1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * Copyright (c) 2019 Joyent, Inc. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 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. 16 * 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 27 * SUCH DAMAGE. 28 * 29 * $FreeBSD$ 30 */ 31 /* 32 * This file and its contents are supplied under the terms of the 33 * Common Development and Distribution License ("CDDL"), version 1.0. 34 * You may only use this file in accordance with the terms of version 35 * 1.0 of the CDDL. 36 * 37 * A full copy of the text of the CDDL should have accompanied this 38 * source. A copy of the CDDL is also available via the Internet at 39 * http://www.illumos.org/license/CDDL. 40 * 41 * Copyright 2014 Pluribus Networks Inc. 42 * Copyright 2018 Joyent, Inc. 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include <sys/param.h> 49 #include <sys/linker_set.h> 50 #include <sys/stat.h> 51 #include <sys/uio.h> 52 #include <sys/ioctl.h> 53 #include <sys/disk.h> 54 55 #include <errno.h> 56 #include <fcntl.h> 57 #include <stdio.h> 58 #include <stdlib.h> 59 #include <stdint.h> 60 #include <string.h> 61 #include <strings.h> 62 #include <unistd.h> 63 #include <assert.h> 64 #include <pthread.h> 65 #include <md5.h> 66 67 #include "bhyverun.h" 68 #include "pci_emul.h" 69 #include "virtio.h" 70 #include "block_if.h" 71 72 #define VTBLK_RINGSZ 128 73 74 _Static_assert(VTBLK_RINGSZ <= BLOCKIF_RING_MAX, "Each ring entry must be able to queue a request"); 75 76 #define VTBLK_S_OK 0 77 #define VTBLK_S_IOERR 1 78 #define VTBLK_S_UNSUPP 2 79 80 #define VTBLK_BLK_ID_BYTES 20 + 1 81 82 /* Capability bits */ 83 #define VTBLK_F_SEG_MAX (1 << 2) /* Maximum request segments */ 84 #define VTBLK_F_BLK_SIZE (1 << 6) /* cfg block size valid */ 85 #define VTBLK_F_FLUSH (1 << 9) /* Cache flush support */ 86 #define VTBLK_F_TOPOLOGY (1 << 10) /* Optimal I/O alignment */ 87 88 /* 89 * Host capabilities 90 */ 91 #define VTBLK_S_HOSTCAPS \ 92 ( VTBLK_F_SEG_MAX | \ 93 VTBLK_F_BLK_SIZE | \ 94 VTBLK_F_FLUSH | \ 95 VTBLK_F_TOPOLOGY | \ 96 VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */ 97 98 /* 99 * Config space "registers" 100 */ 101 struct vtblk_config { 102 uint64_t vbc_capacity; 103 uint32_t vbc_size_max; 104 uint32_t vbc_seg_max; 105 struct { 106 uint16_t cylinders; 107 uint8_t heads; 108 uint8_t sectors; 109 } vbc_geometry; 110 uint32_t vbc_blk_size; 111 struct { 112 uint8_t physical_block_exp; 113 uint8_t alignment_offset; 114 uint16_t min_io_size; 115 uint32_t opt_io_size; 116 } vbc_topology; 117 uint8_t vbc_writeback; 118 } __packed; 119 120 /* 121 * Fixed-size block header 122 */ 123 struct virtio_blk_hdr { 124 #define VBH_OP_READ 0 125 #define VBH_OP_WRITE 1 126 #define VBH_OP_FLUSH 4 127 #define VBH_OP_FLUSH_OUT 5 128 #define VBH_OP_IDENT 8 129 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */ 130 uint32_t vbh_type; 131 uint32_t vbh_ioprio; 132 uint64_t vbh_sector; 133 } __packed; 134 135 /* 136 * Debug printf 137 */ 138 static int pci_vtblk_debug; 139 #define DPRINTF(params) if (pci_vtblk_debug) printf params 140 #define WPRINTF(params) printf params 141 142 struct pci_vtblk_ioreq { 143 struct blockif_req io_req; 144 struct pci_vtblk_softc *io_sc; 145 uint8_t *io_status; 146 uint16_t io_idx; 147 }; 148 149 /* 150 * Per-device softc 151 */ 152 struct pci_vtblk_softc { 153 struct virtio_softc vbsc_vs; 154 pthread_mutex_t vsc_mtx; 155 struct vqueue_info vbsc_vq; 156 struct vtblk_config vbsc_cfg; 157 struct blockif_ctxt *bc; 158 #ifndef __FreeBSD__ 159 int vbsc_wce; 160 #endif 161 char vbsc_ident[VTBLK_BLK_ID_BYTES]; 162 struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ]; 163 }; 164 165 static void pci_vtblk_reset(void *); 166 static void pci_vtblk_notify(void *, struct vqueue_info *); 167 static int pci_vtblk_cfgread(void *, int, int, uint32_t *); 168 static int pci_vtblk_cfgwrite(void *, int, int, uint32_t); 169 #ifndef __FreeBSD__ 170 static void pci_vtblk_apply_feats(void *, uint64_t); 171 #endif 172 173 static struct virtio_consts vtblk_vi_consts = { 174 "vtblk", /* our name */ 175 1, /* we support 1 virtqueue */ 176 sizeof(struct vtblk_config), /* config reg size */ 177 pci_vtblk_reset, /* reset */ 178 pci_vtblk_notify, /* device-wide qnotify */ 179 pci_vtblk_cfgread, /* read PCI config */ 180 pci_vtblk_cfgwrite, /* write PCI config */ 181 #ifndef __FreeBSD__ 182 pci_vtblk_apply_feats, /* apply negotiated features */ 183 #else 184 NULL, /* apply negotiated features */ 185 #endif 186 VTBLK_S_HOSTCAPS, /* our capabilities */ 187 }; 188 189 static void 190 pci_vtblk_reset(void *vsc) 191 { 192 struct pci_vtblk_softc *sc = vsc; 193 194 DPRINTF(("vtblk: device reset requested !\n")); 195 vi_reset_dev(&sc->vbsc_vs); 196 #ifndef __FreeBSD__ 197 /* Disable write cache until FLUSH feature is negotiated */ 198 (void) blockif_set_wce(sc->bc, 0); 199 sc->vbsc_wce = 0; 200 #endif 201 } 202 203 static void 204 pci_vtblk_done_locked(struct pci_vtblk_ioreq *io, int err) 205 { 206 struct pci_vtblk_softc *sc = io->io_sc; 207 208 /* convert errno into a virtio block error return */ 209 if (err == EOPNOTSUPP || err == ENOSYS) 210 *io->io_status = VTBLK_S_UNSUPP; 211 else if (err != 0) 212 *io->io_status = VTBLK_S_IOERR; 213 else 214 *io->io_status = VTBLK_S_OK; 215 216 /* 217 * Return the descriptor back to the host. 218 * We wrote 1 byte (our status) to host. 219 */ 220 vq_relchain(&sc->vbsc_vq, io->io_idx, 1); 221 vq_endchains(&sc->vbsc_vq, 0); 222 } 223 224 static void 225 pci_vtblk_done(struct blockif_req *br, int err) 226 { 227 struct pci_vtblk_ioreq *io = br->br_param; 228 struct pci_vtblk_softc *sc = io->io_sc; 229 230 pthread_mutex_lock(&sc->vsc_mtx); 231 pci_vtblk_done_locked(io, err); 232 pthread_mutex_unlock(&sc->vsc_mtx); 233 } 234 235 static void 236 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq) 237 { 238 struct virtio_blk_hdr *vbh; 239 struct pci_vtblk_ioreq *io; 240 int i, n; 241 int err; 242 ssize_t iolen; 243 int writeop, type; 244 struct iovec iov[BLOCKIF_IOV_MAX + 2]; 245 uint16_t idx, flags[BLOCKIF_IOV_MAX + 2]; 246 247 n = vq_getchain(vq, &idx, iov, BLOCKIF_IOV_MAX + 2, flags); 248 249 /* 250 * The first descriptor will be the read-only fixed header, 251 * and the last is for status (hence +2 above and below). 252 * The remaining iov's are the actual data I/O vectors. 253 * 254 * XXX - note - this fails on crash dump, which does a 255 * VIRTIO_BLK_T_FLUSH with a zero transfer length 256 */ 257 assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2); 258 259 io = &sc->vbsc_ios[idx]; 260 assert((flags[0] & VRING_DESC_F_WRITE) == 0); 261 assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr)); 262 vbh = (struct virtio_blk_hdr *)iov[0].iov_base; 263 memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2)); 264 io->io_req.br_iovcnt = n - 2; 265 io->io_req.br_offset = vbh->vbh_sector * DEV_BSIZE; 266 io->io_status = (uint8_t *)iov[--n].iov_base; 267 assert(iov[n].iov_len == 1); 268 assert(flags[n] & VRING_DESC_F_WRITE); 269 270 /* 271 * XXX 272 * The guest should not be setting the BARRIER flag because 273 * we don't advertise the capability. 274 */ 275 type = vbh->vbh_type & ~VBH_FLAG_BARRIER; 276 writeop = (type == VBH_OP_WRITE); 277 278 iolen = 0; 279 for (i = 1; i < n; i++) { 280 /* 281 * - write op implies read-only descriptor, 282 * - read/ident op implies write-only descriptor, 283 * therefore test the inverse of the descriptor bit 284 * to the op. 285 */ 286 assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop); 287 iolen += iov[i].iov_len; 288 } 289 io->io_req.br_resid = iolen; 290 291 DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld\n\r", 292 writeop ? "write" : "read/ident", iolen, i - 1, 293 io->io_req.br_offset)); 294 295 switch (type) { 296 case VBH_OP_READ: 297 err = blockif_read(sc->bc, &io->io_req); 298 break; 299 case VBH_OP_WRITE: 300 err = blockif_write(sc->bc, &io->io_req); 301 break; 302 case VBH_OP_FLUSH: 303 case VBH_OP_FLUSH_OUT: 304 err = blockif_flush(sc->bc, &io->io_req); 305 break; 306 case VBH_OP_IDENT: 307 /* Assume a single buffer */ 308 /* S/n equal to buffer is not zero-terminated. */ 309 memset(iov[1].iov_base, 0, iov[1].iov_len); 310 strncpy(iov[1].iov_base, sc->vbsc_ident, 311 MIN(iov[1].iov_len, sizeof(sc->vbsc_ident))); 312 pci_vtblk_done_locked(io, 0); 313 return; 314 default: 315 pci_vtblk_done_locked(io, EOPNOTSUPP); 316 return; 317 } 318 assert(err == 0); 319 } 320 321 static void 322 pci_vtblk_notify(void *vsc, struct vqueue_info *vq) 323 { 324 struct pci_vtblk_softc *sc = vsc; 325 326 while (vq_has_descs(vq)) 327 pci_vtblk_proc(sc, vq); 328 } 329 330 static int 331 pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts) 332 { 333 char bident[sizeof("XX:X:X")]; 334 struct blockif_ctxt *bctxt; 335 MD5_CTX mdctx; 336 u_char digest[16]; 337 struct pci_vtblk_softc *sc; 338 off_t size; 339 int i, sectsz, sts, sto; 340 341 if (opts == NULL) { 342 printf("virtio-block: backing device required\n"); 343 return (1); 344 } 345 346 /* 347 * The supplied backing file has to exist 348 */ 349 snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func); 350 bctxt = blockif_open(opts, bident); 351 if (bctxt == NULL) { 352 perror("Could not open backing file"); 353 return (1); 354 } 355 356 size = blockif_size(bctxt); 357 sectsz = blockif_sectsz(bctxt); 358 blockif_psectsz(bctxt, &sts, &sto); 359 360 sc = calloc(1, sizeof(struct pci_vtblk_softc)); 361 sc->bc = bctxt; 362 for (i = 0; i < VTBLK_RINGSZ; i++) { 363 struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i]; 364 io->io_req.br_callback = pci_vtblk_done; 365 io->io_req.br_param = io; 366 io->io_sc = sc; 367 io->io_idx = i; 368 } 369 370 #ifndef __FreeBSD__ 371 /* Disable write cache until FLUSH feature is negotiated */ 372 (void) blockif_set_wce(sc->bc, 0); 373 sc->vbsc_wce = 0; 374 #endif 375 376 pthread_mutex_init(&sc->vsc_mtx, NULL); 377 378 /* init virtio softc and virtqueues */ 379 vi_softc_linkup(&sc->vbsc_vs, &vtblk_vi_consts, sc, pi, &sc->vbsc_vq); 380 sc->vbsc_vs.vs_mtx = &sc->vsc_mtx; 381 382 sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ; 383 /* sc->vbsc_vq.vq_notify = we have no per-queue notify */ 384 385 /* 386 * Create an identifier for the backing file. Use parts of the 387 * md5 sum of the filename 388 */ 389 MD5Init(&mdctx); 390 MD5Update(&mdctx, opts, strlen(opts)); 391 MD5Final(digest, &mdctx); 392 snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES, 393 "BHYVE-%02X%02X-%02X%02X-%02X%02X", 394 digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]); 395 396 /* setup virtio block config space */ 397 sc->vbsc_cfg.vbc_capacity = size / DEV_BSIZE; /* 512-byte units */ 398 sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */ 399 400 /* 401 * If Linux is presented with a seg_max greater than the virtio queue 402 * size, it can stumble into situations where it violates its own 403 * invariants and panics. For safety, we keep seg_max clamped, paying 404 * heed to the two extra descriptors needed for the header and status 405 * of a request. 406 */ 407 sc->vbsc_cfg.vbc_seg_max = MIN(VTBLK_RINGSZ - 2, BLOCKIF_IOV_MAX); 408 sc->vbsc_cfg.vbc_geometry.cylinders = 0; /* no geometry */ 409 sc->vbsc_cfg.vbc_geometry.heads = 0; 410 sc->vbsc_cfg.vbc_geometry.sectors = 0; 411 sc->vbsc_cfg.vbc_blk_size = sectsz; 412 sc->vbsc_cfg.vbc_topology.physical_block_exp = 413 (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0; 414 sc->vbsc_cfg.vbc_topology.alignment_offset = 415 (sto != 0) ? ((sts - sto) / sectsz) : 0; 416 sc->vbsc_cfg.vbc_topology.min_io_size = 0; 417 sc->vbsc_cfg.vbc_topology.opt_io_size = 0; 418 sc->vbsc_cfg.vbc_writeback = 0; 419 420 /* 421 * Should we move some of this into virtio.c? Could 422 * have the device, class, and subdev_0 as fields in 423 * the virtio constants structure. 424 */ 425 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK); 426 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR); 427 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE); 428 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK); 429 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR); 430 431 if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) { 432 blockif_close(sc->bc); 433 free(sc); 434 return (1); 435 } 436 vi_set_io_bar(&sc->vbsc_vs, 0); 437 return (0); 438 } 439 440 static int 441 pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value) 442 { 443 444 DPRINTF(("vtblk: write to readonly reg %d\n\r", offset)); 445 return (1); 446 } 447 448 static int 449 pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval) 450 { 451 struct pci_vtblk_softc *sc = vsc; 452 void *ptr; 453 454 /* our caller has already verified offset and size */ 455 ptr = (uint8_t *)&sc->vbsc_cfg + offset; 456 memcpy(retval, ptr, size); 457 return (0); 458 } 459 460 #ifndef __FreeBSD__ 461 void 462 pci_vtblk_apply_feats(void *vsc, uint64_t caps) 463 { 464 struct pci_vtblk_softc *sc = vsc; 465 const int wce_next = ((caps & VTBLK_F_FLUSH) != 0) ? 1 : 0; 466 467 if (sc->vbsc_wce != wce_next) { 468 (void) blockif_set_wce(sc->bc, wce_next); 469 sc->vbsc_wce = wce_next; 470 } 471 } 472 #endif /* __FreeBSD__ */ 473 474 struct pci_devemu pci_de_vblk = { 475 .pe_emu = "virtio-blk", 476 .pe_init = pci_vtblk_init, 477 .pe_barwrite = vi_pci_write, 478 .pe_barread = vi_pci_read 479 }; 480 PCI_EMUL_SET(pci_de_vblk); 481