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>
42 #include <machine/vmm_snapshot.h>
62 #define VTBLK_BSIZE 512
63 #define VTBLK_RINGSZ 128
65 _Static_assert(VTBLK_RINGSZ <= BLOCKIF_RING_MAX, "Each ring entry must be able to queue a request");
68 #define VTBLK_S_IOERR 1
69 #define VTBLK_S_UNSUPP 2
71 #define VTBLK_BLK_ID_BYTES 20 + 1
74 #define VTBLK_F_BARRIER (1 << 0) /* Does host support barriers? */
75 #define VTBLK_F_SIZE_MAX (1 << 1) /* Indicates maximum segment size */
76 #define VTBLK_F_SEG_MAX (1 << 2) /* Indicates maximum # of segments */
77 #define VTBLK_F_GEOMETRY (1 << 4) /* Legacy geometry available */
78 #define VTBLK_F_RO (1 << 5) /* Disk is read-only */
79 #define VTBLK_F_BLK_SIZE (1 << 6) /* Block size of disk is available*/
80 #define VTBLK_F_SCSI (1 << 7) /* Supports scsi command passthru */
81 #define VTBLK_F_FLUSH (1 << 9) /* Writeback mode enabled after reset */
82 #define VTBLK_F_WCE (1 << 9) /* Legacy alias for FLUSH */
83 #define VTBLK_F_TOPOLOGY (1 << 10) /* Topology information is available */
84 #define VTBLK_F_CONFIG_WCE (1 << 11) /* Writeback mode available in config */
85 #define VTBLK_F_MQ (1 << 12) /* Multi-Queue */
86 #define VTBLK_F_DISCARD (1 << 13) /* Trim blocks */
87 #define VTBLK_F_WRITE_ZEROES (1 << 14) /* Write zeros */
92 #define VTBLK_S_HOSTCAPS \
97 VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */
100 * The current blockif_delete() interface only allows a single delete
103 #define VTBLK_MAX_DISCARD_SEG 1
106 * An arbitrary limit to prevent excessive latency due to large
109 #define VTBLK_MAX_DISCARD_SECT ((16 << 20) / VTBLK_BSIZE) /* 16 MiB */
112 * Config space "registers"
114 struct vtblk_config {
115 uint64_t vbc_capacity;
116 uint32_t vbc_size_max;
117 uint32_t vbc_seg_max;
123 uint32_t vbc_blk_size;
125 uint8_t physical_block_exp;
126 uint8_t alignment_offset;
127 uint16_t min_io_size;
128 uint32_t opt_io_size;
130 uint8_t vbc_writeback;
133 uint32_t max_discard_sectors;
134 uint32_t max_discard_seg;
135 uint32_t discard_sector_alignment;
136 uint32_t max_write_zeroes_sectors;
137 uint32_t max_write_zeroes_seg;
138 uint8_t write_zeroes_may_unmap;
143 * Fixed-size block header
145 struct virtio_blk_hdr {
146 #define VBH_OP_READ 0
147 #define VBH_OP_WRITE 1
148 #define VBH_OP_SCSI_CMD 2
149 #define VBH_OP_SCSI_CMD_OUT 3
150 #define VBH_OP_FLUSH 4
151 #define VBH_OP_FLUSH_OUT 5
152 #define VBH_OP_IDENT 8
153 #define VBH_OP_DISCARD 11
154 #define VBH_OP_WRITE_ZEROES 13
156 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */
165 static int pci_vtblk_debug;
166 #define DPRINTF(params) if (pci_vtblk_debug) PRINTLN params
167 #define WPRINTF(params) PRINTLN params
169 struct pci_vtblk_ioreq {
170 struct blockif_req io_req;
171 struct pci_vtblk_softc *io_sc;
176 struct virtio_blk_discard_write_zeroes {
178 uint32_t num_sectors;
181 uint32_t reserved:31;
188 struct pci_vtblk_softc {
189 struct virtio_softc vbsc_vs;
190 pthread_mutex_t vsc_mtx;
191 struct vqueue_info vbsc_vq;
192 struct vtblk_config vbsc_cfg;
193 struct virtio_consts vbsc_consts;
194 struct blockif_ctxt *bc;
195 char vbsc_ident[VTBLK_BLK_ID_BYTES];
196 struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ];
199 static void pci_vtblk_reset(void *);
200 static void pci_vtblk_notify(void *, struct vqueue_info *);
201 static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
202 static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
203 #ifdef BHYVE_SNAPSHOT
204 static void pci_vtblk_pause(void *);
205 static void pci_vtblk_resume(void *);
206 static int pci_vtblk_snapshot(void *, struct vm_snapshot_meta *);
209 static struct virtio_consts vtblk_vi_consts = {
210 "vtblk", /* our name */
211 1, /* we support 1 virtqueue */
212 sizeof(struct vtblk_config), /* config reg size */
213 pci_vtblk_reset, /* reset */
214 pci_vtblk_notify, /* device-wide qnotify */
215 pci_vtblk_cfgread, /* read PCI config */
216 pci_vtblk_cfgwrite, /* write PCI config */
217 NULL, /* apply negotiated features */
218 VTBLK_S_HOSTCAPS, /* our capabilities */
219 #ifdef BHYVE_SNAPSHOT
220 pci_vtblk_pause, /* pause blockif threads */
221 pci_vtblk_resume, /* resume blockif threads */
222 pci_vtblk_snapshot, /* save / restore device state */
227 pci_vtblk_reset(void *vsc)
229 struct pci_vtblk_softc *sc = vsc;
231 DPRINTF(("vtblk: device reset requested !"));
232 vi_reset_dev(&sc->vbsc_vs);
236 pci_vtblk_done_locked(struct pci_vtblk_ioreq *io, int err)
238 struct pci_vtblk_softc *sc = io->io_sc;
240 /* convert errno into a virtio block error return */
241 if (err == EOPNOTSUPP || err == ENOSYS)
242 *io->io_status = VTBLK_S_UNSUPP;
244 *io->io_status = VTBLK_S_IOERR;
246 *io->io_status = VTBLK_S_OK;
249 * Return the descriptor back to the host.
250 * We wrote 1 byte (our status) to host.
252 vq_relchain(&sc->vbsc_vq, io->io_idx, 1);
253 vq_endchains(&sc->vbsc_vq, 0);
256 #ifdef BHYVE_SNAPSHOT
258 pci_vtblk_pause(void *vsc)
260 struct pci_vtblk_softc *sc = vsc;
262 DPRINTF(("vtblk: device pause requested !\n"));
263 blockif_pause(sc->bc);
267 pci_vtblk_resume(void *vsc)
269 struct pci_vtblk_softc *sc = vsc;
271 DPRINTF(("vtblk: device resume requested !\n"));
272 blockif_resume(sc->bc);
276 pci_vtblk_snapshot(void *vsc, struct vm_snapshot_meta *meta)
279 struct pci_vtblk_softc *sc = vsc;
281 SNAPSHOT_VAR_OR_LEAVE(sc->vbsc_cfg, meta, ret, done);
282 SNAPSHOT_BUF_OR_LEAVE(sc->vbsc_ident, sizeof(sc->vbsc_ident),
291 pci_vtblk_done(struct blockif_req *br, int err)
293 struct pci_vtblk_ioreq *io = br->br_param;
294 struct pci_vtblk_softc *sc = io->io_sc;
296 pthread_mutex_lock(&sc->vsc_mtx);
297 pci_vtblk_done_locked(io, err);
298 pthread_mutex_unlock(&sc->vsc_mtx);
302 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
304 struct virtio_blk_hdr *vbh;
305 struct pci_vtblk_ioreq *io;
310 struct iovec iov[BLOCKIF_IOV_MAX + 2];
311 uint16_t idx, flags[BLOCKIF_IOV_MAX + 2];
312 struct virtio_blk_discard_write_zeroes *discard;
314 n = vq_getchain(vq, &idx, iov, BLOCKIF_IOV_MAX + 2, flags);
317 * The first descriptor will be the read-only fixed header,
318 * and the last is for status (hence +2 above and below).
319 * The remaining iov's are the actual data I/O vectors.
321 * XXX - note - this fails on crash dump, which does a
322 * VIRTIO_BLK_T_FLUSH with a zero transfer length
324 assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2);
326 io = &sc->vbsc_ios[idx];
327 assert((flags[0] & VRING_DESC_F_WRITE) == 0);
328 assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
329 vbh = (struct virtio_blk_hdr *)iov[0].iov_base;
330 memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2));
331 io->io_req.br_iovcnt = n - 2;
332 io->io_req.br_offset = vbh->vbh_sector * VTBLK_BSIZE;
333 io->io_status = (uint8_t *)iov[--n].iov_base;
334 assert(iov[n].iov_len == 1);
335 assert(flags[n] & VRING_DESC_F_WRITE);
339 * The guest should not be setting the BARRIER flag because
340 * we don't advertise the capability.
342 type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
343 writeop = (type == VBH_OP_WRITE || type == VBH_OP_DISCARD);
346 for (i = 1; i < n; i++) {
348 * - write op implies read-only descriptor,
349 * - read/ident op implies write-only descriptor,
350 * therefore test the inverse of the descriptor bit
353 assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop);
354 iolen += iov[i].iov_len;
356 io->io_req.br_resid = iolen;
358 DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld",
359 writeop ? "write/discard" : "read/ident", iolen, i - 1,
360 io->io_req.br_offset));
364 err = blockif_read(sc->bc, &io->io_req);
367 err = blockif_write(sc->bc, &io->io_req);
371 * We currently only support a single request, if the guest
372 * has submitted a request that doesn't conform to the
373 * requirements, we return a error.
375 if (iov[1].iov_len != sizeof (*discard)) {
376 pci_vtblk_done_locked(io, EINVAL);
380 /* The segments to discard are provided rather than data */
381 discard = (struct virtio_blk_discard_write_zeroes *)
385 * virtio v1.1 5.2.6.2:
386 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP
387 * for discard and write zeroes commands if any unknown flag is
388 * set. Furthermore, the device MUST set the status byte to
389 * VIRTIO_BLK_S_UNSUPP for discard commands if the unmap flag
392 * Currently there are no known flags for a DISCARD request.
394 if (discard->flags.unmap != 0 || discard->flags.reserved != 0) {
395 pci_vtblk_done_locked(io, ENOTSUP);
399 /* Make sure the request doesn't exceed our size limit */
400 if (discard->num_sectors > VTBLK_MAX_DISCARD_SECT) {
401 pci_vtblk_done_locked(io, EINVAL);
405 io->io_req.br_offset = discard->sector * VTBLK_BSIZE;
406 io->io_req.br_resid = discard->num_sectors * VTBLK_BSIZE;
407 err = blockif_delete(sc->bc, &io->io_req);
410 case VBH_OP_FLUSH_OUT:
411 err = blockif_flush(sc->bc, &io->io_req);
414 /* Assume a single buffer */
415 /* S/n equal to buffer is not zero-terminated. */
416 memset(iov[1].iov_base, 0, iov[1].iov_len);
417 strncpy(iov[1].iov_base, sc->vbsc_ident,
418 MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
419 pci_vtblk_done_locked(io, 0);
422 pci_vtblk_done_locked(io, EOPNOTSUPP);
429 pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
431 struct pci_vtblk_softc *sc = vsc;
433 while (vq_has_descs(vq))
434 pci_vtblk_proc(sc, vq);
438 pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
440 char bident[sizeof("XX:X:X")];
441 struct blockif_ctxt *bctxt;
444 struct pci_vtblk_softc *sc;
446 int i, sectsz, sts, sto;
449 WPRINTF(("virtio-block: backing device required"));
454 * The supplied backing file has to exist
456 snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func);
457 bctxt = blockif_open(opts, bident);
459 perror("Could not open backing file");
463 size = blockif_size(bctxt);
464 sectsz = blockif_sectsz(bctxt);
465 blockif_psectsz(bctxt, &sts, &sto);
467 sc = calloc(1, sizeof(struct pci_vtblk_softc));
469 for (i = 0; i < VTBLK_RINGSZ; i++) {
470 struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i];
471 io->io_req.br_callback = pci_vtblk_done;
472 io->io_req.br_param = io;
477 bcopy(&vtblk_vi_consts, &sc->vbsc_consts, sizeof (vtblk_vi_consts));
478 if (blockif_candelete(sc->bc))
479 sc->vbsc_consts.vc_hv_caps |= VTBLK_F_DISCARD;
481 pthread_mutex_init(&sc->vsc_mtx, NULL);
483 /* init virtio softc and virtqueues */
484 vi_softc_linkup(&sc->vbsc_vs, &sc->vbsc_consts, sc, pi, &sc->vbsc_vq);
485 sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
487 sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
488 /* sc->vbsc_vq.vq_notify = we have no per-queue notify */
491 * Create an identifier for the backing file. Use parts of the
492 * md5 sum of the filename
495 MD5Update(&mdctx, opts, strlen(opts));
496 MD5Final(digest, &mdctx);
497 snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES,
498 "BHYVE-%02X%02X-%02X%02X-%02X%02X",
499 digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]);
501 /* setup virtio block config space */
502 sc->vbsc_cfg.vbc_capacity = size / VTBLK_BSIZE; /* 512-byte units */
503 sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */
506 * If Linux is presented with a seg_max greater than the virtio queue
507 * size, it can stumble into situations where it violates its own
508 * invariants and panics. For safety, we keep seg_max clamped, paying
509 * heed to the two extra descriptors needed for the header and status
512 sc->vbsc_cfg.vbc_seg_max = MIN(VTBLK_RINGSZ - 2, BLOCKIF_IOV_MAX);
513 sc->vbsc_cfg.vbc_geometry.cylinders = 0; /* no geometry */
514 sc->vbsc_cfg.vbc_geometry.heads = 0;
515 sc->vbsc_cfg.vbc_geometry.sectors = 0;
516 sc->vbsc_cfg.vbc_blk_size = sectsz;
517 sc->vbsc_cfg.vbc_topology.physical_block_exp =
518 (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
519 sc->vbsc_cfg.vbc_topology.alignment_offset =
520 (sto != 0) ? ((sts - sto) / sectsz) : 0;
521 sc->vbsc_cfg.vbc_topology.min_io_size = 0;
522 sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
523 sc->vbsc_cfg.vbc_writeback = 0;
524 sc->vbsc_cfg.max_discard_sectors = VTBLK_MAX_DISCARD_SECT;
525 sc->vbsc_cfg.max_discard_seg = VTBLK_MAX_DISCARD_SEG;
526 sc->vbsc_cfg.discard_sector_alignment = MAX(sectsz, sts) / VTBLK_BSIZE;
529 * Should we move some of this into virtio.c? Could
530 * have the device, class, and subdev_0 as fields in
531 * the virtio constants structure.
533 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
534 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
535 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
536 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
537 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
539 if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) {
540 blockif_close(sc->bc);
544 vi_set_io_bar(&sc->vbsc_vs, 0);
549 pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value)
552 DPRINTF(("vtblk: write to readonly reg %d", offset));
557 pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
559 struct pci_vtblk_softc *sc = vsc;
562 /* our caller has already verified offset and size */
563 ptr = (uint8_t *)&sc->vbsc_cfg + offset;
564 memcpy(retval, ptr, size);
568 struct pci_devemu pci_de_vblk = {
569 .pe_emu = "virtio-blk",
570 .pe_init = pci_vtblk_init,
571 .pe_barwrite = vi_pci_write,
572 .pe_barread = vi_pci_read,
573 #ifdef BHYVE_SNAPSHOT
574 .pe_snapshot = vi_pci_snapshot,
577 PCI_EMUL_SET(pci_de_vblk);