1 /* 2 * Virtio Block Device 3 * 4 * Copyright IBM, Corp. 2007 5 * 6 * Authors: 7 * Anthony Liguori <aliguori@us.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 */ 13 14 #include "qemu/osdep.h" 15 #include "qapi/error.h" 16 #include "qemu/iov.h" 17 #include "qemu/module.h" 18 #include "qemu/error-report.h" 19 #include "trace.h" 20 #include "hw/block/block.h" 21 #include "sysemu/blockdev.h" 22 #include "hw/virtio/virtio-blk.h" 23 #include "dataplane/virtio-blk.h" 24 #include "scsi/constants.h" 25 #ifdef __linux__ 26 # include <scsi/sg.h> 27 #endif 28 #include "hw/virtio/virtio-bus.h" 29 #include "migration/qemu-file-types.h" 30 #include "hw/virtio/virtio-access.h" 31 32 /* Config size before the discard support (hide associated config fields) */ 33 #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \ 34 max_discard_sectors) 35 /* 36 * Starting from the discard feature, we can use this array to properly 37 * set the config size depending on the features enabled. 38 */ 39 static VirtIOFeature feature_sizes[] = { 40 {.flags = 1ULL << VIRTIO_BLK_F_DISCARD, 41 .end = virtio_endof(struct virtio_blk_config, discard_sector_alignment)}, 42 {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES, 43 .end = virtio_endof(struct virtio_blk_config, write_zeroes_may_unmap)}, 44 {} 45 }; 46 47 static void virtio_blk_set_config_size(VirtIOBlock *s, uint64_t host_features) 48 { 49 s->config_size = MAX(VIRTIO_BLK_CFG_SIZE, 50 virtio_feature_get_config_size(feature_sizes, host_features)); 51 52 assert(s->config_size <= sizeof(struct virtio_blk_config)); 53 } 54 55 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq, 56 VirtIOBlockReq *req) 57 { 58 req->dev = s; 59 req->vq = vq; 60 req->qiov.size = 0; 61 req->in_len = 0; 62 req->next = NULL; 63 req->mr_next = NULL; 64 } 65 66 static void virtio_blk_free_request(VirtIOBlockReq *req) 67 { 68 g_free(req); 69 } 70 71 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) 72 { 73 VirtIOBlock *s = req->dev; 74 VirtIODevice *vdev = VIRTIO_DEVICE(s); 75 76 trace_virtio_blk_req_complete(vdev, req, status); 77 78 stb_p(&req->in->status, status); 79 virtqueue_push(req->vq, &req->elem, req->in_len); 80 if (s->dataplane_started && !s->dataplane_disabled) { 81 virtio_blk_data_plane_notify(s->dataplane, req->vq); 82 } else { 83 virtio_notify(vdev, req->vq); 84 } 85 } 86 87 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, 88 bool is_read, bool acct_failed) 89 { 90 VirtIOBlock *s = req->dev; 91 BlockErrorAction action = blk_get_error_action(s->blk, is_read, error); 92 93 if (action == BLOCK_ERROR_ACTION_STOP) { 94 /* Break the link as the next request is going to be parsed from the 95 * ring again. Otherwise we may end up doing a double completion! */ 96 req->mr_next = NULL; 97 req->next = s->rq; 98 s->rq = req; 99 } else if (action == BLOCK_ERROR_ACTION_REPORT) { 100 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); 101 if (acct_failed) { 102 block_acct_failed(blk_get_stats(s->blk), &req->acct); 103 } 104 virtio_blk_free_request(req); 105 } 106 107 blk_error_action(s->blk, action, is_read, error); 108 return action != BLOCK_ERROR_ACTION_IGNORE; 109 } 110 111 static void virtio_blk_rw_complete(void *opaque, int ret) 112 { 113 VirtIOBlockReq *next = opaque; 114 VirtIOBlock *s = next->dev; 115 VirtIODevice *vdev = VIRTIO_DEVICE(s); 116 117 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); 118 while (next) { 119 VirtIOBlockReq *req = next; 120 next = req->mr_next; 121 trace_virtio_blk_rw_complete(vdev, req, ret); 122 123 if (req->qiov.nalloc != -1) { 124 /* If nalloc is != -1 req->qiov is a local copy of the original 125 * external iovec. It was allocated in submit_requests to be 126 * able to merge requests. */ 127 qemu_iovec_destroy(&req->qiov); 128 } 129 130 if (ret) { 131 int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type); 132 bool is_read = !(p & VIRTIO_BLK_T_OUT); 133 /* Note that memory may be dirtied on read failure. If the 134 * virtio request is not completed here, as is the case for 135 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied 136 * correctly during live migration. While this is ugly, 137 * it is acceptable because the device is free to write to 138 * the memory until the request is completed (which will 139 * happen on the other side of the migration). 140 */ 141 if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) { 142 continue; 143 } 144 } 145 146 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 147 block_acct_done(blk_get_stats(s->blk), &req->acct); 148 virtio_blk_free_request(req); 149 } 150 aio_context_release(blk_get_aio_context(s->conf.conf.blk)); 151 } 152 153 static void virtio_blk_flush_complete(void *opaque, int ret) 154 { 155 VirtIOBlockReq *req = opaque; 156 VirtIOBlock *s = req->dev; 157 158 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); 159 if (ret) { 160 if (virtio_blk_handle_rw_error(req, -ret, 0, true)) { 161 goto out; 162 } 163 } 164 165 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 166 block_acct_done(blk_get_stats(s->blk), &req->acct); 167 virtio_blk_free_request(req); 168 169 out: 170 aio_context_release(blk_get_aio_context(s->conf.conf.blk)); 171 } 172 173 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret) 174 { 175 VirtIOBlockReq *req = opaque; 176 VirtIOBlock *s = req->dev; 177 bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) & 178 ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES; 179 180 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); 181 if (ret) { 182 if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) { 183 goto out; 184 } 185 } 186 187 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 188 if (is_write_zeroes) { 189 block_acct_done(blk_get_stats(s->blk), &req->acct); 190 } 191 virtio_blk_free_request(req); 192 193 out: 194 aio_context_release(blk_get_aio_context(s->conf.conf.blk)); 195 } 196 197 #ifdef __linux__ 198 199 typedef struct { 200 VirtIOBlockReq *req; 201 struct sg_io_hdr hdr; 202 } VirtIOBlockIoctlReq; 203 204 static void virtio_blk_ioctl_complete(void *opaque, int status) 205 { 206 VirtIOBlockIoctlReq *ioctl_req = opaque; 207 VirtIOBlockReq *req = ioctl_req->req; 208 VirtIOBlock *s = req->dev; 209 VirtIODevice *vdev = VIRTIO_DEVICE(s); 210 struct virtio_scsi_inhdr *scsi; 211 struct sg_io_hdr *hdr; 212 213 scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; 214 215 if (status) { 216 status = VIRTIO_BLK_S_UNSUPP; 217 virtio_stl_p(vdev, &scsi->errors, 255); 218 goto out; 219 } 220 221 hdr = &ioctl_req->hdr; 222 /* 223 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) 224 * clear the masked_status field [hence status gets cleared too, see 225 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED 226 * status has occurred. However they do set DRIVER_SENSE in driver_status 227 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. 228 */ 229 if (hdr->status == 0 && hdr->sb_len_wr > 0) { 230 hdr->status = CHECK_CONDITION; 231 } 232 233 virtio_stl_p(vdev, &scsi->errors, 234 hdr->status | (hdr->msg_status << 8) | 235 (hdr->host_status << 16) | (hdr->driver_status << 24)); 236 virtio_stl_p(vdev, &scsi->residual, hdr->resid); 237 virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr); 238 virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len); 239 240 out: 241 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); 242 virtio_blk_req_complete(req, status); 243 virtio_blk_free_request(req); 244 aio_context_release(blk_get_aio_context(s->conf.conf.blk)); 245 g_free(ioctl_req); 246 } 247 248 #endif 249 250 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq) 251 { 252 VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq)); 253 254 if (req) { 255 virtio_blk_init_request(s, vq, req); 256 } 257 return req; 258 } 259 260 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req) 261 { 262 int status = VIRTIO_BLK_S_OK; 263 struct virtio_scsi_inhdr *scsi = NULL; 264 VirtIOBlock *blk = req->dev; 265 VirtIODevice *vdev = VIRTIO_DEVICE(blk); 266 VirtQueueElement *elem = &req->elem; 267 268 #ifdef __linux__ 269 int i; 270 VirtIOBlockIoctlReq *ioctl_req; 271 BlockAIOCB *acb; 272 #endif 273 274 /* 275 * We require at least one output segment each for the virtio_blk_outhdr 276 * and the SCSI command block. 277 * 278 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr 279 * and the sense buffer pointer in the input segments. 280 */ 281 if (elem->out_num < 2 || elem->in_num < 3) { 282 status = VIRTIO_BLK_S_IOERR; 283 goto fail; 284 } 285 286 /* 287 * The scsi inhdr is placed in the second-to-last input segment, just 288 * before the regular inhdr. 289 */ 290 scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base; 291 292 if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) { 293 status = VIRTIO_BLK_S_UNSUPP; 294 goto fail; 295 } 296 297 /* 298 * No support for bidirection commands yet. 299 */ 300 if (elem->out_num > 2 && elem->in_num > 3) { 301 status = VIRTIO_BLK_S_UNSUPP; 302 goto fail; 303 } 304 305 #ifdef __linux__ 306 ioctl_req = g_new0(VirtIOBlockIoctlReq, 1); 307 ioctl_req->req = req; 308 ioctl_req->hdr.interface_id = 'S'; 309 ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len; 310 ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base; 311 ioctl_req->hdr.dxfer_len = 0; 312 313 if (elem->out_num > 2) { 314 /* 315 * If there are more than the minimally required 2 output segments 316 * there is write payload starting from the third iovec. 317 */ 318 ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV; 319 ioctl_req->hdr.iovec_count = elem->out_num - 2; 320 321 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { 322 ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len; 323 } 324 325 ioctl_req->hdr.dxferp = elem->out_sg + 2; 326 327 } else if (elem->in_num > 3) { 328 /* 329 * If we have more than 3 input segments the guest wants to actually 330 * read data. 331 */ 332 ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV; 333 ioctl_req->hdr.iovec_count = elem->in_num - 3; 334 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { 335 ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len; 336 } 337 338 ioctl_req->hdr.dxferp = elem->in_sg; 339 } else { 340 /* 341 * Some SCSI commands don't actually transfer any data. 342 */ 343 ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE; 344 } 345 346 ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base; 347 ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len; 348 349 acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr, 350 virtio_blk_ioctl_complete, ioctl_req); 351 if (!acb) { 352 g_free(ioctl_req); 353 status = VIRTIO_BLK_S_UNSUPP; 354 goto fail; 355 } 356 return -EINPROGRESS; 357 #else 358 abort(); 359 #endif 360 361 fail: 362 /* Just put anything nonzero so that the ioctl fails in the guest. */ 363 if (scsi) { 364 virtio_stl_p(vdev, &scsi->errors, 255); 365 } 366 return status; 367 } 368 369 static void virtio_blk_handle_scsi(VirtIOBlockReq *req) 370 { 371 int status; 372 373 status = virtio_blk_handle_scsi_req(req); 374 if (status != -EINPROGRESS) { 375 virtio_blk_req_complete(req, status); 376 virtio_blk_free_request(req); 377 } 378 } 379 380 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb, 381 int start, int num_reqs, int niov) 382 { 383 QEMUIOVector *qiov = &mrb->reqs[start]->qiov; 384 int64_t sector_num = mrb->reqs[start]->sector_num; 385 bool is_write = mrb->is_write; 386 387 if (num_reqs > 1) { 388 int i; 389 struct iovec *tmp_iov = qiov->iov; 390 int tmp_niov = qiov->niov; 391 392 /* mrb->reqs[start]->qiov was initialized from external so we can't 393 * modify it here. We need to initialize it locally and then add the 394 * external iovecs. */ 395 qemu_iovec_init(qiov, niov); 396 397 for (i = 0; i < tmp_niov; i++) { 398 qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len); 399 } 400 401 for (i = start + 1; i < start + num_reqs; i++) { 402 qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0, 403 mrb->reqs[i]->qiov.size); 404 mrb->reqs[i - 1]->mr_next = mrb->reqs[i]; 405 } 406 407 trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev), 408 mrb, start, num_reqs, 409 sector_num << BDRV_SECTOR_BITS, 410 qiov->size, is_write); 411 block_acct_merge_done(blk_get_stats(blk), 412 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ, 413 num_reqs - 1); 414 } 415 416 if (is_write) { 417 blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0, 418 virtio_blk_rw_complete, mrb->reqs[start]); 419 } else { 420 blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0, 421 virtio_blk_rw_complete, mrb->reqs[start]); 422 } 423 } 424 425 static int multireq_compare(const void *a, const void *b) 426 { 427 const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a, 428 *req2 = *(VirtIOBlockReq **)b; 429 430 /* 431 * Note that we can't simply subtract sector_num1 from sector_num2 432 * here as that could overflow the return value. 433 */ 434 if (req1->sector_num > req2->sector_num) { 435 return 1; 436 } else if (req1->sector_num < req2->sector_num) { 437 return -1; 438 } else { 439 return 0; 440 } 441 } 442 443 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb) 444 { 445 int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0; 446 uint32_t max_transfer; 447 int64_t sector_num = 0; 448 449 if (mrb->num_reqs == 1) { 450 submit_requests(blk, mrb, 0, 1, -1); 451 mrb->num_reqs = 0; 452 return; 453 } 454 455 max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk); 456 457 qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs), 458 &multireq_compare); 459 460 for (i = 0; i < mrb->num_reqs; i++) { 461 VirtIOBlockReq *req = mrb->reqs[i]; 462 if (num_reqs > 0) { 463 /* 464 * NOTE: We cannot merge the requests in below situations: 465 * 1. requests are not sequential 466 * 2. merge would exceed maximum number of IOVs 467 * 3. merge would exceed maximum transfer length of backend device 468 */ 469 if (sector_num + nb_sectors != req->sector_num || 470 niov > blk_get_max_iov(blk) - req->qiov.niov || 471 req->qiov.size > max_transfer || 472 nb_sectors > (max_transfer - 473 req->qiov.size) / BDRV_SECTOR_SIZE) { 474 submit_requests(blk, mrb, start, num_reqs, niov); 475 num_reqs = 0; 476 } 477 } 478 479 if (num_reqs == 0) { 480 sector_num = req->sector_num; 481 nb_sectors = niov = 0; 482 start = i; 483 } 484 485 nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE; 486 niov += req->qiov.niov; 487 num_reqs++; 488 } 489 490 submit_requests(blk, mrb, start, num_reqs, niov); 491 mrb->num_reqs = 0; 492 } 493 494 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) 495 { 496 VirtIOBlock *s = req->dev; 497 498 block_acct_start(blk_get_stats(s->blk), &req->acct, 0, 499 BLOCK_ACCT_FLUSH); 500 501 /* 502 * Make sure all outstanding writes are posted to the backing device. 503 */ 504 if (mrb->is_write && mrb->num_reqs > 0) { 505 virtio_blk_submit_multireq(s->blk, mrb); 506 } 507 blk_aio_flush(s->blk, virtio_blk_flush_complete, req); 508 } 509 510 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev, 511 uint64_t sector, size_t size) 512 { 513 uint64_t nb_sectors = size >> BDRV_SECTOR_BITS; 514 uint64_t total_sectors; 515 516 if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) { 517 return false; 518 } 519 if (sector & dev->sector_mask) { 520 return false; 521 } 522 if (size % dev->conf.conf.logical_block_size) { 523 return false; 524 } 525 blk_get_geometry(dev->blk, &total_sectors); 526 if (sector > total_sectors || nb_sectors > total_sectors - sector) { 527 return false; 528 } 529 return true; 530 } 531 532 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req, 533 struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes) 534 { 535 VirtIOBlock *s = req->dev; 536 VirtIODevice *vdev = VIRTIO_DEVICE(s); 537 uint64_t sector; 538 uint32_t num_sectors, flags, max_sectors; 539 uint8_t err_status; 540 int bytes; 541 542 sector = virtio_ldq_p(vdev, &dwz_hdr->sector); 543 num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors); 544 flags = virtio_ldl_p(vdev, &dwz_hdr->flags); 545 max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors : 546 s->conf.max_discard_sectors; 547 548 /* 549 * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check 550 * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in 551 * the integer variable. 552 */ 553 if (unlikely(num_sectors > max_sectors)) { 554 err_status = VIRTIO_BLK_S_IOERR; 555 goto err; 556 } 557 558 bytes = num_sectors << BDRV_SECTOR_BITS; 559 560 if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) { 561 err_status = VIRTIO_BLK_S_IOERR; 562 goto err; 563 } 564 565 /* 566 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard 567 * and write zeroes commands if any unknown flag is set. 568 */ 569 if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { 570 err_status = VIRTIO_BLK_S_UNSUPP; 571 goto err; 572 } 573 574 if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */ 575 int blk_aio_flags = 0; 576 577 if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { 578 blk_aio_flags |= BDRV_REQ_MAY_UNMAP; 579 } 580 581 block_acct_start(blk_get_stats(s->blk), &req->acct, bytes, 582 BLOCK_ACCT_WRITE); 583 584 blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS, 585 bytes, blk_aio_flags, 586 virtio_blk_discard_write_zeroes_complete, req); 587 } else { /* VIRTIO_BLK_T_DISCARD */ 588 /* 589 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for 590 * discard commands if the unmap flag is set. 591 */ 592 if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { 593 err_status = VIRTIO_BLK_S_UNSUPP; 594 goto err; 595 } 596 597 blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes, 598 virtio_blk_discard_write_zeroes_complete, req); 599 } 600 601 return VIRTIO_BLK_S_OK; 602 603 err: 604 if (is_write_zeroes) { 605 block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE); 606 } 607 return err_status; 608 } 609 610 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) 611 { 612 uint32_t type; 613 struct iovec *in_iov = req->elem.in_sg; 614 struct iovec *out_iov = req->elem.out_sg; 615 unsigned in_num = req->elem.in_num; 616 unsigned out_num = req->elem.out_num; 617 VirtIOBlock *s = req->dev; 618 VirtIODevice *vdev = VIRTIO_DEVICE(s); 619 620 if (req->elem.out_num < 1 || req->elem.in_num < 1) { 621 virtio_error(vdev, "virtio-blk missing headers"); 622 return -1; 623 } 624 625 if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out, 626 sizeof(req->out)) != sizeof(req->out))) { 627 virtio_error(vdev, "virtio-blk request outhdr too short"); 628 return -1; 629 } 630 631 iov_discard_front(&out_iov, &out_num, sizeof(req->out)); 632 633 if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { 634 virtio_error(vdev, "virtio-blk request inhdr too short"); 635 return -1; 636 } 637 638 /* We always touch the last byte, so just see how big in_iov is. */ 639 req->in_len = iov_size(in_iov, in_num); 640 req->in = (void *)in_iov[in_num - 1].iov_base 641 + in_iov[in_num - 1].iov_len 642 - sizeof(struct virtio_blk_inhdr); 643 iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr)); 644 645 type = virtio_ldl_p(vdev, &req->out.type); 646 647 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER 648 * is an optional flag. Although a guest should not send this flag if 649 * not negotiated we ignored it in the past. So keep ignoring it. */ 650 switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) { 651 case VIRTIO_BLK_T_IN: 652 { 653 bool is_write = type & VIRTIO_BLK_T_OUT; 654 req->sector_num = virtio_ldq_p(vdev, &req->out.sector); 655 656 if (is_write) { 657 qemu_iovec_init_external(&req->qiov, out_iov, out_num); 658 trace_virtio_blk_handle_write(vdev, req, req->sector_num, 659 req->qiov.size / BDRV_SECTOR_SIZE); 660 } else { 661 qemu_iovec_init_external(&req->qiov, in_iov, in_num); 662 trace_virtio_blk_handle_read(vdev, req, req->sector_num, 663 req->qiov.size / BDRV_SECTOR_SIZE); 664 } 665 666 if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) { 667 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); 668 block_acct_invalid(blk_get_stats(s->blk), 669 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); 670 virtio_blk_free_request(req); 671 return 0; 672 } 673 674 block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size, 675 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); 676 677 /* merge would exceed maximum number of requests or IO direction 678 * changes */ 679 if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS || 680 is_write != mrb->is_write || 681 !s->conf.request_merging)) { 682 virtio_blk_submit_multireq(s->blk, mrb); 683 } 684 685 assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS); 686 mrb->reqs[mrb->num_reqs++] = req; 687 mrb->is_write = is_write; 688 break; 689 } 690 case VIRTIO_BLK_T_FLUSH: 691 virtio_blk_handle_flush(req, mrb); 692 break; 693 case VIRTIO_BLK_T_SCSI_CMD: 694 virtio_blk_handle_scsi(req); 695 break; 696 case VIRTIO_BLK_T_GET_ID: 697 { 698 /* 699 * NB: per existing s/n string convention the string is 700 * terminated by '\0' only when shorter than buffer. 701 */ 702 const char *serial = s->conf.serial ? s->conf.serial : ""; 703 size_t size = MIN(strlen(serial) + 1, 704 MIN(iov_size(in_iov, in_num), 705 VIRTIO_BLK_ID_BYTES)); 706 iov_from_buf(in_iov, in_num, 0, serial, size); 707 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 708 virtio_blk_free_request(req); 709 break; 710 } 711 /* 712 * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with 713 * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement, 714 * so we must mask it for these requests, then we will check if it is set. 715 */ 716 case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT: 717 case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT: 718 { 719 struct virtio_blk_discard_write_zeroes dwz_hdr; 720 size_t out_len = iov_size(out_iov, out_num); 721 bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) == 722 VIRTIO_BLK_T_WRITE_ZEROES; 723 uint8_t err_status; 724 725 /* 726 * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains 727 * more than one segment. 728 */ 729 if (unlikely(!(type & VIRTIO_BLK_T_OUT) || 730 out_len > sizeof(dwz_hdr))) { 731 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); 732 virtio_blk_free_request(req); 733 return 0; 734 } 735 736 if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr, 737 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) { 738 virtio_error(vdev, "virtio-blk discard/write_zeroes header" 739 " too short"); 740 return -1; 741 } 742 743 err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr, 744 is_write_zeroes); 745 if (err_status != VIRTIO_BLK_S_OK) { 746 virtio_blk_req_complete(req, err_status); 747 virtio_blk_free_request(req); 748 } 749 750 break; 751 } 752 default: 753 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); 754 virtio_blk_free_request(req); 755 } 756 return 0; 757 } 758 759 bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq) 760 { 761 VirtIOBlockReq *req; 762 MultiReqBuffer mrb = {}; 763 bool progress = false; 764 765 aio_context_acquire(blk_get_aio_context(s->blk)); 766 blk_io_plug(s->blk); 767 768 do { 769 virtio_queue_set_notification(vq, 0); 770 771 while ((req = virtio_blk_get_request(s, vq))) { 772 progress = true; 773 if (virtio_blk_handle_request(req, &mrb)) { 774 virtqueue_detach_element(req->vq, &req->elem, 0); 775 virtio_blk_free_request(req); 776 break; 777 } 778 } 779 780 virtio_queue_set_notification(vq, 1); 781 } while (!virtio_queue_empty(vq)); 782 783 if (mrb.num_reqs) { 784 virtio_blk_submit_multireq(s->blk, &mrb); 785 } 786 787 blk_io_unplug(s->blk); 788 aio_context_release(blk_get_aio_context(s->blk)); 789 return progress; 790 } 791 792 static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq) 793 { 794 virtio_blk_handle_vq(s, vq); 795 } 796 797 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) 798 { 799 VirtIOBlock *s = (VirtIOBlock *)vdev; 800 801 if (s->dataplane) { 802 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start 803 * dataplane here instead of waiting for .set_status(). 804 */ 805 virtio_device_start_ioeventfd(vdev); 806 if (!s->dataplane_disabled) { 807 return; 808 } 809 } 810 virtio_blk_handle_output_do(s, vq); 811 } 812 813 static void virtio_blk_dma_restart_bh(void *opaque) 814 { 815 VirtIOBlock *s = opaque; 816 VirtIOBlockReq *req = s->rq; 817 MultiReqBuffer mrb = {}; 818 819 qemu_bh_delete(s->bh); 820 s->bh = NULL; 821 822 s->rq = NULL; 823 824 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); 825 while (req) { 826 VirtIOBlockReq *next = req->next; 827 if (virtio_blk_handle_request(req, &mrb)) { 828 /* Device is now broken and won't do any processing until it gets 829 * reset. Already queued requests will be lost: let's purge them. 830 */ 831 while (req) { 832 next = req->next; 833 virtqueue_detach_element(req->vq, &req->elem, 0); 834 virtio_blk_free_request(req); 835 req = next; 836 } 837 break; 838 } 839 req = next; 840 } 841 842 if (mrb.num_reqs) { 843 virtio_blk_submit_multireq(s->blk, &mrb); 844 } 845 blk_dec_in_flight(s->conf.conf.blk); 846 aio_context_release(blk_get_aio_context(s->conf.conf.blk)); 847 } 848 849 static void virtio_blk_dma_restart_cb(void *opaque, int running, 850 RunState state) 851 { 852 VirtIOBlock *s = opaque; 853 854 if (!running) { 855 return; 856 } 857 858 if (!s->bh) { 859 /* FIXME The data plane is not started yet, so these requests are 860 * processed in the main thread. */ 861 s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk), 862 virtio_blk_dma_restart_bh, s); 863 blk_inc_in_flight(s->conf.conf.blk); 864 qemu_bh_schedule(s->bh); 865 } 866 } 867 868 static void virtio_blk_reset(VirtIODevice *vdev) 869 { 870 VirtIOBlock *s = VIRTIO_BLK(vdev); 871 AioContext *ctx; 872 VirtIOBlockReq *req; 873 874 ctx = blk_get_aio_context(s->blk); 875 aio_context_acquire(ctx); 876 blk_drain(s->blk); 877 878 /* We drop queued requests after blk_drain() because blk_drain() itself can 879 * produce them. */ 880 while (s->rq) { 881 req = s->rq; 882 s->rq = req->next; 883 virtqueue_detach_element(req->vq, &req->elem, 0); 884 virtio_blk_free_request(req); 885 } 886 887 aio_context_release(ctx); 888 889 assert(!s->dataplane_started); 890 blk_set_enable_write_cache(s->blk, s->original_wce); 891 } 892 893 /* coalesce internal state, copy to pci i/o region 0 894 */ 895 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) 896 { 897 VirtIOBlock *s = VIRTIO_BLK(vdev); 898 BlockConf *conf = &s->conf.conf; 899 struct virtio_blk_config blkcfg; 900 uint64_t capacity; 901 int64_t length; 902 int blk_size = conf->logical_block_size; 903 904 blk_get_geometry(s->blk, &capacity); 905 memset(&blkcfg, 0, sizeof(blkcfg)); 906 virtio_stq_p(vdev, &blkcfg.capacity, capacity); 907 virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2); 908 virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls); 909 virtio_stl_p(vdev, &blkcfg.blk_size, blk_size); 910 virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size); 911 virtio_stw_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size); 912 blkcfg.geometry.heads = conf->heads; 913 /* 914 * We must ensure that the block device capacity is a multiple of 915 * the logical block size. If that is not the case, let's use 916 * sector_mask to adopt the geometry to have a correct picture. 917 * For those devices where the capacity is ok for the given geometry 918 * we don't touch the sector value of the geometry, since some devices 919 * (like s390 dasd) need a specific value. Here the capacity is already 920 * cyls*heads*secs*blk_size and the sector value is not block size 921 * divided by 512 - instead it is the amount of blk_size blocks 922 * per track (cylinder). 923 */ 924 length = blk_getlength(s->blk); 925 if (length > 0 && length / conf->heads / conf->secs % blk_size) { 926 blkcfg.geometry.sectors = conf->secs & ~s->sector_mask; 927 } else { 928 blkcfg.geometry.sectors = conf->secs; 929 } 930 blkcfg.size_max = 0; 931 blkcfg.physical_block_exp = get_physical_block_exp(conf); 932 blkcfg.alignment_offset = 0; 933 blkcfg.wce = blk_enable_write_cache(s->blk); 934 virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues); 935 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) { 936 virtio_stl_p(vdev, &blkcfg.max_discard_sectors, 937 s->conf.max_discard_sectors); 938 virtio_stl_p(vdev, &blkcfg.discard_sector_alignment, 939 blk_size >> BDRV_SECTOR_BITS); 940 /* 941 * We support only one segment per request since multiple segments 942 * are not widely used and there are no userspace APIs that allow 943 * applications to submit multiple segments in a single call. 944 */ 945 virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1); 946 } 947 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) { 948 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors, 949 s->conf.max_write_zeroes_sectors); 950 blkcfg.write_zeroes_may_unmap = 1; 951 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1); 952 } 953 memcpy(config, &blkcfg, s->config_size); 954 } 955 956 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config) 957 { 958 VirtIOBlock *s = VIRTIO_BLK(vdev); 959 struct virtio_blk_config blkcfg; 960 961 memcpy(&blkcfg, config, s->config_size); 962 963 aio_context_acquire(blk_get_aio_context(s->blk)); 964 blk_set_enable_write_cache(s->blk, blkcfg.wce != 0); 965 aio_context_release(blk_get_aio_context(s->blk)); 966 } 967 968 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features, 969 Error **errp) 970 { 971 VirtIOBlock *s = VIRTIO_BLK(vdev); 972 973 /* Firstly sync all virtio-blk possible supported features */ 974 features |= s->host_features; 975 976 virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX); 977 virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY); 978 virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY); 979 virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE); 980 if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) { 981 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) { 982 error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0"); 983 return 0; 984 } 985 } else { 986 virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT); 987 virtio_add_feature(&features, VIRTIO_BLK_F_SCSI); 988 } 989 990 if (blk_enable_write_cache(s->blk)) { 991 virtio_add_feature(&features, VIRTIO_BLK_F_WCE); 992 } 993 if (blk_is_read_only(s->blk)) { 994 virtio_add_feature(&features, VIRTIO_BLK_F_RO); 995 } 996 if (s->conf.num_queues > 1) { 997 virtio_add_feature(&features, VIRTIO_BLK_F_MQ); 998 } 999 1000 return features; 1001 } 1002 1003 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status) 1004 { 1005 VirtIOBlock *s = VIRTIO_BLK(vdev); 1006 1007 if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) { 1008 assert(!s->dataplane_started); 1009 } 1010 1011 if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1012 return; 1013 } 1014 1015 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send 1016 * cache flushes. Thus, the "auto writethrough" behavior is never 1017 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature. 1018 * Leaving it enabled would break the following sequence: 1019 * 1020 * Guest started with "-drive cache=writethrough" 1021 * Guest sets status to 0 1022 * Guest sets DRIVER bit in status field 1023 * Guest reads host features (WCE=0, CONFIG_WCE=1) 1024 * Guest writes guest features (WCE=0, CONFIG_WCE=1) 1025 * Guest writes 1 to the WCE configuration field (writeback mode) 1026 * Guest sets DRIVER_OK bit in status field 1027 * 1028 * s->blk would erroneously be placed in writethrough mode. 1029 */ 1030 if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) { 1031 aio_context_acquire(blk_get_aio_context(s->blk)); 1032 blk_set_enable_write_cache(s->blk, 1033 virtio_vdev_has_feature(vdev, 1034 VIRTIO_BLK_F_WCE)); 1035 aio_context_release(blk_get_aio_context(s->blk)); 1036 } 1037 } 1038 1039 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f) 1040 { 1041 VirtIOBlock *s = VIRTIO_BLK(vdev); 1042 VirtIOBlockReq *req = s->rq; 1043 1044 while (req) { 1045 qemu_put_sbyte(f, 1); 1046 1047 if (s->conf.num_queues > 1) { 1048 qemu_put_be32(f, virtio_get_queue_index(req->vq)); 1049 } 1050 1051 qemu_put_virtqueue_element(f, &req->elem); 1052 req = req->next; 1053 } 1054 qemu_put_sbyte(f, 0); 1055 } 1056 1057 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f, 1058 int version_id) 1059 { 1060 VirtIOBlock *s = VIRTIO_BLK(vdev); 1061 1062 while (qemu_get_sbyte(f)) { 1063 unsigned nvqs = s->conf.num_queues; 1064 unsigned vq_idx = 0; 1065 VirtIOBlockReq *req; 1066 1067 if (nvqs > 1) { 1068 vq_idx = qemu_get_be32(f); 1069 1070 if (vq_idx >= nvqs) { 1071 error_report("Invalid virtqueue index in request list: %#x", 1072 vq_idx); 1073 return -EINVAL; 1074 } 1075 } 1076 1077 req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq)); 1078 virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req); 1079 req->next = s->rq; 1080 s->rq = req; 1081 } 1082 1083 return 0; 1084 } 1085 1086 static void virtio_blk_resize(void *opaque) 1087 { 1088 VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 1089 1090 virtio_notify_config(vdev); 1091 } 1092 1093 static const BlockDevOps virtio_block_ops = { 1094 .resize_cb = virtio_blk_resize, 1095 }; 1096 1097 static void virtio_blk_device_realize(DeviceState *dev, Error **errp) 1098 { 1099 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1100 VirtIOBlock *s = VIRTIO_BLK(dev); 1101 VirtIOBlkConf *conf = &s->conf; 1102 Error *err = NULL; 1103 unsigned i; 1104 1105 if (!conf->conf.blk) { 1106 error_setg(errp, "drive property not set"); 1107 return; 1108 } 1109 if (!blk_is_inserted(conf->conf.blk)) { 1110 error_setg(errp, "Device needs media, but drive is empty"); 1111 return; 1112 } 1113 if (!conf->num_queues) { 1114 error_setg(errp, "num-queues property must be larger than 0"); 1115 return; 1116 } 1117 if (!is_power_of_2(conf->queue_size) || 1118 conf->queue_size > VIRTQUEUE_MAX_SIZE) { 1119 error_setg(errp, "invalid queue-size property (%" PRIu16 "), " 1120 "must be a power of 2 (max %d)", 1121 conf->queue_size, VIRTQUEUE_MAX_SIZE); 1122 return; 1123 } 1124 1125 if (!blkconf_apply_backend_options(&conf->conf, 1126 blk_is_read_only(conf->conf.blk), true, 1127 errp)) { 1128 return; 1129 } 1130 s->original_wce = blk_enable_write_cache(conf->conf.blk); 1131 if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) { 1132 return; 1133 } 1134 1135 blkconf_blocksizes(&conf->conf); 1136 1137 if (conf->conf.logical_block_size > 1138 conf->conf.physical_block_size) { 1139 error_setg(errp, 1140 "logical_block_size > physical_block_size not supported"); 1141 return; 1142 } 1143 1144 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) && 1145 (!conf->max_discard_sectors || 1146 conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) { 1147 error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")" 1148 ", must be between 1 and %d", 1149 conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS); 1150 return; 1151 } 1152 1153 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) && 1154 (!conf->max_write_zeroes_sectors || 1155 conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) { 1156 error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32 1157 "), must be between 1 and %d", 1158 conf->max_write_zeroes_sectors, 1159 (int)BDRV_REQUEST_MAX_SECTORS); 1160 return; 1161 } 1162 1163 virtio_blk_set_config_size(s, s->host_features); 1164 1165 virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, s->config_size); 1166 1167 s->blk = conf->conf.blk; 1168 s->rq = NULL; 1169 s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1; 1170 1171 for (i = 0; i < conf->num_queues; i++) { 1172 virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output); 1173 } 1174 virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err); 1175 if (err != NULL) { 1176 error_propagate(errp, err); 1177 virtio_cleanup(vdev); 1178 return; 1179 } 1180 1181 s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); 1182 blk_set_dev_ops(s->blk, &virtio_block_ops, s); 1183 blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size); 1184 1185 blk_iostatus_enable(s->blk); 1186 } 1187 1188 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp) 1189 { 1190 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1191 VirtIOBlock *s = VIRTIO_BLK(dev); 1192 1193 virtio_blk_data_plane_destroy(s->dataplane); 1194 s->dataplane = NULL; 1195 qemu_del_vm_change_state_handler(s->change); 1196 blockdev_mark_auto_del(s->blk); 1197 virtio_cleanup(vdev); 1198 } 1199 1200 static void virtio_blk_instance_init(Object *obj) 1201 { 1202 VirtIOBlock *s = VIRTIO_BLK(obj); 1203 1204 device_add_bootindex_property(obj, &s->conf.conf.bootindex, 1205 "bootindex", "/disk@0,0", 1206 DEVICE(obj), NULL); 1207 } 1208 1209 static const VMStateDescription vmstate_virtio_blk = { 1210 .name = "virtio-blk", 1211 .minimum_version_id = 2, 1212 .version_id = 2, 1213 .fields = (VMStateField[]) { 1214 VMSTATE_VIRTIO_DEVICE, 1215 VMSTATE_END_OF_LIST() 1216 }, 1217 }; 1218 1219 static Property virtio_blk_properties[] = { 1220 DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf), 1221 DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf), 1222 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf), 1223 DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial), 1224 DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features, 1225 VIRTIO_BLK_F_CONFIG_WCE, true), 1226 #ifdef __linux__ 1227 DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features, 1228 VIRTIO_BLK_F_SCSI, false), 1229 #endif 1230 DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0, 1231 true), 1232 DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1), 1233 DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 128), 1234 DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD, 1235 IOThread *), 1236 DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features, 1237 VIRTIO_BLK_F_DISCARD, true), 1238 DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features, 1239 VIRTIO_BLK_F_WRITE_ZEROES, true), 1240 DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock, 1241 conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS), 1242 DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock, 1243 conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS), 1244 DEFINE_PROP_END_OF_LIST(), 1245 }; 1246 1247 static void virtio_blk_class_init(ObjectClass *klass, void *data) 1248 { 1249 DeviceClass *dc = DEVICE_CLASS(klass); 1250 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 1251 1252 dc->props = virtio_blk_properties; 1253 dc->vmsd = &vmstate_virtio_blk; 1254 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); 1255 vdc->realize = virtio_blk_device_realize; 1256 vdc->unrealize = virtio_blk_device_unrealize; 1257 vdc->get_config = virtio_blk_update_config; 1258 vdc->set_config = virtio_blk_set_config; 1259 vdc->get_features = virtio_blk_get_features; 1260 vdc->set_status = virtio_blk_set_status; 1261 vdc->reset = virtio_blk_reset; 1262 vdc->save = virtio_blk_save_device; 1263 vdc->load = virtio_blk_load_device; 1264 vdc->start_ioeventfd = virtio_blk_data_plane_start; 1265 vdc->stop_ioeventfd = virtio_blk_data_plane_stop; 1266 } 1267 1268 static const TypeInfo virtio_blk_info = { 1269 .name = TYPE_VIRTIO_BLK, 1270 .parent = TYPE_VIRTIO_DEVICE, 1271 .instance_size = sizeof(VirtIOBlock), 1272 .instance_init = virtio_blk_instance_init, 1273 .class_init = virtio_blk_class_init, 1274 }; 1275 1276 static void virtio_register_types(void) 1277 { 1278 type_register_static(&virtio_blk_info); 1279 } 1280 1281 type_init(virtio_register_types) 1282