, &sto);
421 sc = calloc(1, sizeof(struct pci_vtblk_softc));
423 for (i = 0; i < VTBLK_RINGSZ; i++) {
424 struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i];
425 io->io_req.br_callback = pci_vtblk_done;
426 io->io_req.br_param = io;
431 bcopy(&vtblk_vi_consts, &sc->vbsc_consts, sizeof (vtblk_vi_consts));
432 if (blockif_candelete(sc->bc))
433 sc->vbsc_consts.vc_hv_caps |= VTBLK_F_DISCARD;
435 pthread_mutex_init(&sc->vsc_mtx, NULL);
437 /* init virtio softc and virtqueues */
438 vi_softc_linkup(&sc->vbsc_vs, &sc->vbsc_consts, sc, pi, &sc->vbsc_vq);
439 sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
441 sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
442 /* sc->vbsc_vq.vq_notify = we have no per-queue notify */
445 * Create an identifier for the backing file. Use parts of the
446 * md5 sum of the filename
449 MD5Update(&mdctx, opts, strlen(opts));
450 MD5Final(digest, &mdctx);
451 snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES,
452 "BHYVE-%02X%02X-%02X%02X-%02X%02X",
453 digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
455 /* setup virtio block config space */
456 sc->vbsc_cfg.vbc_capacity = size / VTBLK_BSIZE; /* 512-byte units */
457 sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */
460 * If Linux is presented with a seg_max greater than the virtio queue
461 * size, it can stumble into situations where it violates its own
462 * invariants and panics. For safety, we keep seg_max clamped, paying
463 * heed to the two extra descriptors needed for the header and status
466 sc->vbsc_cfg.vbc_seg_max = MIN(VTBLK_RINGSZ - 2, BLOCKIF_IOV_MAX);
467 sc->vbsc_cfg.vbc_geometry.cylinders = 0; /* no geometry */
468 sc->vbsc_cfg.vbc_geometry.heads = 0;
469 sc->vbsc_cfg.vbc_geometry.sectors = 0;
470 sc->vbsc_cfg.vbc_blk_size = sectsz;
471 sc->vbsc_cfg.vbc_topology.physical_block_exp =
472 (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
473 sc->vbsc_cfg.vbc_topology.alignment_offset =
474 (sto != 0) ? ((sts - sto) / sectsz) : 0;
475 sc->vbsc_cfg.vbc_topology.min_io_size = 0;
476 sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
477 sc->vbsc_cfg.vbc_writeback = 0;
478 sc->vbsc_cfg.max_discard_sectors = VTBLK_MAX_DISCARD_SECT;
479 sc->vbsc_cfg.max_discard_seg = VTBLK_MAX_DISCARD_SEG;
480 sc->vbsc_cfg.discard_sector_alignment = sectsz / VTBLK_BSIZE;
483 * Should we move some of this into virtio.c? Could
484 * have the device, class, and subdev_0 as fields in
485 * the virtio constants structure.
487 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
488 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
489 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
490 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
491 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
493 if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) {
494 blockif_close(sc->bc);
498 vi_set_io_bar(&sc->vbsc_vs, 0);
503 pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
506 DPRINTF(("vtblk: write to readonly reg %d", offset));
511 pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
513 struct pci_vtblk_softc *sc = vsc;
516 /* our caller has already verified offset and size */
517 ptr = (uint8_t *)&sc->vbsc_cfg + offset;
518 memcpy(retval, ptr, size);
522 struct pci_devemu pci_de_vblk = {
523 .pe_emu = "virtio-blk",
524 .pe_init = pci_vtblk_init,
525 .pe_barwrite = vi_pci_write,
526 .pe_barread = vi_pci_read
528 PCI_EMUL_SET(pci_de_vblk);