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