2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2011 NetApp, Inc.
6 * Copyright 2020-2021 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
30 #include <sys/param.h>
31 #include <sys/linker_set.h>
34 #include <sys/ioctl.h>
37 #include <machine/vmm_snapshot.h>
58 #define VTBLK_BSIZE 512
59 #define VTBLK_RINGSZ 128
61 _Static_assert(VTBLK_RINGSZ <= BLOCKIF_RING_MAX, "Each ring entry must be able to queue a request");
64 #define VTBLK_S_IOERR 1
65 #define VTBLK_S_UNSUPP 2
67 #define VTBLK_BLK_ID_BYTES 20 + 1
70 #define VTBLK_F_BARRIER (1 << 0) /* Does host support barriers? */
71 #define VTBLK_F_SIZE_MAX (1 << 1) /* Indicates maximum segment size */
72 #define VTBLK_F_SEG_MAX (1 << 2) /* Indicates maximum # of segments */
73 #define VTBLK_F_GEOMETRY (1 << 4) /* Legacy geometry available */
74 #define VTBLK_F_RO (1 << 5) /* Disk is read-only */
75 #define VTBLK_F_BLK_SIZE (1 << 6) /* Block size of disk is available*/
76 #define VTBLK_F_SCSI (1 << 7) /* Supports scsi command passthru */
77 #define VTBLK_F_FLUSH (1 << 9) /* Writeback mode enabled after reset */
78 #define VTBLK_F_WCE (1 << 9) /* Legacy alias for FLUSH */
79 #define VTBLK_F_TOPOLOGY (1 << 10) /* Topology information is available */
80 #define VTBLK_F_CONFIG_WCE (1 << 11) /* Writeback mode available in config */
81 #define VTBLK_F_MQ (1 << 12) /* Multi-Queue */
82 #define VTBLK_F_DISCARD (1 << 13) /* Trim blocks */
83 #define VTBLK_F_WRITE_ZEROES (1 << 14) /* Write zeros */
88 #define VTBLK_S_HOSTCAPS \
93 VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */
96 * The current blockif_delete() interface only allows a single delete
99 #define VTBLK_MAX_DISCARD_SEG 1
102 * An arbitrary limit to prevent excessive latency due to large
105 #define VTBLK_MAX_DISCARD_SECT ((16 << 20) / VTBLK_BSIZE) /* 16 MiB */
108 * Config space "registers"
110 struct vtblk_config {
111 uint64_t vbc_capacity;
112 uint32_t vbc_size_max;
113 uint32_t vbc_seg_max;
119 uint32_t vbc_blk_size;
121 uint8_t physical_block_exp;
122 uint8_t alignment_offset;
123 uint16_t min_io_size;
124 uint32_t opt_io_size;
126 uint8_t vbc_writeback;
129 uint32_t max_discard_sectors;
130 uint32_t max_discard_seg;
131 uint32_t discard_sector_alignment;
132 uint32_t max_write_zeroes_sectors;
133 uint32_t max_write_zeroes_seg;
134 uint8_t write_zeroes_may_unmap;
139 * Fixed-size block header
141 struct virtio_blk_hdr {
142 #define VBH_OP_READ 0
143 #define VBH_OP_WRITE 1
144 #define VBH_OP_SCSI_CMD 2
145 #define VBH_OP_SCSI_CMD_OUT 3
146 #define VBH_OP_FLUSH 4
147 #define VBH_OP_FLUSH_OUT 5
148 #define VBH_OP_IDENT 8
149 #define VBH_OP_DISCARD 11
150 #define VBH_OP_WRITE_ZEROES 13
152 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */
161 static int pci_vtblk_debug;
162 #define DPRINTF(params) if (pci_vtblk_debug) PRINTLN params
163 #define WPRINTF(params) PRINTLN params
165 struct pci_vtblk_ioreq {
166 struct blockif_req io_req;
167 struct pci_vtblk_softc *io_sc;
172 struct virtio_blk_discard_write_zeroes {
174 uint32_t num_sectors;
177 uint32_t reserved:31;
184 struct pci_vtblk_softc {
185 struct virtio_softc vbsc_vs;
186 pthread_mutex_t vsc_mtx;
187 struct vqueue_info vbsc_vq;
188 struct vtblk_config vbsc_cfg;
189 struct virtio_consts vbsc_consts;
190 struct blockif_ctxt *bc;
191 char vbsc_ident[VTBLK_BLK_ID_BYTES];
192 struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ];
195 static void pci_vtblk_reset(void *);
196 static void pci_vtblk_notify(void *, struct vqueue_info *);
197 static int pci_vtblk_cfgread(void *, int, int, uint32_t *);
198 static int pci_vtblk_cfgwrite(void *, int, int, uint32_t);
199 #ifdef BHYVE_SNAPSHOT
200 static void pci_vtblk_pause(void *);
201 static void pci_vtblk_resume(void *);
202 static int pci_vtblk_snapshot(void *, struct vm_snapshot_meta *);
205 static struct virtio_consts vtblk_vi_consts = {
208 .vc_cfgsize = sizeof(struct vtblk_config),
209 .vc_reset = pci_vtblk_reset,
210 .vc_qnotify = pci_vtblk_notify,
211 .vc_cfgread = pci_vtblk_cfgread,
212 .vc_cfgwrite = pci_vtblk_cfgwrite,
213 .vc_apply_features = NULL,
214 .vc_hv_caps = VTBLK_S_HOSTCAPS,
215 #ifdef BHYVE_SNAPSHOT
216 .vc_pause = pci_vtblk_pause,
217 .vc_resume = pci_vtblk_resume,
218 .vc_snapshot = pci_vtblk_snapshot,
223 pci_vtblk_reset(void *vsc)
225 struct pci_vtblk_softc *sc = vsc;
227 DPRINTF(("vtblk: device reset requested !"));
228 vi_reset_dev(&sc->vbsc_vs);
232 pci_vtblk_done_locked(struct pci_vtblk_ioreq *io, int err)
234 struct pci_vtblk_softc *sc = io->io_sc;
236 /* convert errno into a virtio block error return */
237 if (err == EOPNOTSUPP || err == ENOSYS)
238 *io->io_status = VTBLK_S_UNSUPP;
240 *io->io_status = VTBLK_S_IOERR;
242 *io->io_status = VTBLK_S_OK;
245 * Return the descriptor back to the host.
246 * We wrote 1 byte (our status) to host.
248 vq_relchain(&sc->vbsc_vq, io->io_idx, 1);
249 vq_endchains(&sc->vbsc_vq, 0);
252 #ifdef BHYVE_SNAPSHOT
254 pci_vtblk_pause(void *vsc)
256 struct pci_vtblk_softc *sc = vsc;
258 DPRINTF(("vtblk: device pause requested !\n"));
259 blockif_pause(sc->bc);
263 pci_vtblk_resume(void *vsc)
265 struct pci_vtblk_softc *sc = vsc;
267 DPRINTF(("vtblk: device resume requested !\n"));
268 blockif_resume(sc->bc);
272 pci_vtblk_snapshot(void *vsc, struct vm_snapshot_meta *meta)
275 struct pci_vtblk_softc *sc = vsc;
277 SNAPSHOT_VAR_OR_LEAVE(sc->vbsc_cfg, meta, ret, done);
278 SNAPSHOT_BUF_OR_LEAVE(sc->vbsc_ident, sizeof(sc->vbsc_ident),
287 pci_vtblk_done(struct blockif_req *br, int err)
289 struct pci_vtblk_ioreq *io = br->br_param;
290 struct pci_vtblk_softc *sc = io->io_sc;
292 pthread_mutex_lock(&sc->vsc_mtx);
293 pci_vtblk_done_locked(io, err);
294 pthread_mutex_unlock(&sc->vsc_mtx);
298 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq)
300 struct virtio_blk_hdr *vbh;
301 struct pci_vtblk_ioreq *io;
307 struct iovec iov[BLOCKIF_IOV_MAX + 2];
308 struct virtio_blk_discard_write_zeroes *discard;
310 n = vq_getchain(vq, iov, BLOCKIF_IOV_MAX + 2, &req);
313 * The first descriptor will be the read-only fixed header,
314 * and the last is for status (hence +2 above and below).
315 * The remaining iov's are the actual data I/O vectors.
317 * XXX - note - this fails on crash dump, which does a
318 * VIRTIO_BLK_T_FLUSH with a zero transfer length
320 assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2);
322 io = &sc->vbsc_ios[req.idx];
323 assert(req.readable != 0);
324 assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr));
325 vbh = (struct virtio_blk_hdr *)iov[0].iov_base;
326 memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2));
327 io->io_req.br_iovcnt = n - 2;
328 io->io_req.br_offset = vbh->vbh_sector * VTBLK_BSIZE;
329 io->io_status = (uint8_t *)iov[--n].iov_base;
330 assert(req.writable != 0);
331 assert(iov[n].iov_len == 1);
335 * The guest should not be setting the BARRIER flag because
336 * we don't advertise the capability.
338 type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
339 writeop = (type == VBH_OP_WRITE || type == VBH_OP_DISCARD);
341 * - Write op implies read-only descriptor
342 * - Read/ident op implies write-only descriptor
344 * By taking away either the read-only fixed header or the write-only
345 * status iovec, the following condition should hold true.
347 assert(n == (writeop ? req.readable : req.writable));
350 for (i = 1; i < n; i++) {
351 iolen += iov[i].iov_len;
353 io->io_req.br_resid = iolen;
355 DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld",
356 writeop ? "write/discard" : "read/ident", iolen, i - 1,
357 io->io_req.br_offset));
361 err = blockif_read(sc->bc, &io->io_req);
364 err = blockif_write(sc->bc, &io->io_req);
368 * We currently only support a single request, if the guest
369 * has submitted a request that doesn't conform to the
370 * requirements, we return a error.
372 if (iov[1].iov_len != sizeof (*discard)) {
373 pci_vtblk_done_locked(io, EINVAL);
377 /* The segments to discard are provided rather than data */
378 discard = (struct virtio_blk_discard_write_zeroes *)
382 * virtio v1.1 5.2.6.2:
383 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP
384 * for discard and write zeroes commands if any unknown flag is
385 * set. Furthermore, the device MUST set the status byte to
386 * VIRTIO_BLK_S_UNSUPP for discard commands if the unmap flag
389 * Currently there are no known flags for a DISCARD request.
391 if (discard->flags.unmap != 0 || discard->flags.reserved != 0) {
392 pci_vtblk_done_locked(io, ENOTSUP);
396 /* Make sure the request doesn't exceed our size limit */
397 if (discard->num_sectors > VTBLK_MAX_DISCARD_SECT) {
398 pci_vtblk_done_locked(io, EINVAL);
402 io->io_req.br_offset = discard->sector * VTBLK_BSIZE;
403 io->io_req.br_resid = discard->num_sectors * VTBLK_BSIZE;
404 err = blockif_delete(sc->bc, &io->io_req);
407 case VBH_OP_FLUSH_OUT:
408 err = blockif_flush(sc->bc, &io->io_req);
411 /* Assume a single buffer */
412 /* S/n equal to buffer is not zero-terminated. */
413 memset(iov[1].iov_base, 0, iov[1].iov_len);
414 strncpy(iov[1].iov_base, sc->vbsc_ident,
415 MIN(iov[1].iov_len, sizeof(sc->vbsc_ident)));
416 pci_vtblk_done_locked(io, 0);
419 pci_vtblk_done_locked(io, EOPNOTSUPP);
426 pci_vtblk_notify(void *vsc, struct vqueue_info *vq)
428 struct pci_vtblk_softc *sc = vsc;
430 while (vq_has_descs(vq))
431 pci_vtblk_proc(sc, vq);
435 pci_vtblk_resized(struct blockif_ctxt *bctxt __unused, void *arg,
438 struct pci_vtblk_softc *sc;
442 sc->vbsc_cfg.vbc_capacity = new_size / VTBLK_BSIZE; /* 512-byte units */
443 vi_interrupt(&sc->vbsc_vs, VIRTIO_PCI_ISR_CONFIG,
444 sc->vbsc_vs.vs_msix_cfg_idx);
448 pci_vtblk_init(struct pci_devinst *pi, nvlist_t *nvl)
450 char bident[sizeof("XXX:XXX")];
451 struct blockif_ctxt *bctxt;
452 const char *path, *serial;
455 struct pci_vtblk_softc *sc;
457 int i, sectsz, sts, sto;
460 * The supplied backing file has to exist
462 snprintf(bident, sizeof(bident), "%u:%u", pi->pi_slot, pi->pi_func);
463 bctxt = blockif_open(nvl, bident);
465 perror("Could not open backing file");
469 if (blockif_add_boot_device(pi, bctxt)) {
470 perror("Invalid boot device");
474 size = blockif_size(bctxt);
475 sectsz = blockif_sectsz(bctxt);
476 blockif_psectsz(bctxt, &sts, &sto);
478 sc = calloc(1, sizeof(struct pci_vtblk_softc));
480 for (i = 0; i < VTBLK_RINGSZ; i++) {
481 struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i];
482 io->io_req.br_callback = pci_vtblk_done;
483 io->io_req.br_param = io;
488 bcopy(&vtblk_vi_consts, &sc->vbsc_consts, sizeof (vtblk_vi_consts));
489 if (blockif_candelete(sc->bc))
490 sc->vbsc_consts.vc_hv_caps |= VTBLK_F_DISCARD;
492 pthread_mutex_init(&sc->vsc_mtx, NULL);
494 /* init virtio softc and virtqueues */
495 vi_softc_linkup(&sc->vbsc_vs, &sc->vbsc_consts, sc, pi, &sc->vbsc_vq);
496 sc->vbsc_vs.vs_mtx = &sc->vsc_mtx;
498 sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ;
499 /* sc->vbsc_vq.vq_notify = we have no per-queue notify */
502 * If an explicit identifier is not given, create an
503 * identifier using parts of the md5 sum of the filename.
505 bzero(sc->vbsc_ident, VTBLK_BLK_ID_BYTES);
506 if ((serial = get_config_value_node(nvl, "serial")) != NULL ||
507 (serial = get_config_value_node(nvl, "ser")) != NULL) {
508 strlcpy(sc->vbsc_ident, serial, VTBLK_BLK_ID_BYTES);
510 path = get_config_value_node(nvl, "path");
512 MD5Update(&mdctx, path, strlen(path));
513 MD5Final(digest, &mdctx);
514 snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES,
515 "BHYVE-%02X%02X-%02X%02X-%02X%02X",
516 digest[0], digest[1], digest[2], digest[3], digest[4],
520 /* setup virtio block config space */
521 sc->vbsc_cfg.vbc_capacity = size / VTBLK_BSIZE; /* 512-byte units */
522 sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */
525 * If Linux is presented with a seg_max greater than the virtio queue
526 * size, it can stumble into situations where it violates its own
527 * invariants and panics. For safety, we keep seg_max clamped, paying
528 * heed to the two extra descriptors needed for the header and status
531 sc->vbsc_cfg.vbc_seg_max = MIN(VTBLK_RINGSZ - 2, BLOCKIF_IOV_MAX);
532 sc->vbsc_cfg.vbc_geometry.cylinders = 0; /* no geometry */
533 sc->vbsc_cfg.vbc_geometry.heads = 0;
534 sc->vbsc_cfg.vbc_geometry.sectors = 0;
535 sc->vbsc_cfg.vbc_blk_size = sectsz;
536 sc->vbsc_cfg.vbc_topology.physical_block_exp =
537 (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0;
538 sc->vbsc_cfg.vbc_topology.alignment_offset =
539 (sto != 0) ? ((sts - sto) / sectsz) : 0;
540 sc->vbsc_cfg.vbc_topology.min_io_size = 0;
541 sc->vbsc_cfg.vbc_topology.opt_io_size = 0;
542 sc->vbsc_cfg.vbc_writeback = 0;
543 sc->vbsc_cfg.max_discard_sectors = VTBLK_MAX_DISCARD_SECT;
544 sc->vbsc_cfg.max_discard_seg = VTBLK_MAX_DISCARD_SEG;
545 sc->vbsc_cfg.discard_sector_alignment = MAX(sectsz, sts) / VTBLK_BSIZE;
548 * Should we move some of this into virtio.c? Could
549 * have the device, class, and subdev_0 as fields in
550 * the virtio constants structure.
552 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
553 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
554 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
555 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_BLOCK);
556 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
558 if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) {
559 blockif_close(sc->bc);
563 vi_set_io_bar(&sc->vbsc_vs, 0);
564 blockif_register_resize_callback(sc->bc, pci_vtblk_resized, sc);
569 pci_vtblk_cfgwrite(void *vsc __unused, int offset, int size __unused,
570 uint32_t value __unused)
573 DPRINTF(("vtblk: write to readonly reg %d", offset));
578 pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval)
580 struct pci_vtblk_softc *sc = vsc;
583 /* our caller has already verified offset and size */
584 ptr = (uint8_t *)&sc->vbsc_cfg + offset;
585 memcpy(retval, ptr, size);
589 static const struct pci_devemu pci_de_vblk = {
590 .pe_emu = "virtio-blk",
591 .pe_init = pci_vtblk_init,
592 .pe_legacy_config = blockif_legacy_config,
593 .pe_barwrite = vi_pci_write,
594 .pe_barread = vi_pci_read,
595 #ifdef BHYVE_SNAPSHOT
596 .pe_snapshot = vi_pci_snapshot,
597 .pe_pause = vi_pci_pause,
598 .pe_resume = vi_pci_resume,
601 PCI_EMUL_SET(pci_de_vblk);