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