1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * Copyright 2020 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 <machine/vmm_snapshot.h> 43 44 #include <errno.h> 45 #include <fcntl.h> 46 #include <stdio.h> 47 #include <stdlib.h> 48 #include <stdint.h> 49 #include <string.h> 50 #include <strings.h> 51 #include <unistd.h> 52 #include <assert.h> 53 #include <pthread.h> 54 #include <md5.h> 55 56 #include "bhyverun.h" 57 #include "config.h" 58 #include "debug.h" 59 #include "pci_emul.h" 60 #include "virtio.h" 61 #include "block_if.h" 62 63 #define VTBLK_BSIZE 512 64 #define VTBLK_RINGSZ 128 65 66 _Static_assert(VTBLK_RINGSZ <= BLOCKIF_RING_MAX, "Each ring entry must be able to queue a request"); 67 68 #define VTBLK_S_OK 0 69 #define VTBLK_S_IOERR 1 70 #define VTBLK_S_UNSUPP 2 71 72 #define VTBLK_BLK_ID_BYTES 20 + 1 73 74 /* Capability bits */ 75 #define VTBLK_F_BARRIER (1 << 0) /* Does host support barriers? */ 76 #define VTBLK_F_SIZE_MAX (1 << 1) /* Indicates maximum segment size */ 77 #define VTBLK_F_SEG_MAX (1 << 2) /* Indicates maximum # of segments */ 78 #define VTBLK_F_GEOMETRY (1 << 4) /* Legacy geometry available */ 79 #define VTBLK_F_RO (1 << 5) /* Disk is read-only */ 80 #define VTBLK_F_BLK_SIZE (1 << 6) /* Block size of disk is available*/ 81 #define VTBLK_F_SCSI (1 << 7) /* Supports scsi command passthru */ 82 #define VTBLK_F_FLUSH (1 << 9) /* Writeback mode enabled after reset */ 83 #define VTBLK_F_WCE (1 << 9) /* Legacy alias for FLUSH */ 84 #define VTBLK_F_TOPOLOGY (1 << 10) /* Topology information is available */ 85 #define VTBLK_F_CONFIG_WCE (1 << 11) /* Writeback mode available in config */ 86 #define VTBLK_F_MQ (1 << 12) /* Multi-Queue */ 87 #define VTBLK_F_DISCARD (1 << 13) /* Trim blocks */ 88 #define VTBLK_F_WRITE_ZEROES (1 << 14) /* Write zeros */ 89 90 /* 91 * Host capabilities 92 */ 93 #define VTBLK_S_HOSTCAPS \ 94 ( VTBLK_F_SEG_MAX | \ 95 VTBLK_F_BLK_SIZE | \ 96 VTBLK_F_FLUSH | \ 97 VTBLK_F_TOPOLOGY | \ 98 VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */ 99 100 /* 101 * The current blockif_delete() interface only allows a single delete 102 * request at a time. 103 */ 104 #define VTBLK_MAX_DISCARD_SEG 1 105 106 /* 107 * An arbitrary limit to prevent excessive latency due to large 108 * delete requests. 109 */ 110 #define VTBLK_MAX_DISCARD_SECT ((16 << 20) / VTBLK_BSIZE) /* 16 MiB */ 111 112 /* 113 * Config space "registers" 114 */ 115 struct vtblk_config { 116 uint64_t vbc_capacity; 117 uint32_t vbc_size_max; 118 uint32_t vbc_seg_max; 119 struct { 120 uint16_t cylinders; 121 uint8_t heads; 122 uint8_t sectors; 123 } vbc_geometry; 124 uint32_t vbc_blk_size; 125 struct { 126 uint8_t physical_block_exp; 127 uint8_t alignment_offset; 128 uint16_t min_io_size; 129 uint32_t opt_io_size; 130 } vbc_topology; 131 uint8_t vbc_writeback; 132 uint8_t unused0[1]; 133 uint16_t num_queues; 134 uint32_t max_discard_sectors; 135 uint32_t max_discard_seg; 136 uint32_t discard_sector_alignment; 137 uint32_t max_write_zeroes_sectors; 138 uint32_t max_write_zeroes_seg; 139 uint8_t write_zeroes_may_unmap; 140 uint8_t unused1[3]; 141 } __packed; 142 143 /* 144 * Fixed-size block header 145 */ 146 struct virtio_blk_hdr { 147 #define VBH_OP_READ 0 148 #define VBH_OP_WRITE 1 149 #define VBH_OP_SCSI_CMD 2 150 #define VBH_OP_SCSI_CMD_OUT 3 151 #define VBH_OP_FLUSH 4 152 #define VBH_OP_FLUSH_OUT 5 153 #define VBH_OP_IDENT 8 154 #define VBH_OP_DISCARD 11 155 #define VBH_OP_WRITE_ZEROES 13 156 157 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */ 158 uint32_t vbh_type; 159 uint32_t vbh_ioprio; 160 uint64_t vbh_sector; 161 } __packed; 162 163 /* 164 * Debug printf 165 */ 166 static int pci_vtblk_debug; 167 #define DPRINTF(params) if (pci_vtblk_debug) PRINTLN params 168 #define WPRINTF(params) PRINTLN params 169 170 struct pci_vtblk_ioreq { 171 struct blockif_req io_req; 172 struct pci_vtblk_softc *io_sc; 173 uint8_t *io_status; 174 uint16_t io_idx; 175 }; 176 177 struct virtio_blk_discard_write_zeroes { 178 uint64_t sector; 179 uint32_t num_sectors; 180 struct { 181 uint32_t unmap:1; 182 uint32_t reserved:31; 183 } flags; 184 }; 185 186 /* 187 * Per-device softc 188 */ 189 struct pci_vtblk_softc { 190 struct virtio_softc vbsc_vs; 191 pthread_mutex_t vsc_mtx; 192 struct vqueue_info vbsc_vq; 193 struct vtblk_config vbsc_cfg; 194 struct virtio_consts vbsc_consts; 195 struct blockif_ctxt *bc; 196 char vbsc_ident[VTBLK_BLK_ID_BYTES]; 197 struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ]; 198 }; 199 200 static void pci_vtblk_reset(void *); 201 static void pci_vtblk_notify(void *, struct vqueue_info *); 202 static int pci_vtblk_cfgread(void *, int, int, uint32_t *); 203 static int pci_vtblk_cfgwrite(void *, int, int, uint32_t); 204 #ifdef BHYVE_SNAPSHOT 205 static void pci_vtblk_pause(void *); 206 static void pci_vtblk_resume(void *); 207 static int pci_vtblk_snapshot(void *, struct vm_snapshot_meta *); 208 #endif 209 210 static struct virtio_consts vtblk_vi_consts = { 211 "vtblk", /* our name */ 212 1, /* we support 1 virtqueue */ 213 sizeof(struct vtblk_config), /* config reg size */ 214 pci_vtblk_reset, /* reset */ 215 pci_vtblk_notify, /* device-wide qnotify */ 216 pci_vtblk_cfgread, /* read PCI config */ 217 pci_vtblk_cfgwrite, /* write PCI config */ 218 NULL, /* apply negotiated features */ 219 VTBLK_S_HOSTCAPS, /* our capabilities */ 220 #ifdef BHYVE_SNAPSHOT 221 pci_vtblk_pause, /* pause blockif threads */ 222 pci_vtblk_resume, /* resume blockif threads */ 223 pci_vtblk_snapshot, /* save / restore device state */ 224 #endif 225 }; 226 227 static void 228 pci_vtblk_reset(void *vsc) 229 { 230 struct pci_vtblk_softc *sc = vsc; 231 232 DPRINTF(("vtblk: device reset requested !")); 233 vi_reset_dev(&sc->vbsc_vs); 234 } 235 236 static void 237 pci_vtblk_done_locked(struct pci_vtblk_ioreq *io, int err) 238 { 239 struct pci_vtblk_softc *sc = io->io_sc; 240 241 /* convert errno into a virtio block error return */ 242 if (err == EOPNOTSUPP || err == ENOSYS) 243 *io->io_status = VTBLK_S_UNSUPP; 244 else if (err != 0) 245 *io->io_status = VTBLK_S_IOERR; 246 else 247 *io->io_status = VTBLK_S_OK; 248 249 /* 250 * Return the descriptor back to the host. 251 * We wrote 1 byte (our status) to host. 252 */ 253 vq_relchain(&sc->vbsc_vq, io->io_idx, 1); 254 vq_endchains(&sc->vbsc_vq, 0); 255 } 256 257 #ifdef BHYVE_SNAPSHOT 258 static void 259 pci_vtblk_pause(void *vsc) 260 { 261 struct pci_vtblk_softc *sc = vsc; 262 263 DPRINTF(("vtblk: device pause requested !\n")); 264 blockif_pause(sc->bc); 265 } 266 267 static void 268 pci_vtblk_resume(void *vsc) 269 { 270 struct pci_vtblk_softc *sc = vsc; 271 272 DPRINTF(("vtblk: device resume requested !\n")); 273 blockif_resume(sc->bc); 274 } 275 276 static int 277 pci_vtblk_snapshot(void *vsc, struct vm_snapshot_meta *meta) 278 { 279 int ret; 280 struct pci_vtblk_softc *sc = vsc; 281 282 SNAPSHOT_VAR_OR_LEAVE(sc->vbsc_cfg, meta, ret, done); 283 SNAPSHOT_BUF_OR_LEAVE(sc->vbsc_ident, sizeof(sc->vbsc_ident), 284 meta, ret, done); 285 286 done: 287 return (ret); 288 } 289 #endif 290 291 static void 292 pci_vtblk_done(struct blockif_req *br, int err) 293 { 294 struct pci_vtblk_ioreq *io = br->br_param; 295 struct pci_vtblk_softc *sc = io->io_sc; 296 297 pthread_mutex_lock(&sc->vsc_mtx); 298 pci_vtblk_done_locked(io, err); 299 pthread_mutex_unlock(&sc->vsc_mtx); 300 } 301 302 static void 303 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq) 304 { 305 struct virtio_blk_hdr *vbh; 306 struct pci_vtblk_ioreq *io; 307 int i, n; 308 int err; 309 ssize_t iolen; 310 int writeop, type; 311 struct iovec iov[BLOCKIF_IOV_MAX + 2]; 312 uint16_t idx, flags[BLOCKIF_IOV_MAX + 2]; 313 struct virtio_blk_discard_write_zeroes *discard; 314 315 n = vq_getchain(vq, &idx, iov, BLOCKIF_IOV_MAX + 2, flags); 316 317 /* 318 * The first descriptor will be the read-only fixed header, 319 * and the last is for status (hence +2 above and below). 320 * The remaining iov's are the actual data I/O vectors. 321 * 322 * XXX - note - this fails on crash dump, which does a 323 * VIRTIO_BLK_T_FLUSH with a zero transfer length 324 */ 325 assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2); 326 327 io = &sc->vbsc_ios[idx]; 328 assert((flags[0] & VRING_DESC_F_WRITE) == 0); 329 assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr)); 330 vbh = (struct virtio_blk_hdr *)iov[0].iov_base; 331 memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2)); 332 io->io_req.br_iovcnt = n - 2; 333 io->io_req.br_offset = vbh->vbh_sector * VTBLK_BSIZE; 334 io->io_status = (uint8_t *)iov[--n].iov_base; 335 assert(iov[n].iov_len == 1); 336 assert(flags[n] & VRING_DESC_F_WRITE); 337 338 /* 339 * XXX 340 * The guest should not be setting the BARRIER flag because 341 * we don't advertise the capability. 342 */ 343 type = vbh->vbh_type & ~VBH_FLAG_BARRIER; 344 writeop = (type == VBH_OP_WRITE || type == VBH_OP_DISCARD); 345 346 iolen = 0; 347 for (i = 1; i < n; i++) { 348 /* 349 * - write op implies read-only descriptor, 350 * - read/ident op implies write-only descriptor, 351 * therefore test the inverse of the descriptor bit 352 * to the op. 353 */ 354 assert(((flags[i] & VRING_DESC_F_WRITE) == 0) == writeop); 355 iolen += iov[i].iov_len; 356 } 357 io->io_req.br_resid = iolen; 358 359 DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld", 360 writeop ? "write/discard" : "read/ident", iolen, i - 1, 361 io->io_req.br_offset)); 362 363 switch (type) { 364 case VBH_OP_READ: 365 err = blockif_read(sc->bc, &io->io_req); 366 break; 367 case VBH_OP_WRITE: 368 err = blockif_write(sc->bc, &io->io_req); 369 break; 370 case VBH_OP_DISCARD: 371 /* 372 * We currently only support a single request, if the guest 373 * has submitted a request that doesn't conform to the 374 * requirements, we return a error. 375 */ 376 if (iov[1].iov_len != sizeof (*discard)) { 377 pci_vtblk_done_locked(io, EINVAL); 378 return; 379 } 380 381 /* The segments to discard are provided rather than data */ 382 discard = (struct virtio_blk_discard_write_zeroes *) 383 iov[1].iov_base; 384 385 /* 386 * virtio v1.1 5.2.6.2: 387 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP 388 * for discard and write zeroes commands if any unknown flag is 389 * set. Furthermore, the device MUST set the status byte to 390 * VIRTIO_BLK_S_UNSUPP for discard commands if the unmap flag 391 * is set. 392 * 393 * Currently there are no known flags for a DISCARD request. 394 */ 395 if (discard->flags.unmap != 0 || discard->flags.reserved != 0) { 396 pci_vtblk_done_locked(io, ENOTSUP); 397 return; 398 } 399 400 /* Make sure the request doesn't exceed our size limit */ 401 if (discard->num_sectors > VTBLK_MAX_DISCARD_SECT) { 402 pci_vtblk_done_locked(io, EINVAL); 403 return; 404 } 405 406 io->io_req.br_offset = discard->sector * VTBLK_BSIZE; 407 io->io_req.br_resid = discard->num_sectors * VTBLK_BSIZE; 408 err = blockif_delete(sc->bc, &io->io_req); 409 break; 410 case VBH_OP_FLUSH: 411 case VBH_OP_FLUSH_OUT: 412 err = blockif_flush(sc->bc, &io->io_req); 413 break; 414 case VBH_OP_IDENT: 415 /* Assume a single buffer */ 416 /* S/n equal to buffer is not zero-terminated. */ 417 memset(iov[1].iov_base, 0, iov[1].iov_len); 418 strncpy(iov[1].iov_base, sc->vbsc_ident, 419 MIN(iov[1].iov_len, sizeof(sc->vbsc_ident))); 420 pci_vtblk_done_locked(io, 0); 421 return; 422 default: 423 pci_vtblk_done_locked(io, EOPNOTSUPP); 424 return; 425 } 426 assert(err == 0); 427 } 428 429 static void 430 pci_vtblk_notify(void *vsc, struct vqueue_info *vq) 431 { 432 struct pci_vtblk_softc *sc = vsc; 433 434 while (vq_has_descs(vq)) 435 pci_vtblk_proc(sc, vq); 436 } 437 438 static int 439 pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, nvlist_t *nvl) 440 { 441 char bident[sizeof("XX:X:X")]; 442 struct blockif_ctxt *bctxt; 443 const char *path; 444 MD5_CTX mdctx; 445 u_char digest[16]; 446 struct pci_vtblk_softc *sc; 447 off_t size; 448 int i, sectsz, sts, sto; 449 450 /* 451 * The supplied backing file has to exist 452 */ 453 snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func); 454 bctxt = blockif_open(nvl, bident); 455 if (bctxt == NULL) { 456 perror("Could not open backing file"); 457 return (1); 458 } 459 460 size = blockif_size(bctxt); 461 sectsz = blockif_sectsz(bctxt); 462 blockif_psectsz(bctxt, &sts, &sto); 463 464 sc = calloc(1, sizeof(struct pci_vtblk_softc)); 465 sc->bc = bctxt; 466 for (i = 0; i < VTBLK_RINGSZ; i++) { 467 struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i]; 468 io->io_req.br_callback = pci_vtblk_done; 469 io->io_req.br_param = io; 470 io->io_sc = sc; 471 io->io_idx = i; 472 } 473 474 bcopy(&vtblk_vi_consts, &sc->vbsc_consts, sizeof (vtblk_vi_consts)); 475 if (blockif_candelete(sc->bc)) 476 sc->vbsc_consts.vc_hv_caps |= VTBLK_F_DISCARD; 477 478 pthread_mutex_init(&sc->vsc_mtx, NULL); 479 480 /* init virtio softc and virtqueues */ 481 vi_softc_linkup(&sc->vbsc_vs, &sc->vbsc_consts, sc, pi, &sc->vbsc_vq); 482 sc->vbsc_vs.vs_mtx = &sc->vsc_mtx; 483 484 sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ; 485 /* sc->vbsc_vq.vq_notify = we have no per-queue notify */ 486 487 /* 488 * Create an identifier for the backing file. Use parts of the 489 * md5 sum of the filename 490 */ 491 path = get_config_value_node(nvl, "path"); 492 MD5Init(&mdctx); 493 MD5Update(&mdctx, path, strlen(path)); 494 MD5Final(digest, &mdctx); 495 snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES, 496 "BHYVE-%02X%02X-%02X%02X-%02X%02X", 497 digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]); 498 499 /* setup virtio block config space */ 500 sc->vbsc_cfg.vbc_capacity = size / VTBLK_BSIZE; /* 512-byte units */ 501 sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */ 502 503 /* 504 * If Linux is presented with a seg_max greater than the virtio queue 505 * size, it can stumble into situations where it violates its own 506 * invariants and panics. For safety, we keep seg_max clamped, paying 507 * heed to the two extra descriptors needed for the header and status 508 * of a request. 509 */ 510 sc->vbsc_cfg.vbc_seg_max = MIN(VTBLK_RINGSZ - 2, BLOCKIF_IOV_MAX); 511 sc->vbsc_cfg.vbc_geometry.cylinders = 0; /* no geometry */ 512 sc->vbsc_cfg.vbc_geometry.heads = 0; 513 sc->vbsc_cfg.vbc_geometry.sectors = 0; 514 sc->vbsc_cfg.vbc_blk_size = sectsz; 515 sc->vbsc_cfg.vbc_topology.physical_block_exp = 516 (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0; 517 sc->vbsc_cfg.vbc_topology.alignment_offset = 518 (sto != 0) ? ((sts - sto) / sectsz) : 0; 519 sc->vbsc_cfg.vbc_topology.min_io_size = 0; 520 sc->vbsc_cfg.vbc_topology.opt_io_size = 0; 521 sc->vbsc_cfg.vbc_writeback = 0; 522 sc->vbsc_cfg.max_discard_sectors = VTBLK_MAX_DISCARD_SECT; 523 sc->vbsc_cfg.max_discard_seg = VTBLK_MAX_DISCARD_SEG; 524 sc->vbsc_cfg.discard_sector_alignment = MAX(sectsz, sts) / VTBLK_BSIZE; 525 526 /* 527 * Should we move some of this into virtio.c? Could 528 * have the device, class, and subdev_0 as fields in 529 * the virtio constants structure. 530 */ 531 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK); 532 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR); 533 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE); 534 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_BLOCK); 535 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR); 536 537 if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) { 538 blockif_close(sc->bc); 539 free(sc); 540 return (1); 541 } 542 vi_set_io_bar(&sc->vbsc_vs, 0); 543 return (0); 544 } 545 546 static int 547 pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value) 548 { 549 550 DPRINTF(("vtblk: write to readonly reg %d", offset)); 551 return (1); 552 } 553 554 static int 555 pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval) 556 { 557 struct pci_vtblk_softc *sc = vsc; 558 void *ptr; 559 560 /* our caller has already verified offset and size */ 561 ptr = (uint8_t *)&sc->vbsc_cfg + offset; 562 memcpy(retval, ptr, size); 563 return (0); 564 } 565 566 struct pci_devemu pci_de_vblk = { 567 .pe_emu = "virtio-blk", 568 .pe_init = pci_vtblk_init, 569 .pe_legacy_config = blockif_legacy_config, 570 .pe_barwrite = vi_pci_write, 571 .pe_barread = vi_pci_read, 572 #ifdef BHYVE_SNAPSHOT 573 .pe_snapshot = vi_pci_snapshot, 574 #endif 575 }; 576 PCI_EMUL_SET(pci_de_vblk); 577