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 "qemu/defer-call.h" 16 #include "qapi/error.h" 17 #include "qemu/iov.h" 18 #include "qemu/module.h" 19 #include "qemu/error-report.h" 20 #include "qemu/main-loop.h" 21 #include "block/block_int.h" 22 #include "trace.h" 23 #include "hw/block/block.h" 24 #include "hw/qdev-properties.h" 25 #include "sysemu/blockdev.h" 26 #include "sysemu/block-ram-registrar.h" 27 #include "sysemu/sysemu.h" 28 #include "sysemu/runstate.h" 29 #include "hw/virtio/virtio-blk.h" 30 #include "dataplane/virtio-blk.h" 31 #include "scsi/constants.h" 32 #ifdef __linux__ 33 # include <scsi/sg.h> 34 #endif 35 #include "hw/virtio/virtio-bus.h" 36 #include "migration/qemu-file-types.h" 37 #include "hw/virtio/virtio-access.h" 38 #include "hw/virtio/virtio-blk-common.h" 39 #include "qemu/coroutine.h" 40 41 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq, 42 VirtIOBlockReq *req) 43 { 44 req->dev = s; 45 req->vq = vq; 46 req->qiov.size = 0; 47 req->in_len = 0; 48 req->next = NULL; 49 req->mr_next = NULL; 50 } 51 52 static void virtio_blk_free_request(VirtIOBlockReq *req) 53 { 54 g_free(req); 55 } 56 57 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) 58 { 59 VirtIOBlock *s = req->dev; 60 VirtIODevice *vdev = VIRTIO_DEVICE(s); 61 62 trace_virtio_blk_req_complete(vdev, req, status); 63 64 stb_p(&req->in->status, status); 65 iov_discard_undo(&req->inhdr_undo); 66 iov_discard_undo(&req->outhdr_undo); 67 virtqueue_push(req->vq, &req->elem, req->in_len); 68 if (s->dataplane_started && !s->dataplane_disabled) { 69 virtio_blk_data_plane_notify(s->dataplane, req->vq); 70 } else { 71 virtio_notify(vdev, req->vq); 72 } 73 } 74 75 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, 76 bool is_read, bool acct_failed) 77 { 78 VirtIOBlock *s = req->dev; 79 BlockErrorAction action = blk_get_error_action(s->blk, is_read, error); 80 81 if (action == BLOCK_ERROR_ACTION_STOP) { 82 /* Break the link as the next request is going to be parsed from the 83 * ring again. Otherwise we may end up doing a double completion! */ 84 req->mr_next = NULL; 85 86 WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 87 req->next = s->rq; 88 s->rq = req; 89 } 90 } else if (action == BLOCK_ERROR_ACTION_REPORT) { 91 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); 92 if (acct_failed) { 93 block_acct_failed(blk_get_stats(s->blk), &req->acct); 94 } 95 virtio_blk_free_request(req); 96 } 97 98 blk_error_action(s->blk, action, is_read, error); 99 return action != BLOCK_ERROR_ACTION_IGNORE; 100 } 101 102 static void virtio_blk_rw_complete(void *opaque, int ret) 103 { 104 VirtIOBlockReq *next = opaque; 105 VirtIOBlock *s = next->dev; 106 VirtIODevice *vdev = VIRTIO_DEVICE(s); 107 108 while (next) { 109 VirtIOBlockReq *req = next; 110 next = req->mr_next; 111 trace_virtio_blk_rw_complete(vdev, req, ret); 112 113 if (req->qiov.nalloc != -1) { 114 /* If nalloc is != -1 req->qiov is a local copy of the original 115 * external iovec. It was allocated in submit_requests to be 116 * able to merge requests. */ 117 qemu_iovec_destroy(&req->qiov); 118 } 119 120 if (ret) { 121 int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type); 122 bool is_read = !(p & VIRTIO_BLK_T_OUT); 123 /* Note that memory may be dirtied on read failure. If the 124 * virtio request is not completed here, as is the case for 125 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied 126 * correctly during live migration. While this is ugly, 127 * it is acceptable because the device is free to write to 128 * the memory until the request is completed (which will 129 * happen on the other side of the migration). 130 */ 131 if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) { 132 continue; 133 } 134 } 135 136 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 137 block_acct_done(blk_get_stats(s->blk), &req->acct); 138 virtio_blk_free_request(req); 139 } 140 } 141 142 static void virtio_blk_flush_complete(void *opaque, int ret) 143 { 144 VirtIOBlockReq *req = opaque; 145 VirtIOBlock *s = req->dev; 146 147 if (ret && virtio_blk_handle_rw_error(req, -ret, 0, true)) { 148 return; 149 } 150 151 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 152 block_acct_done(blk_get_stats(s->blk), &req->acct); 153 virtio_blk_free_request(req); 154 } 155 156 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret) 157 { 158 VirtIOBlockReq *req = opaque; 159 VirtIOBlock *s = req->dev; 160 bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) & 161 ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES; 162 163 if (ret && virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) { 164 return; 165 } 166 167 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 168 if (is_write_zeroes) { 169 block_acct_done(blk_get_stats(s->blk), &req->acct); 170 } 171 virtio_blk_free_request(req); 172 } 173 174 #ifdef __linux__ 175 176 typedef struct { 177 VirtIOBlockReq *req; 178 struct sg_io_hdr hdr; 179 } VirtIOBlockIoctlReq; 180 181 static void virtio_blk_ioctl_complete(void *opaque, int status) 182 { 183 VirtIOBlockIoctlReq *ioctl_req = opaque; 184 VirtIOBlockReq *req = ioctl_req->req; 185 VirtIOBlock *s = req->dev; 186 VirtIODevice *vdev = VIRTIO_DEVICE(s); 187 struct virtio_scsi_inhdr *scsi; 188 struct sg_io_hdr *hdr; 189 190 scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; 191 192 if (status) { 193 status = VIRTIO_BLK_S_UNSUPP; 194 virtio_stl_p(vdev, &scsi->errors, 255); 195 goto out; 196 } 197 198 hdr = &ioctl_req->hdr; 199 /* 200 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) 201 * clear the masked_status field [hence status gets cleared too, see 202 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED 203 * status has occurred. However they do set DRIVER_SENSE in driver_status 204 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. 205 */ 206 if (hdr->status == 0 && hdr->sb_len_wr > 0) { 207 hdr->status = CHECK_CONDITION; 208 } 209 210 virtio_stl_p(vdev, &scsi->errors, 211 hdr->status | (hdr->msg_status << 8) | 212 (hdr->host_status << 16) | (hdr->driver_status << 24)); 213 virtio_stl_p(vdev, &scsi->residual, hdr->resid); 214 virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr); 215 virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len); 216 217 out: 218 virtio_blk_req_complete(req, status); 219 virtio_blk_free_request(req); 220 g_free(ioctl_req); 221 } 222 223 #endif 224 225 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq) 226 { 227 VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq)); 228 229 if (req) { 230 virtio_blk_init_request(s, vq, req); 231 } 232 return req; 233 } 234 235 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req) 236 { 237 int status = VIRTIO_BLK_S_OK; 238 struct virtio_scsi_inhdr *scsi = NULL; 239 VirtIOBlock *blk = req->dev; 240 VirtIODevice *vdev = VIRTIO_DEVICE(blk); 241 VirtQueueElement *elem = &req->elem; 242 243 #ifdef __linux__ 244 int i; 245 VirtIOBlockIoctlReq *ioctl_req; 246 BlockAIOCB *acb; 247 #endif 248 249 /* 250 * We require at least one output segment each for the virtio_blk_outhdr 251 * and the SCSI command block. 252 * 253 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr 254 * and the sense buffer pointer in the input segments. 255 */ 256 if (elem->out_num < 2 || elem->in_num < 3) { 257 status = VIRTIO_BLK_S_IOERR; 258 goto fail; 259 } 260 261 /* 262 * The scsi inhdr is placed in the second-to-last input segment, just 263 * before the regular inhdr. 264 */ 265 scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base; 266 267 if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) { 268 status = VIRTIO_BLK_S_UNSUPP; 269 goto fail; 270 } 271 272 /* 273 * No support for bidirection commands yet. 274 */ 275 if (elem->out_num > 2 && elem->in_num > 3) { 276 status = VIRTIO_BLK_S_UNSUPP; 277 goto fail; 278 } 279 280 #ifdef __linux__ 281 ioctl_req = g_new0(VirtIOBlockIoctlReq, 1); 282 ioctl_req->req = req; 283 ioctl_req->hdr.interface_id = 'S'; 284 ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len; 285 ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base; 286 ioctl_req->hdr.dxfer_len = 0; 287 288 if (elem->out_num > 2) { 289 /* 290 * If there are more than the minimally required 2 output segments 291 * there is write payload starting from the third iovec. 292 */ 293 ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV; 294 ioctl_req->hdr.iovec_count = elem->out_num - 2; 295 296 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { 297 ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len; 298 } 299 300 ioctl_req->hdr.dxferp = elem->out_sg + 2; 301 302 } else if (elem->in_num > 3) { 303 /* 304 * If we have more than 3 input segments the guest wants to actually 305 * read data. 306 */ 307 ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV; 308 ioctl_req->hdr.iovec_count = elem->in_num - 3; 309 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { 310 ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len; 311 } 312 313 ioctl_req->hdr.dxferp = elem->in_sg; 314 } else { 315 /* 316 * Some SCSI commands don't actually transfer any data. 317 */ 318 ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE; 319 } 320 321 ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base; 322 ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len; 323 324 acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr, 325 virtio_blk_ioctl_complete, ioctl_req); 326 if (!acb) { 327 g_free(ioctl_req); 328 status = VIRTIO_BLK_S_UNSUPP; 329 goto fail; 330 } 331 return -EINPROGRESS; 332 #else 333 abort(); 334 #endif 335 336 fail: 337 /* Just put anything nonzero so that the ioctl fails in the guest. */ 338 if (scsi) { 339 virtio_stl_p(vdev, &scsi->errors, 255); 340 } 341 return status; 342 } 343 344 static void virtio_blk_handle_scsi(VirtIOBlockReq *req) 345 { 346 int status; 347 348 status = virtio_blk_handle_scsi_req(req); 349 if (status != -EINPROGRESS) { 350 virtio_blk_req_complete(req, status); 351 virtio_blk_free_request(req); 352 } 353 } 354 355 static inline void submit_requests(VirtIOBlock *s, MultiReqBuffer *mrb, 356 int start, int num_reqs, int niov) 357 { 358 BlockBackend *blk = s->blk; 359 QEMUIOVector *qiov = &mrb->reqs[start]->qiov; 360 int64_t sector_num = mrb->reqs[start]->sector_num; 361 bool is_write = mrb->is_write; 362 BdrvRequestFlags flags = 0; 363 364 if (num_reqs > 1) { 365 int i; 366 struct iovec *tmp_iov = qiov->iov; 367 int tmp_niov = qiov->niov; 368 369 /* mrb->reqs[start]->qiov was initialized from external so we can't 370 * modify it here. We need to initialize it locally and then add the 371 * external iovecs. */ 372 qemu_iovec_init(qiov, niov); 373 374 for (i = 0; i < tmp_niov; i++) { 375 qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len); 376 } 377 378 for (i = start + 1; i < start + num_reqs; i++) { 379 qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0, 380 mrb->reqs[i]->qiov.size); 381 mrb->reqs[i - 1]->mr_next = mrb->reqs[i]; 382 } 383 384 trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev), 385 mrb, start, num_reqs, 386 sector_num << BDRV_SECTOR_BITS, 387 qiov->size, is_write); 388 block_acct_merge_done(blk_get_stats(blk), 389 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ, 390 num_reqs - 1); 391 } 392 393 if (blk_ram_registrar_ok(&s->blk_ram_registrar)) { 394 flags |= BDRV_REQ_REGISTERED_BUF; 395 } 396 397 if (is_write) { 398 blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 399 flags, virtio_blk_rw_complete, 400 mrb->reqs[start]); 401 } else { 402 blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 403 flags, virtio_blk_rw_complete, 404 mrb->reqs[start]); 405 } 406 } 407 408 static int multireq_compare(const void *a, const void *b) 409 { 410 const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a, 411 *req2 = *(VirtIOBlockReq **)b; 412 413 /* 414 * Note that we can't simply subtract sector_num1 from sector_num2 415 * here as that could overflow the return value. 416 */ 417 if (req1->sector_num > req2->sector_num) { 418 return 1; 419 } else if (req1->sector_num < req2->sector_num) { 420 return -1; 421 } else { 422 return 0; 423 } 424 } 425 426 static void virtio_blk_submit_multireq(VirtIOBlock *s, MultiReqBuffer *mrb) 427 { 428 int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0; 429 uint32_t max_transfer; 430 int64_t sector_num = 0; 431 432 if (mrb->num_reqs == 1) { 433 submit_requests(s, mrb, 0, 1, -1); 434 mrb->num_reqs = 0; 435 return; 436 } 437 438 max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk); 439 440 qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs), 441 &multireq_compare); 442 443 for (i = 0; i < mrb->num_reqs; i++) { 444 VirtIOBlockReq *req = mrb->reqs[i]; 445 if (num_reqs > 0) { 446 /* 447 * NOTE: We cannot merge the requests in below situations: 448 * 1. requests are not sequential 449 * 2. merge would exceed maximum number of IOVs 450 * 3. merge would exceed maximum transfer length of backend device 451 */ 452 if (sector_num + nb_sectors != req->sector_num || 453 niov > blk_get_max_iov(s->blk) - req->qiov.niov || 454 req->qiov.size > max_transfer || 455 nb_sectors > (max_transfer - 456 req->qiov.size) / BDRV_SECTOR_SIZE) { 457 submit_requests(s, mrb, start, num_reqs, niov); 458 num_reqs = 0; 459 } 460 } 461 462 if (num_reqs == 0) { 463 sector_num = req->sector_num; 464 nb_sectors = niov = 0; 465 start = i; 466 } 467 468 nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE; 469 niov += req->qiov.niov; 470 num_reqs++; 471 } 472 473 submit_requests(s, mrb, start, num_reqs, niov); 474 mrb->num_reqs = 0; 475 } 476 477 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) 478 { 479 VirtIOBlock *s = req->dev; 480 481 block_acct_start(blk_get_stats(s->blk), &req->acct, 0, 482 BLOCK_ACCT_FLUSH); 483 484 /* 485 * Make sure all outstanding writes are posted to the backing device. 486 */ 487 if (mrb->is_write && mrb->num_reqs > 0) { 488 virtio_blk_submit_multireq(s, mrb); 489 } 490 blk_aio_flush(s->blk, virtio_blk_flush_complete, req); 491 } 492 493 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev, 494 uint64_t sector, size_t size) 495 { 496 uint64_t nb_sectors = size >> BDRV_SECTOR_BITS; 497 uint64_t total_sectors; 498 499 if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) { 500 return false; 501 } 502 if (sector & dev->sector_mask) { 503 return false; 504 } 505 if (size % dev->conf.conf.logical_block_size) { 506 return false; 507 } 508 blk_get_geometry(dev->blk, &total_sectors); 509 if (sector > total_sectors || nb_sectors > total_sectors - sector) { 510 return false; 511 } 512 return true; 513 } 514 515 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req, 516 struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes) 517 { 518 VirtIOBlock *s = req->dev; 519 VirtIODevice *vdev = VIRTIO_DEVICE(s); 520 uint64_t sector; 521 uint32_t num_sectors, flags, max_sectors; 522 uint8_t err_status; 523 int bytes; 524 525 sector = virtio_ldq_p(vdev, &dwz_hdr->sector); 526 num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors); 527 flags = virtio_ldl_p(vdev, &dwz_hdr->flags); 528 max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors : 529 s->conf.max_discard_sectors; 530 531 /* 532 * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check 533 * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in 534 * the integer variable. 535 */ 536 if (unlikely(num_sectors > max_sectors)) { 537 err_status = VIRTIO_BLK_S_IOERR; 538 goto err; 539 } 540 541 bytes = num_sectors << BDRV_SECTOR_BITS; 542 543 if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) { 544 err_status = VIRTIO_BLK_S_IOERR; 545 goto err; 546 } 547 548 /* 549 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard 550 * and write zeroes commands if any unknown flag is set. 551 */ 552 if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { 553 err_status = VIRTIO_BLK_S_UNSUPP; 554 goto err; 555 } 556 557 if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */ 558 int blk_aio_flags = 0; 559 560 if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { 561 blk_aio_flags |= BDRV_REQ_MAY_UNMAP; 562 } 563 564 block_acct_start(blk_get_stats(s->blk), &req->acct, bytes, 565 BLOCK_ACCT_WRITE); 566 567 blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS, 568 bytes, blk_aio_flags, 569 virtio_blk_discard_write_zeroes_complete, req); 570 } else { /* VIRTIO_BLK_T_DISCARD */ 571 /* 572 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for 573 * discard commands if the unmap flag is set. 574 */ 575 if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { 576 err_status = VIRTIO_BLK_S_UNSUPP; 577 goto err; 578 } 579 580 blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes, 581 virtio_blk_discard_write_zeroes_complete, req); 582 } 583 584 return VIRTIO_BLK_S_OK; 585 586 err: 587 if (is_write_zeroes) { 588 block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE); 589 } 590 return err_status; 591 } 592 593 typedef struct ZoneCmdData { 594 VirtIOBlockReq *req; 595 struct iovec *in_iov; 596 unsigned in_num; 597 union { 598 struct { 599 unsigned int nr_zones; 600 BlockZoneDescriptor *zones; 601 } zone_report_data; 602 struct { 603 int64_t offset; 604 } zone_append_data; 605 }; 606 } ZoneCmdData; 607 608 /* 609 * check zoned_request: error checking before issuing requests. If all checks 610 * passed, return true. 611 * append: true if only zone append requests issued. 612 */ 613 static bool check_zoned_request(VirtIOBlock *s, int64_t offset, int64_t len, 614 bool append, uint8_t *status) { 615 BlockDriverState *bs = blk_bs(s->blk); 616 int index; 617 618 if (!virtio_has_feature(s->host_features, VIRTIO_BLK_F_ZONED)) { 619 *status = VIRTIO_BLK_S_UNSUPP; 620 return false; 621 } 622 623 if (offset < 0 || len < 0 || len > (bs->total_sectors << BDRV_SECTOR_BITS) 624 || offset > (bs->total_sectors << BDRV_SECTOR_BITS) - len) { 625 *status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 626 return false; 627 } 628 629 if (append) { 630 if (bs->bl.write_granularity) { 631 if ((offset % bs->bl.write_granularity) != 0) { 632 *status = VIRTIO_BLK_S_ZONE_UNALIGNED_WP; 633 return false; 634 } 635 } 636 637 index = offset / bs->bl.zone_size; 638 if (BDRV_ZT_IS_CONV(bs->wps->wp[index])) { 639 *status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 640 return false; 641 } 642 643 if (len / 512 > bs->bl.max_append_sectors) { 644 if (bs->bl.max_append_sectors == 0) { 645 *status = VIRTIO_BLK_S_UNSUPP; 646 } else { 647 *status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 648 } 649 return false; 650 } 651 } 652 return true; 653 } 654 655 static void virtio_blk_zone_report_complete(void *opaque, int ret) 656 { 657 ZoneCmdData *data = opaque; 658 VirtIOBlockReq *req = data->req; 659 VirtIODevice *vdev = VIRTIO_DEVICE(req->dev); 660 struct iovec *in_iov = data->in_iov; 661 unsigned in_num = data->in_num; 662 int64_t zrp_size, n, j = 0; 663 int64_t nz = data->zone_report_data.nr_zones; 664 int8_t err_status = VIRTIO_BLK_S_OK; 665 666 trace_virtio_blk_zone_report_complete(vdev, req, nz, ret); 667 if (ret) { 668 err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 669 goto out; 670 } 671 672 struct virtio_blk_zone_report zrp_hdr = (struct virtio_blk_zone_report) { 673 .nr_zones = cpu_to_le64(nz), 674 }; 675 zrp_size = sizeof(struct virtio_blk_zone_report) 676 + sizeof(struct virtio_blk_zone_descriptor) * nz; 677 n = iov_from_buf(in_iov, in_num, 0, &zrp_hdr, sizeof(zrp_hdr)); 678 if (n != sizeof(zrp_hdr)) { 679 virtio_error(vdev, "Driver provided input buffer that is too small!"); 680 err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 681 goto out; 682 } 683 684 for (size_t i = sizeof(zrp_hdr); i < zrp_size; 685 i += sizeof(struct virtio_blk_zone_descriptor), ++j) { 686 struct virtio_blk_zone_descriptor desc = 687 (struct virtio_blk_zone_descriptor) { 688 .z_start = cpu_to_le64(data->zone_report_data.zones[j].start 689 >> BDRV_SECTOR_BITS), 690 .z_cap = cpu_to_le64(data->zone_report_data.zones[j].cap 691 >> BDRV_SECTOR_BITS), 692 .z_wp = cpu_to_le64(data->zone_report_data.zones[j].wp 693 >> BDRV_SECTOR_BITS), 694 }; 695 696 switch (data->zone_report_data.zones[j].type) { 697 case BLK_ZT_CONV: 698 desc.z_type = VIRTIO_BLK_ZT_CONV; 699 break; 700 case BLK_ZT_SWR: 701 desc.z_type = VIRTIO_BLK_ZT_SWR; 702 break; 703 case BLK_ZT_SWP: 704 desc.z_type = VIRTIO_BLK_ZT_SWP; 705 break; 706 default: 707 g_assert_not_reached(); 708 } 709 710 switch (data->zone_report_data.zones[j].state) { 711 case BLK_ZS_RDONLY: 712 desc.z_state = VIRTIO_BLK_ZS_RDONLY; 713 break; 714 case BLK_ZS_OFFLINE: 715 desc.z_state = VIRTIO_BLK_ZS_OFFLINE; 716 break; 717 case BLK_ZS_EMPTY: 718 desc.z_state = VIRTIO_BLK_ZS_EMPTY; 719 break; 720 case BLK_ZS_CLOSED: 721 desc.z_state = VIRTIO_BLK_ZS_CLOSED; 722 break; 723 case BLK_ZS_FULL: 724 desc.z_state = VIRTIO_BLK_ZS_FULL; 725 break; 726 case BLK_ZS_EOPEN: 727 desc.z_state = VIRTIO_BLK_ZS_EOPEN; 728 break; 729 case BLK_ZS_IOPEN: 730 desc.z_state = VIRTIO_BLK_ZS_IOPEN; 731 break; 732 case BLK_ZS_NOT_WP: 733 desc.z_state = VIRTIO_BLK_ZS_NOT_WP; 734 break; 735 default: 736 g_assert_not_reached(); 737 } 738 739 /* TODO: it takes O(n^2) time complexity. Optimizations required. */ 740 n = iov_from_buf(in_iov, in_num, i, &desc, sizeof(desc)); 741 if (n != sizeof(desc)) { 742 virtio_error(vdev, "Driver provided input buffer " 743 "for descriptors that is too small!"); 744 err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 745 } 746 } 747 748 out: 749 virtio_blk_req_complete(req, err_status); 750 virtio_blk_free_request(req); 751 g_free(data->zone_report_data.zones); 752 g_free(data); 753 } 754 755 static void virtio_blk_handle_zone_report(VirtIOBlockReq *req, 756 struct iovec *in_iov, 757 unsigned in_num) 758 { 759 VirtIOBlock *s = req->dev; 760 VirtIODevice *vdev = VIRTIO_DEVICE(s); 761 unsigned int nr_zones; 762 ZoneCmdData *data; 763 int64_t zone_size, offset; 764 uint8_t err_status; 765 766 if (req->in_len < sizeof(struct virtio_blk_inhdr) + 767 sizeof(struct virtio_blk_zone_report) + 768 sizeof(struct virtio_blk_zone_descriptor)) { 769 virtio_error(vdev, "in buffer too small for zone report"); 770 return; 771 } 772 773 /* start byte offset of the zone report */ 774 offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS; 775 if (!check_zoned_request(s, offset, 0, false, &err_status)) { 776 goto out; 777 } 778 nr_zones = (req->in_len - sizeof(struct virtio_blk_inhdr) - 779 sizeof(struct virtio_blk_zone_report)) / 780 sizeof(struct virtio_blk_zone_descriptor); 781 trace_virtio_blk_handle_zone_report(vdev, req, 782 offset >> BDRV_SECTOR_BITS, nr_zones); 783 784 zone_size = sizeof(BlockZoneDescriptor) * nr_zones; 785 data = g_malloc(sizeof(ZoneCmdData)); 786 data->req = req; 787 data->in_iov = in_iov; 788 data->in_num = in_num; 789 data->zone_report_data.nr_zones = nr_zones; 790 data->zone_report_data.zones = g_malloc(zone_size), 791 792 blk_aio_zone_report(s->blk, offset, &data->zone_report_data.nr_zones, 793 data->zone_report_data.zones, 794 virtio_blk_zone_report_complete, data); 795 return; 796 out: 797 virtio_blk_req_complete(req, err_status); 798 virtio_blk_free_request(req); 799 } 800 801 static void virtio_blk_zone_mgmt_complete(void *opaque, int ret) 802 { 803 VirtIOBlockReq *req = opaque; 804 VirtIOBlock *s = req->dev; 805 VirtIODevice *vdev = VIRTIO_DEVICE(s); 806 int8_t err_status = VIRTIO_BLK_S_OK; 807 trace_virtio_blk_zone_mgmt_complete(vdev, req,ret); 808 809 if (ret) { 810 err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 811 } 812 813 virtio_blk_req_complete(req, err_status); 814 virtio_blk_free_request(req); 815 } 816 817 static int virtio_blk_handle_zone_mgmt(VirtIOBlockReq *req, BlockZoneOp op) 818 { 819 VirtIOBlock *s = req->dev; 820 VirtIODevice *vdev = VIRTIO_DEVICE(s); 821 BlockDriverState *bs = blk_bs(s->blk); 822 int64_t offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS; 823 uint64_t len; 824 uint64_t capacity = bs->total_sectors << BDRV_SECTOR_BITS; 825 uint8_t err_status = VIRTIO_BLK_S_OK; 826 827 uint32_t type = virtio_ldl_p(vdev, &req->out.type); 828 if (type == VIRTIO_BLK_T_ZONE_RESET_ALL) { 829 /* Entire drive capacity */ 830 offset = 0; 831 len = capacity; 832 trace_virtio_blk_handle_zone_reset_all(vdev, req, 0, 833 bs->total_sectors); 834 } else { 835 if (bs->bl.zone_size > capacity - offset) { 836 /* The zoned device allows the last smaller zone. */ 837 len = capacity - bs->bl.zone_size * (bs->bl.nr_zones - 1); 838 } else { 839 len = bs->bl.zone_size; 840 } 841 trace_virtio_blk_handle_zone_mgmt(vdev, req, op, 842 offset >> BDRV_SECTOR_BITS, 843 len >> BDRV_SECTOR_BITS); 844 } 845 846 if (!check_zoned_request(s, offset, len, false, &err_status)) { 847 goto out; 848 } 849 850 blk_aio_zone_mgmt(s->blk, op, offset, len, 851 virtio_blk_zone_mgmt_complete, req); 852 853 return 0; 854 out: 855 virtio_blk_req_complete(req, err_status); 856 virtio_blk_free_request(req); 857 return err_status; 858 } 859 860 static void virtio_blk_zone_append_complete(void *opaque, int ret) 861 { 862 ZoneCmdData *data = opaque; 863 VirtIOBlockReq *req = data->req; 864 VirtIODevice *vdev = VIRTIO_DEVICE(req->dev); 865 int64_t append_sector, n; 866 uint8_t err_status = VIRTIO_BLK_S_OK; 867 868 if (ret) { 869 err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 870 goto out; 871 } 872 873 virtio_stq_p(vdev, &append_sector, 874 data->zone_append_data.offset >> BDRV_SECTOR_BITS); 875 n = iov_from_buf(data->in_iov, data->in_num, 0, &append_sector, 876 sizeof(append_sector)); 877 if (n != sizeof(append_sector)) { 878 virtio_error(vdev, "Driver provided input buffer less than size of " 879 "append_sector"); 880 err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD; 881 goto out; 882 } 883 trace_virtio_blk_zone_append_complete(vdev, req, append_sector, ret); 884 885 out: 886 virtio_blk_req_complete(req, err_status); 887 virtio_blk_free_request(req); 888 g_free(data); 889 } 890 891 static int virtio_blk_handle_zone_append(VirtIOBlockReq *req, 892 struct iovec *out_iov, 893 struct iovec *in_iov, 894 uint64_t out_num, 895 unsigned in_num) { 896 VirtIOBlock *s = req->dev; 897 VirtIODevice *vdev = VIRTIO_DEVICE(s); 898 uint8_t err_status = VIRTIO_BLK_S_OK; 899 900 int64_t offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS; 901 int64_t len = iov_size(out_iov, out_num); 902 903 trace_virtio_blk_handle_zone_append(vdev, req, offset >> BDRV_SECTOR_BITS); 904 if (!check_zoned_request(s, offset, len, true, &err_status)) { 905 goto out; 906 } 907 908 ZoneCmdData *data = g_malloc(sizeof(ZoneCmdData)); 909 data->req = req; 910 data->in_iov = in_iov; 911 data->in_num = in_num; 912 data->zone_append_data.offset = offset; 913 qemu_iovec_init_external(&req->qiov, out_iov, out_num); 914 915 block_acct_start(blk_get_stats(s->blk), &req->acct, len, 916 BLOCK_ACCT_ZONE_APPEND); 917 918 blk_aio_zone_append(s->blk, &data->zone_append_data.offset, &req->qiov, 0, 919 virtio_blk_zone_append_complete, data); 920 return 0; 921 922 out: 923 virtio_blk_req_complete(req, err_status); 924 virtio_blk_free_request(req); 925 return err_status; 926 } 927 928 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) 929 { 930 uint32_t type; 931 struct iovec *in_iov = req->elem.in_sg; 932 struct iovec *out_iov = req->elem.out_sg; 933 unsigned in_num = req->elem.in_num; 934 unsigned out_num = req->elem.out_num; 935 VirtIOBlock *s = req->dev; 936 VirtIODevice *vdev = VIRTIO_DEVICE(s); 937 938 if (req->elem.out_num < 1 || req->elem.in_num < 1) { 939 virtio_error(vdev, "virtio-blk missing headers"); 940 return -1; 941 } 942 943 if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out, 944 sizeof(req->out)) != sizeof(req->out))) { 945 virtio_error(vdev, "virtio-blk request outhdr too short"); 946 return -1; 947 } 948 949 iov_discard_front_undoable(&out_iov, &out_num, sizeof(req->out), 950 &req->outhdr_undo); 951 952 if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { 953 virtio_error(vdev, "virtio-blk request inhdr too short"); 954 iov_discard_undo(&req->outhdr_undo); 955 return -1; 956 } 957 958 /* We always touch the last byte, so just see how big in_iov is. */ 959 req->in_len = iov_size(in_iov, in_num); 960 req->in = (void *)in_iov[in_num - 1].iov_base 961 + in_iov[in_num - 1].iov_len 962 - sizeof(struct virtio_blk_inhdr); 963 iov_discard_back_undoable(in_iov, &in_num, sizeof(struct virtio_blk_inhdr), 964 &req->inhdr_undo); 965 966 type = virtio_ldl_p(vdev, &req->out.type); 967 968 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER 969 * is an optional flag. Although a guest should not send this flag if 970 * not negotiated we ignored it in the past. So keep ignoring it. */ 971 switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) { 972 case VIRTIO_BLK_T_IN: 973 { 974 bool is_write = type & VIRTIO_BLK_T_OUT; 975 req->sector_num = virtio_ldq_p(vdev, &req->out.sector); 976 977 if (is_write) { 978 qemu_iovec_init_external(&req->qiov, out_iov, out_num); 979 trace_virtio_blk_handle_write(vdev, req, req->sector_num, 980 req->qiov.size / BDRV_SECTOR_SIZE); 981 } else { 982 qemu_iovec_init_external(&req->qiov, in_iov, in_num); 983 trace_virtio_blk_handle_read(vdev, req, req->sector_num, 984 req->qiov.size / BDRV_SECTOR_SIZE); 985 } 986 987 if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) { 988 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); 989 block_acct_invalid(blk_get_stats(s->blk), 990 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); 991 virtio_blk_free_request(req); 992 return 0; 993 } 994 995 block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size, 996 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); 997 998 /* merge would exceed maximum number of requests or IO direction 999 * changes */ 1000 if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS || 1001 is_write != mrb->is_write || 1002 !s->conf.request_merging)) { 1003 virtio_blk_submit_multireq(s, mrb); 1004 } 1005 1006 assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS); 1007 mrb->reqs[mrb->num_reqs++] = req; 1008 mrb->is_write = is_write; 1009 break; 1010 } 1011 case VIRTIO_BLK_T_FLUSH: 1012 virtio_blk_handle_flush(req, mrb); 1013 break; 1014 case VIRTIO_BLK_T_ZONE_REPORT: 1015 virtio_blk_handle_zone_report(req, in_iov, in_num); 1016 break; 1017 case VIRTIO_BLK_T_ZONE_OPEN: 1018 virtio_blk_handle_zone_mgmt(req, BLK_ZO_OPEN); 1019 break; 1020 case VIRTIO_BLK_T_ZONE_CLOSE: 1021 virtio_blk_handle_zone_mgmt(req, BLK_ZO_CLOSE); 1022 break; 1023 case VIRTIO_BLK_T_ZONE_FINISH: 1024 virtio_blk_handle_zone_mgmt(req, BLK_ZO_FINISH); 1025 break; 1026 case VIRTIO_BLK_T_ZONE_RESET: 1027 virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET); 1028 break; 1029 case VIRTIO_BLK_T_ZONE_RESET_ALL: 1030 virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET); 1031 break; 1032 case VIRTIO_BLK_T_SCSI_CMD: 1033 virtio_blk_handle_scsi(req); 1034 break; 1035 case VIRTIO_BLK_T_GET_ID: 1036 { 1037 /* 1038 * NB: per existing s/n string convention the string is 1039 * terminated by '\0' only when shorter than buffer. 1040 */ 1041 const char *serial = s->conf.serial ? s->conf.serial : ""; 1042 size_t size = MIN(strlen(serial) + 1, 1043 MIN(iov_size(in_iov, in_num), 1044 VIRTIO_BLK_ID_BYTES)); 1045 iov_from_buf(in_iov, in_num, 0, serial, size); 1046 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); 1047 virtio_blk_free_request(req); 1048 break; 1049 } 1050 case VIRTIO_BLK_T_ZONE_APPEND & ~VIRTIO_BLK_T_OUT: 1051 /* 1052 * Passing out_iov/out_num and in_iov/in_num is not safe 1053 * to access req->elem.out_sg directly because it may be 1054 * modified by virtio_blk_handle_request(). 1055 */ 1056 virtio_blk_handle_zone_append(req, out_iov, in_iov, out_num, in_num); 1057 break; 1058 /* 1059 * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with 1060 * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement, 1061 * so we must mask it for these requests, then we will check if it is set. 1062 */ 1063 case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT: 1064 case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT: 1065 { 1066 struct virtio_blk_discard_write_zeroes dwz_hdr; 1067 size_t out_len = iov_size(out_iov, out_num); 1068 bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) == 1069 VIRTIO_BLK_T_WRITE_ZEROES; 1070 uint8_t err_status; 1071 1072 /* 1073 * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains 1074 * more than one segment. 1075 */ 1076 if (unlikely(!(type & VIRTIO_BLK_T_OUT) || 1077 out_len > sizeof(dwz_hdr))) { 1078 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); 1079 virtio_blk_free_request(req); 1080 return 0; 1081 } 1082 1083 if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr, 1084 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) { 1085 iov_discard_undo(&req->inhdr_undo); 1086 iov_discard_undo(&req->outhdr_undo); 1087 virtio_error(vdev, "virtio-blk discard/write_zeroes header" 1088 " too short"); 1089 return -1; 1090 } 1091 1092 err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr, 1093 is_write_zeroes); 1094 if (err_status != VIRTIO_BLK_S_OK) { 1095 virtio_blk_req_complete(req, err_status); 1096 virtio_blk_free_request(req); 1097 } 1098 1099 break; 1100 } 1101 default: 1102 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); 1103 virtio_blk_free_request(req); 1104 } 1105 return 0; 1106 } 1107 1108 void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq) 1109 { 1110 VirtIOBlockReq *req; 1111 MultiReqBuffer mrb = {}; 1112 bool suppress_notifications = virtio_queue_get_notification(vq); 1113 1114 defer_call_begin(); 1115 1116 do { 1117 if (suppress_notifications) { 1118 virtio_queue_set_notification(vq, 0); 1119 } 1120 1121 while ((req = virtio_blk_get_request(s, vq))) { 1122 if (virtio_blk_handle_request(req, &mrb)) { 1123 virtqueue_detach_element(req->vq, &req->elem, 0); 1124 virtio_blk_free_request(req); 1125 break; 1126 } 1127 } 1128 1129 if (suppress_notifications) { 1130 virtio_queue_set_notification(vq, 1); 1131 } 1132 } while (!virtio_queue_empty(vq)); 1133 1134 if (mrb.num_reqs) { 1135 virtio_blk_submit_multireq(s, &mrb); 1136 } 1137 1138 defer_call_end(); 1139 } 1140 1141 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) 1142 { 1143 VirtIOBlock *s = (VirtIOBlock *)vdev; 1144 1145 if (s->dataplane && !s->dataplane_started) { 1146 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start 1147 * dataplane here instead of waiting for .set_status(). 1148 */ 1149 virtio_device_start_ioeventfd(vdev); 1150 if (!s->dataplane_disabled) { 1151 return; 1152 } 1153 } 1154 virtio_blk_handle_vq(s, vq); 1155 } 1156 1157 static void virtio_blk_dma_restart_bh(void *opaque) 1158 { 1159 VirtIOBlock *s = opaque; 1160 1161 VirtIOBlockReq *req; 1162 MultiReqBuffer mrb = {}; 1163 1164 WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 1165 req = s->rq; 1166 s->rq = NULL; 1167 } 1168 1169 while (req) { 1170 VirtIOBlockReq *next = req->next; 1171 if (virtio_blk_handle_request(req, &mrb)) { 1172 /* Device is now broken and won't do any processing until it gets 1173 * reset. Already queued requests will be lost: let's purge them. 1174 */ 1175 while (req) { 1176 next = req->next; 1177 virtqueue_detach_element(req->vq, &req->elem, 0); 1178 virtio_blk_free_request(req); 1179 req = next; 1180 } 1181 break; 1182 } 1183 req = next; 1184 } 1185 1186 if (mrb.num_reqs) { 1187 virtio_blk_submit_multireq(s, &mrb); 1188 } 1189 1190 /* Paired with inc in virtio_blk_dma_restart_cb() */ 1191 blk_dec_in_flight(s->conf.conf.blk); 1192 } 1193 1194 static void virtio_blk_dma_restart_cb(void *opaque, bool running, 1195 RunState state) 1196 { 1197 VirtIOBlock *s = opaque; 1198 1199 if (!running) { 1200 return; 1201 } 1202 1203 /* Paired with dec in virtio_blk_dma_restart_bh() */ 1204 blk_inc_in_flight(s->conf.conf.blk); 1205 1206 aio_bh_schedule_oneshot(blk_get_aio_context(s->conf.conf.blk), 1207 virtio_blk_dma_restart_bh, s); 1208 } 1209 1210 static void virtio_blk_reset(VirtIODevice *vdev) 1211 { 1212 VirtIOBlock *s = VIRTIO_BLK(vdev); 1213 VirtIOBlockReq *req; 1214 1215 /* Dataplane has stopped... */ 1216 assert(!s->dataplane_started); 1217 1218 /* ...but requests may still be in flight. */ 1219 blk_drain(s->blk); 1220 1221 /* We drop queued requests after blk_drain() because blk_drain() itself can 1222 * produce them. */ 1223 WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 1224 while (s->rq) { 1225 req = s->rq; 1226 s->rq = req->next; 1227 1228 /* No other threads can access req->vq here */ 1229 virtqueue_detach_element(req->vq, &req->elem, 0); 1230 1231 virtio_blk_free_request(req); 1232 } 1233 } 1234 1235 blk_set_enable_write_cache(s->blk, s->original_wce); 1236 } 1237 1238 /* coalesce internal state, copy to pci i/o region 0 1239 */ 1240 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) 1241 { 1242 VirtIOBlock *s = VIRTIO_BLK(vdev); 1243 BlockConf *conf = &s->conf.conf; 1244 BlockDriverState *bs = blk_bs(s->blk); 1245 struct virtio_blk_config blkcfg; 1246 uint64_t capacity; 1247 int64_t length; 1248 int blk_size = conf->logical_block_size; 1249 1250 blk_get_geometry(s->blk, &capacity); 1251 memset(&blkcfg, 0, sizeof(blkcfg)); 1252 virtio_stq_p(vdev, &blkcfg.capacity, capacity); 1253 virtio_stl_p(vdev, &blkcfg.seg_max, 1254 s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2); 1255 virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls); 1256 virtio_stl_p(vdev, &blkcfg.blk_size, blk_size); 1257 virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size); 1258 virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size); 1259 blkcfg.geometry.heads = conf->heads; 1260 /* 1261 * We must ensure that the block device capacity is a multiple of 1262 * the logical block size. If that is not the case, let's use 1263 * sector_mask to adopt the geometry to have a correct picture. 1264 * For those devices where the capacity is ok for the given geometry 1265 * we don't touch the sector value of the geometry, since some devices 1266 * (like s390 dasd) need a specific value. Here the capacity is already 1267 * cyls*heads*secs*blk_size and the sector value is not block size 1268 * divided by 512 - instead it is the amount of blk_size blocks 1269 * per track (cylinder). 1270 */ 1271 length = blk_getlength(s->blk); 1272 if (length > 0 && length / conf->heads / conf->secs % blk_size) { 1273 blkcfg.geometry.sectors = conf->secs & ~s->sector_mask; 1274 } else { 1275 blkcfg.geometry.sectors = conf->secs; 1276 } 1277 blkcfg.size_max = 0; 1278 blkcfg.physical_block_exp = get_physical_block_exp(conf); 1279 blkcfg.alignment_offset = 0; 1280 blkcfg.wce = blk_enable_write_cache(s->blk); 1281 virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues); 1282 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) { 1283 uint32_t discard_granularity = conf->discard_granularity; 1284 if (discard_granularity == -1 || !s->conf.report_discard_granularity) { 1285 discard_granularity = blk_size; 1286 } 1287 virtio_stl_p(vdev, &blkcfg.max_discard_sectors, 1288 s->conf.max_discard_sectors); 1289 virtio_stl_p(vdev, &blkcfg.discard_sector_alignment, 1290 discard_granularity >> BDRV_SECTOR_BITS); 1291 /* 1292 * We support only one segment per request since multiple segments 1293 * are not widely used and there are no userspace APIs that allow 1294 * applications to submit multiple segments in a single call. 1295 */ 1296 virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1); 1297 } 1298 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) { 1299 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors, 1300 s->conf.max_write_zeroes_sectors); 1301 blkcfg.write_zeroes_may_unmap = 1; 1302 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1); 1303 } 1304 if (bs->bl.zoned != BLK_Z_NONE) { 1305 switch (bs->bl.zoned) { 1306 case BLK_Z_HM: 1307 blkcfg.zoned.model = VIRTIO_BLK_Z_HM; 1308 break; 1309 case BLK_Z_HA: 1310 blkcfg.zoned.model = VIRTIO_BLK_Z_HA; 1311 break; 1312 default: 1313 g_assert_not_reached(); 1314 } 1315 1316 virtio_stl_p(vdev, &blkcfg.zoned.zone_sectors, 1317 bs->bl.zone_size / 512); 1318 virtio_stl_p(vdev, &blkcfg.zoned.max_active_zones, 1319 bs->bl.max_active_zones); 1320 virtio_stl_p(vdev, &blkcfg.zoned.max_open_zones, 1321 bs->bl.max_open_zones); 1322 virtio_stl_p(vdev, &blkcfg.zoned.write_granularity, blk_size); 1323 virtio_stl_p(vdev, &blkcfg.zoned.max_append_sectors, 1324 bs->bl.max_append_sectors); 1325 } else { 1326 blkcfg.zoned.model = VIRTIO_BLK_Z_NONE; 1327 } 1328 memcpy(config, &blkcfg, s->config_size); 1329 } 1330 1331 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config) 1332 { 1333 VirtIOBlock *s = VIRTIO_BLK(vdev); 1334 struct virtio_blk_config blkcfg; 1335 1336 memcpy(&blkcfg, config, s->config_size); 1337 1338 blk_set_enable_write_cache(s->blk, blkcfg.wce != 0); 1339 } 1340 1341 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features, 1342 Error **errp) 1343 { 1344 VirtIOBlock *s = VIRTIO_BLK(vdev); 1345 1346 /* Firstly sync all virtio-blk possible supported features */ 1347 features |= s->host_features; 1348 1349 virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX); 1350 virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY); 1351 virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY); 1352 virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE); 1353 if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) { 1354 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) { 1355 error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0"); 1356 return 0; 1357 } 1358 } else { 1359 virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT); 1360 virtio_add_feature(&features, VIRTIO_BLK_F_SCSI); 1361 } 1362 1363 if (blk_enable_write_cache(s->blk) || 1364 (s->conf.x_enable_wce_if_config_wce && 1365 virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) { 1366 virtio_add_feature(&features, VIRTIO_BLK_F_WCE); 1367 } 1368 if (!blk_is_writable(s->blk)) { 1369 virtio_add_feature(&features, VIRTIO_BLK_F_RO); 1370 } 1371 if (s->conf.num_queues > 1) { 1372 virtio_add_feature(&features, VIRTIO_BLK_F_MQ); 1373 } 1374 1375 return features; 1376 } 1377 1378 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status) 1379 { 1380 VirtIOBlock *s = VIRTIO_BLK(vdev); 1381 1382 if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) { 1383 assert(!s->dataplane_started); 1384 } 1385 1386 if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1387 return; 1388 } 1389 1390 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send 1391 * cache flushes. Thus, the "auto writethrough" behavior is never 1392 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature. 1393 * Leaving it enabled would break the following sequence: 1394 * 1395 * Guest started with "-drive cache=writethrough" 1396 * Guest sets status to 0 1397 * Guest sets DRIVER bit in status field 1398 * Guest reads host features (WCE=0, CONFIG_WCE=1) 1399 * Guest writes guest features (WCE=0, CONFIG_WCE=1) 1400 * Guest writes 1 to the WCE configuration field (writeback mode) 1401 * Guest sets DRIVER_OK bit in status field 1402 * 1403 * s->blk would erroneously be placed in writethrough mode. 1404 */ 1405 if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) { 1406 blk_set_enable_write_cache(s->blk, 1407 virtio_vdev_has_feature(vdev, 1408 VIRTIO_BLK_F_WCE)); 1409 } 1410 } 1411 1412 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f) 1413 { 1414 VirtIOBlock *s = VIRTIO_BLK(vdev); 1415 1416 WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 1417 VirtIOBlockReq *req = s->rq; 1418 1419 while (req) { 1420 qemu_put_sbyte(f, 1); 1421 1422 if (s->conf.num_queues > 1) { 1423 qemu_put_be32(f, virtio_get_queue_index(req->vq)); 1424 } 1425 1426 qemu_put_virtqueue_element(vdev, f, &req->elem); 1427 req = req->next; 1428 } 1429 } 1430 1431 qemu_put_sbyte(f, 0); 1432 } 1433 1434 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f, 1435 int version_id) 1436 { 1437 VirtIOBlock *s = VIRTIO_BLK(vdev); 1438 1439 while (qemu_get_sbyte(f)) { 1440 unsigned nvqs = s->conf.num_queues; 1441 unsigned vq_idx = 0; 1442 VirtIOBlockReq *req; 1443 1444 if (nvqs > 1) { 1445 vq_idx = qemu_get_be32(f); 1446 1447 if (vq_idx >= nvqs) { 1448 error_report("Invalid virtqueue index in request list: %#x", 1449 vq_idx); 1450 return -EINVAL; 1451 } 1452 } 1453 1454 req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq)); 1455 virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req); 1456 1457 WITH_QEMU_LOCK_GUARD(&s->rq_lock) { 1458 req->next = s->rq; 1459 s->rq = req; 1460 } 1461 } 1462 1463 return 0; 1464 } 1465 1466 static void virtio_resize_cb(void *opaque) 1467 { 1468 VirtIODevice *vdev = opaque; 1469 1470 assert(qemu_get_current_aio_context() == qemu_get_aio_context()); 1471 virtio_notify_config(vdev); 1472 } 1473 1474 static void virtio_blk_resize(void *opaque) 1475 { 1476 VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 1477 1478 /* 1479 * virtio_notify_config() needs to acquire the global mutex, 1480 * so it can't be called from an iothread. Instead, schedule 1481 * it to be run in the main context BH. 1482 */ 1483 aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev); 1484 } 1485 1486 /* Suspend virtqueue ioeventfd processing during drain */ 1487 static void virtio_blk_drained_begin(void *opaque) 1488 { 1489 VirtIOBlock *s = opaque; 1490 VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 1491 AioContext *ctx = blk_get_aio_context(s->conf.conf.blk); 1492 1493 if (!s->dataplane || !s->dataplane_started) { 1494 return; 1495 } 1496 1497 for (uint16_t i = 0; i < s->conf.num_queues; i++) { 1498 VirtQueue *vq = virtio_get_queue(vdev, i); 1499 virtio_queue_aio_detach_host_notifier(vq, ctx); 1500 } 1501 } 1502 1503 /* Resume virtqueue ioeventfd processing after drain */ 1504 static void virtio_blk_drained_end(void *opaque) 1505 { 1506 VirtIOBlock *s = opaque; 1507 VirtIODevice *vdev = VIRTIO_DEVICE(opaque); 1508 AioContext *ctx = blk_get_aio_context(s->conf.conf.blk); 1509 1510 if (!s->dataplane || !s->dataplane_started) { 1511 return; 1512 } 1513 1514 for (uint16_t i = 0; i < s->conf.num_queues; i++) { 1515 VirtQueue *vq = virtio_get_queue(vdev, i); 1516 virtio_queue_aio_attach_host_notifier(vq, ctx); 1517 } 1518 } 1519 1520 static const BlockDevOps virtio_block_ops = { 1521 .resize_cb = virtio_blk_resize, 1522 .drained_begin = virtio_blk_drained_begin, 1523 .drained_end = virtio_blk_drained_end, 1524 }; 1525 1526 static void virtio_blk_device_realize(DeviceState *dev, Error **errp) 1527 { 1528 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1529 VirtIOBlock *s = VIRTIO_BLK(dev); 1530 VirtIOBlkConf *conf = &s->conf; 1531 Error *err = NULL; 1532 unsigned i; 1533 1534 if (!conf->conf.blk) { 1535 error_setg(errp, "drive property not set"); 1536 return; 1537 } 1538 if (!blk_is_inserted(conf->conf.blk)) { 1539 error_setg(errp, "Device needs media, but drive is empty"); 1540 return; 1541 } 1542 if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) { 1543 conf->num_queues = 1; 1544 } 1545 if (!conf->num_queues) { 1546 error_setg(errp, "num-queues property must be larger than 0"); 1547 return; 1548 } 1549 if (conf->queue_size <= 2) { 1550 error_setg(errp, "invalid queue-size property (%" PRIu16 "), " 1551 "must be > 2", conf->queue_size); 1552 return; 1553 } 1554 if (!is_power_of_2(conf->queue_size) || 1555 conf->queue_size > VIRTQUEUE_MAX_SIZE) { 1556 error_setg(errp, "invalid queue-size property (%" PRIu16 "), " 1557 "must be a power of 2 (max %d)", 1558 conf->queue_size, VIRTQUEUE_MAX_SIZE); 1559 return; 1560 } 1561 1562 if (!blkconf_apply_backend_options(&conf->conf, 1563 !blk_supports_write_perm(conf->conf.blk), 1564 true, errp)) { 1565 return; 1566 } 1567 s->original_wce = blk_enable_write_cache(conf->conf.blk); 1568 if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) { 1569 return; 1570 } 1571 1572 if (!blkconf_blocksizes(&conf->conf, errp)) { 1573 return; 1574 } 1575 1576 BlockDriverState *bs = blk_bs(conf->conf.blk); 1577 if (bs->bl.zoned != BLK_Z_NONE) { 1578 virtio_add_feature(&s->host_features, VIRTIO_BLK_F_ZONED); 1579 if (bs->bl.zoned == BLK_Z_HM) { 1580 virtio_clear_feature(&s->host_features, VIRTIO_BLK_F_DISCARD); 1581 } 1582 } 1583 1584 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) && 1585 (!conf->max_discard_sectors || 1586 conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) { 1587 error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")" 1588 ", must be between 1 and %d", 1589 conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS); 1590 return; 1591 } 1592 1593 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) && 1594 (!conf->max_write_zeroes_sectors || 1595 conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) { 1596 error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32 1597 "), must be between 1 and %d", 1598 conf->max_write_zeroes_sectors, 1599 (int)BDRV_REQUEST_MAX_SECTORS); 1600 return; 1601 } 1602 1603 s->config_size = virtio_get_config_size(&virtio_blk_cfg_size_params, 1604 s->host_features); 1605 virtio_init(vdev, VIRTIO_ID_BLOCK, s->config_size); 1606 1607 qemu_mutex_init(&s->rq_lock); 1608 1609 s->blk = conf->conf.blk; 1610 s->rq = NULL; 1611 s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1; 1612 1613 for (i = 0; i < conf->num_queues; i++) { 1614 virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output); 1615 } 1616 qemu_coroutine_inc_pool_size(conf->num_queues * conf->queue_size / 2); 1617 virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err); 1618 if (err != NULL) { 1619 error_propagate(errp, err); 1620 for (i = 0; i < conf->num_queues; i++) { 1621 virtio_del_queue(vdev, i); 1622 } 1623 virtio_cleanup(vdev); 1624 return; 1625 } 1626 1627 /* 1628 * This must be after virtio_init() so virtio_blk_dma_restart_cb() gets 1629 * called after ->start_ioeventfd() has already set blk's AioContext. 1630 */ 1631 s->change = 1632 qdev_add_vm_change_state_handler(dev, virtio_blk_dma_restart_cb, s); 1633 1634 blk_ram_registrar_init(&s->blk_ram_registrar, s->blk); 1635 blk_set_dev_ops(s->blk, &virtio_block_ops, s); 1636 1637 blk_iostatus_enable(s->blk); 1638 1639 add_boot_device_lchs(dev, "/disk@0,0", 1640 conf->conf.lcyls, 1641 conf->conf.lheads, 1642 conf->conf.lsecs); 1643 } 1644 1645 static void virtio_blk_device_unrealize(DeviceState *dev) 1646 { 1647 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1648 VirtIOBlock *s = VIRTIO_BLK(dev); 1649 VirtIOBlkConf *conf = &s->conf; 1650 unsigned i; 1651 1652 blk_drain(s->blk); 1653 del_boot_device_lchs(dev, "/disk@0,0"); 1654 virtio_blk_data_plane_destroy(s->dataplane); 1655 s->dataplane = NULL; 1656 for (i = 0; i < conf->num_queues; i++) { 1657 virtio_del_queue(vdev, i); 1658 } 1659 qemu_coroutine_dec_pool_size(conf->num_queues * conf->queue_size / 2); 1660 qemu_mutex_destroy(&s->rq_lock); 1661 blk_ram_registrar_destroy(&s->blk_ram_registrar); 1662 qemu_del_vm_change_state_handler(s->change); 1663 blockdev_mark_auto_del(s->blk); 1664 virtio_cleanup(vdev); 1665 } 1666 1667 static void virtio_blk_instance_init(Object *obj) 1668 { 1669 VirtIOBlock *s = VIRTIO_BLK(obj); 1670 1671 device_add_bootindex_property(obj, &s->conf.conf.bootindex, 1672 "bootindex", "/disk@0,0", 1673 DEVICE(obj)); 1674 } 1675 1676 static const VMStateDescription vmstate_virtio_blk = { 1677 .name = "virtio-blk", 1678 .minimum_version_id = 2, 1679 .version_id = 2, 1680 .fields = (VMStateField[]) { 1681 VMSTATE_VIRTIO_DEVICE, 1682 VMSTATE_END_OF_LIST() 1683 }, 1684 }; 1685 1686 static Property virtio_blk_properties[] = { 1687 DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf), 1688 DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf), 1689 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf), 1690 DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial), 1691 DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features, 1692 VIRTIO_BLK_F_CONFIG_WCE, true), 1693 #ifdef __linux__ 1694 DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features, 1695 VIRTIO_BLK_F_SCSI, false), 1696 #endif 1697 DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0, 1698 true), 1699 DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1700 VIRTIO_BLK_AUTO_NUM_QUEUES), 1701 DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256), 1702 DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true), 1703 DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD, 1704 IOThread *), 1705 DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features, 1706 VIRTIO_BLK_F_DISCARD, true), 1707 DEFINE_PROP_BOOL("report-discard-granularity", VirtIOBlock, 1708 conf.report_discard_granularity, true), 1709 DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features, 1710 VIRTIO_BLK_F_WRITE_ZEROES, true), 1711 DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock, 1712 conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS), 1713 DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock, 1714 conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS), 1715 DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock, 1716 conf.x_enable_wce_if_config_wce, true), 1717 DEFINE_PROP_END_OF_LIST(), 1718 }; 1719 1720 static void virtio_blk_class_init(ObjectClass *klass, void *data) 1721 { 1722 DeviceClass *dc = DEVICE_CLASS(klass); 1723 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 1724 1725 device_class_set_props(dc, virtio_blk_properties); 1726 dc->vmsd = &vmstate_virtio_blk; 1727 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); 1728 vdc->realize = virtio_blk_device_realize; 1729 vdc->unrealize = virtio_blk_device_unrealize; 1730 vdc->get_config = virtio_blk_update_config; 1731 vdc->set_config = virtio_blk_set_config; 1732 vdc->get_features = virtio_blk_get_features; 1733 vdc->set_status = virtio_blk_set_status; 1734 vdc->reset = virtio_blk_reset; 1735 vdc->save = virtio_blk_save_device; 1736 vdc->load = virtio_blk_load_device; 1737 vdc->start_ioeventfd = virtio_blk_data_plane_start; 1738 vdc->stop_ioeventfd = virtio_blk_data_plane_stop; 1739 } 1740 1741 static const TypeInfo virtio_blk_info = { 1742 .name = TYPE_VIRTIO_BLK, 1743 .parent = TYPE_VIRTIO_DEVICE, 1744 .instance_size = sizeof(VirtIOBlock), 1745 .instance_init = virtio_blk_instance_init, 1746 .class_init = virtio_blk_class_init, 1747 }; 1748 1749 static void virtio_register_types(void) 1750 { 1751 type_register_static(&virtio_blk_info); 1752 } 1753 1754 type_init(virtio_register_types) 1755