4 * Copyright IBM, Corp. 2007
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu-common.h"
18 #include "qemu/error-report.h"
20 #include "hw/block/block.h"
21 #include "sysemu/block-backend.h"
22 #include "sysemu/blockdev.h"
23 #include "hw/virtio/virtio-blk.h"
24 #include "dataplane/virtio-blk.h"
25 #include "block/scsi.h"
29 #include "hw/virtio/virtio-bus.h"
30 #include "hw/virtio/virtio-access.h"
32 void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
43 void virtio_blk_free_request(VirtIOBlockReq *req)
50 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
52 VirtIOBlock *s = req->dev;
53 VirtIODevice *vdev = VIRTIO_DEVICE(s);
55 trace_virtio_blk_req_complete(req, status);
57 stb_p(&req->in->status, status);
58 virtqueue_push(req->vq, &req->elem, req->in_len);
59 if (s->dataplane_started && !s->dataplane_disabled) {
60 virtio_blk_data_plane_notify(s->dataplane, req->vq);
62 virtio_notify(vdev, req->vq);
66 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
69 BlockErrorAction action = blk_get_error_action(req->dev->blk,
71 VirtIOBlock *s = req->dev;
73 if (action == BLOCK_ERROR_ACTION_STOP) {
74 /* Break the link as the next request is going to be parsed from the
75 * ring again. Otherwise we may end up doing a double completion! */
79 } else if (action == BLOCK_ERROR_ACTION_REPORT) {
80 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
81 block_acct_failed(blk_get_stats(s->blk), &req->acct);
82 virtio_blk_free_request(req);
85 blk_error_action(s->blk, action, is_read, error);
86 return action != BLOCK_ERROR_ACTION_IGNORE;
89 static void virtio_blk_rw_complete(void *opaque, int ret)
91 VirtIOBlockReq *next = opaque;
94 VirtIOBlockReq *req = next;
96 trace_virtio_blk_rw_complete(req, ret);
98 if (req->qiov.nalloc != -1) {
99 /* If nalloc is != 1 req->qiov is a local copy of the original
100 * external iovec. It was allocated in submit_merged_requests
101 * to be able to merge requests. */
102 qemu_iovec_destroy(&req->qiov);
106 int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
107 bool is_read = !(p & VIRTIO_BLK_T_OUT);
108 /* Note that memory may be dirtied on read failure. If the
109 * virtio request is not completed here, as is the case for
110 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
111 * correctly during live migration. While this is ugly,
112 * it is acceptable because the device is free to write to
113 * the memory until the request is completed (which will
114 * happen on the other side of the migration).
116 if (virtio_blk_handle_rw_error(req, -ret, is_read)) {
121 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
122 block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
123 virtio_blk_free_request(req);
127 static void virtio_blk_flush_complete(void *opaque, int ret)
129 VirtIOBlockReq *req = opaque;
132 if (virtio_blk_handle_rw_error(req, -ret, 0)) {
137 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
138 block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
139 virtio_blk_free_request(req);
146 struct sg_io_hdr hdr;
147 } VirtIOBlockIoctlReq;
149 static void virtio_blk_ioctl_complete(void *opaque, int status)
151 VirtIOBlockIoctlReq *ioctl_req = opaque;
152 VirtIOBlockReq *req = ioctl_req->req;
153 VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
154 struct virtio_scsi_inhdr *scsi;
155 struct sg_io_hdr *hdr;
157 scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
160 status = VIRTIO_BLK_S_UNSUPP;
161 virtio_stl_p(vdev, &scsi->errors, 255);
165 hdr = &ioctl_req->hdr;
167 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
168 * clear the masked_status field [hence status gets cleared too, see
169 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
170 * status has occurred. However they do set DRIVER_SENSE in driver_status
171 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
173 if (hdr->status == 0 && hdr->sb_len_wr > 0) {
174 hdr->status = CHECK_CONDITION;
177 virtio_stl_p(vdev, &scsi->errors,
178 hdr->status | (hdr->msg_status << 8) |
179 (hdr->host_status << 16) | (hdr->driver_status << 24));
180 virtio_stl_p(vdev, &scsi->residual, hdr->resid);
181 virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
182 virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
185 virtio_blk_req_complete(req, status);
186 virtio_blk_free_request(req);
192 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
194 VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
197 virtio_blk_init_request(s, vq, req);
202 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
204 int status = VIRTIO_BLK_S_OK;
205 struct virtio_scsi_inhdr *scsi = NULL;
206 VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
207 VirtQueueElement *elem = &req->elem;
208 VirtIOBlock *blk = req->dev;
212 VirtIOBlockIoctlReq *ioctl_req;
217 * We require at least one output segment each for the virtio_blk_outhdr
218 * and the SCSI command block.
220 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
221 * and the sense buffer pointer in the input segments.
223 if (elem->out_num < 2 || elem->in_num < 3) {
224 status = VIRTIO_BLK_S_IOERR;
229 * The scsi inhdr is placed in the second-to-last input segment, just
230 * before the regular inhdr.
232 scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
234 if (!blk->conf.scsi) {
235 status = VIRTIO_BLK_S_UNSUPP;
240 * No support for bidirection commands yet.
242 if (elem->out_num > 2 && elem->in_num > 3) {
243 status = VIRTIO_BLK_S_UNSUPP;
248 ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
249 ioctl_req->req = req;
250 ioctl_req->hdr.interface_id = 'S';
251 ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
252 ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
253 ioctl_req->hdr.dxfer_len = 0;
255 if (elem->out_num > 2) {
257 * If there are more than the minimally required 2 output segments
258 * there is write payload starting from the third iovec.
260 ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
261 ioctl_req->hdr.iovec_count = elem->out_num - 2;
263 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
264 ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
267 ioctl_req->hdr.dxferp = elem->out_sg + 2;
269 } else if (elem->in_num > 3) {
271 * If we have more than 3 input segments the guest wants to actually
274 ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
275 ioctl_req->hdr.iovec_count = elem->in_num - 3;
276 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
277 ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
280 ioctl_req->hdr.dxferp = elem->in_sg;
283 * Some SCSI commands don't actually transfer any data.
285 ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
288 ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
289 ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
291 acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
292 virtio_blk_ioctl_complete, ioctl_req);
295 status = VIRTIO_BLK_S_UNSUPP;
304 /* Just put anything nonzero so that the ioctl fails in the guest. */
306 virtio_stl_p(vdev, &scsi->errors, 255);
311 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
315 status = virtio_blk_handle_scsi_req(req);
316 if (status != -EINPROGRESS) {
317 virtio_blk_req_complete(req, status);
318 virtio_blk_free_request(req);
322 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
323 int start, int num_reqs, int niov)
325 QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
326 int64_t sector_num = mrb->reqs[start]->sector_num;
327 bool is_write = mrb->is_write;
331 struct iovec *tmp_iov = qiov->iov;
332 int tmp_niov = qiov->niov;
334 /* mrb->reqs[start]->qiov was initialized from external so we can't
335 * modify it here. We need to initialize it locally and then add the
336 * external iovecs. */
337 qemu_iovec_init(qiov, niov);
339 for (i = 0; i < tmp_niov; i++) {
340 qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
343 for (i = start + 1; i < start + num_reqs; i++) {
344 qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
345 mrb->reqs[i]->qiov.size);
346 mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
349 trace_virtio_blk_submit_multireq(mrb, start, num_reqs,
350 sector_num << BDRV_SECTOR_BITS,
351 qiov->size, is_write);
352 block_acct_merge_done(blk_get_stats(blk),
353 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
358 blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
359 virtio_blk_rw_complete, mrb->reqs[start]);
361 blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
362 virtio_blk_rw_complete, mrb->reqs[start]);
366 static int multireq_compare(const void *a, const void *b)
368 const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
369 *req2 = *(VirtIOBlockReq **)b;
372 * Note that we can't simply subtract sector_num1 from sector_num2
373 * here as that could overflow the return value.
375 if (req1->sector_num > req2->sector_num) {
377 } else if (req1->sector_num < req2->sector_num) {
384 void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
386 int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
387 int max_xfer_len = 0;
388 int64_t sector_num = 0;
390 if (mrb->num_reqs == 1) {
391 submit_requests(blk, mrb, 0, 1, -1);
396 max_xfer_len = blk_get_max_transfer_length(mrb->reqs[0]->dev->blk);
397 max_xfer_len = MIN_NON_ZERO(max_xfer_len, BDRV_REQUEST_MAX_SECTORS);
399 qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
402 for (i = 0; i < mrb->num_reqs; i++) {
403 VirtIOBlockReq *req = mrb->reqs[i];
406 * NOTE: We cannot merge the requests in below situations:
407 * 1. requests are not sequential
408 * 2. merge would exceed maximum number of IOVs
409 * 3. merge would exceed maximum transfer length of backend device
411 if (sector_num + nb_sectors != req->sector_num ||
412 niov > blk_get_max_iov(blk) - req->qiov.niov ||
413 req->qiov.size / BDRV_SECTOR_SIZE > max_xfer_len ||
414 nb_sectors > max_xfer_len - req->qiov.size / BDRV_SECTOR_SIZE) {
415 submit_requests(blk, mrb, start, num_reqs, niov);
421 sector_num = req->sector_num;
422 nb_sectors = niov = 0;
426 nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
427 niov += req->qiov.niov;
431 submit_requests(blk, mrb, start, num_reqs, niov);
435 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
437 block_acct_start(blk_get_stats(req->dev->blk), &req->acct, 0,
441 * Make sure all outstanding writes are posted to the backing device.
443 if (mrb->is_write && mrb->num_reqs > 0) {
444 virtio_blk_submit_multireq(req->dev->blk, mrb);
446 blk_aio_flush(req->dev->blk, virtio_blk_flush_complete, req);
449 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
450 uint64_t sector, size_t size)
452 uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
453 uint64_t total_sectors;
455 if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
458 if (sector & dev->sector_mask) {
461 if (size % dev->conf.conf.logical_block_size) {
464 blk_get_geometry(dev->blk, &total_sectors);
465 if (sector > total_sectors || nb_sectors > total_sectors - sector) {
471 void virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
474 struct iovec *in_iov = req->elem.in_sg;
475 struct iovec *iov = req->elem.out_sg;
476 unsigned in_num = req->elem.in_num;
477 unsigned out_num = req->elem.out_num;
479 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
480 error_report("virtio-blk missing headers");
484 if (unlikely(iov_to_buf(iov, out_num, 0, &req->out,
485 sizeof(req->out)) != sizeof(req->out))) {
486 error_report("virtio-blk request outhdr too short");
490 iov_discard_front(&iov, &out_num, sizeof(req->out));
492 if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
493 error_report("virtio-blk request inhdr too short");
497 /* We always touch the last byte, so just see how big in_iov is. */
498 req->in_len = iov_size(in_iov, in_num);
499 req->in = (void *)in_iov[in_num - 1].iov_base
500 + in_iov[in_num - 1].iov_len
501 - sizeof(struct virtio_blk_inhdr);
502 iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
504 type = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
506 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
507 * is an optional flag. Although a guest should not send this flag if
508 * not negotiated we ignored it in the past. So keep ignoring it. */
509 switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
510 case VIRTIO_BLK_T_IN:
512 bool is_write = type & VIRTIO_BLK_T_OUT;
513 req->sector_num = virtio_ldq_p(VIRTIO_DEVICE(req->dev),
517 qemu_iovec_init_external(&req->qiov, iov, out_num);
518 trace_virtio_blk_handle_write(req, req->sector_num,
519 req->qiov.size / BDRV_SECTOR_SIZE);
521 qemu_iovec_init_external(&req->qiov, in_iov, in_num);
522 trace_virtio_blk_handle_read(req, req->sector_num,
523 req->qiov.size / BDRV_SECTOR_SIZE);
526 if (!virtio_blk_sect_range_ok(req->dev, req->sector_num,
528 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
529 block_acct_invalid(blk_get_stats(req->dev->blk),
530 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
531 virtio_blk_free_request(req);
535 block_acct_start(blk_get_stats(req->dev->blk),
536 &req->acct, req->qiov.size,
537 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
539 /* merge would exceed maximum number of requests or IO direction
541 if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
542 is_write != mrb->is_write ||
543 !req->dev->conf.request_merging)) {
544 virtio_blk_submit_multireq(req->dev->blk, mrb);
547 assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
548 mrb->reqs[mrb->num_reqs++] = req;
549 mrb->is_write = is_write;
552 case VIRTIO_BLK_T_FLUSH:
553 virtio_blk_handle_flush(req, mrb);
555 case VIRTIO_BLK_T_SCSI_CMD:
556 virtio_blk_handle_scsi(req);
558 case VIRTIO_BLK_T_GET_ID:
560 VirtIOBlock *s = req->dev;
563 * NB: per existing s/n string convention the string is
564 * terminated by '\0' only when shorter than buffer.
566 const char *serial = s->conf.serial ? s->conf.serial : "";
567 size_t size = MIN(strlen(serial) + 1,
568 MIN(iov_size(in_iov, in_num),
569 VIRTIO_BLK_ID_BYTES));
570 iov_from_buf(in_iov, in_num, 0, serial, size);
571 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
572 virtio_blk_free_request(req);
576 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
577 virtio_blk_free_request(req);
581 void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
584 MultiReqBuffer mrb = {};
588 while ((req = virtio_blk_get_request(s, vq))) {
589 virtio_blk_handle_request(req, &mrb);
593 virtio_blk_submit_multireq(s->blk, &mrb);
596 blk_io_unplug(s->blk);
599 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
601 VirtIOBlock *s = (VirtIOBlock *)vdev;
604 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
605 * dataplane here instead of waiting for .set_status().
607 virtio_blk_data_plane_start(s->dataplane);
608 if (!s->dataplane_disabled) {
612 virtio_blk_handle_vq(s, vq);
615 static void virtio_blk_dma_restart_bh(void *opaque)
617 VirtIOBlock *s = opaque;
618 VirtIOBlockReq *req = s->rq;
619 MultiReqBuffer mrb = {};
621 qemu_bh_delete(s->bh);
627 VirtIOBlockReq *next = req->next;
628 virtio_blk_handle_request(req, &mrb);
633 virtio_blk_submit_multireq(s->blk, &mrb);
637 static void virtio_blk_dma_restart_cb(void *opaque, int running,
640 VirtIOBlock *s = opaque;
647 s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
648 virtio_blk_dma_restart_bh, s);
649 qemu_bh_schedule(s->bh);
653 static void virtio_blk_reset(VirtIODevice *vdev)
655 VirtIOBlock *s = VIRTIO_BLK(vdev);
659 * This should cancel pending requests, but can't do nicely until there
660 * are per-device request lists.
662 ctx = blk_get_aio_context(s->blk);
663 aio_context_acquire(ctx);
667 virtio_blk_data_plane_stop(s->dataplane);
669 aio_context_release(ctx);
671 blk_set_enable_write_cache(s->blk, s->original_wce);
674 /* coalesce internal state, copy to pci i/o region 0
676 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
678 VirtIOBlock *s = VIRTIO_BLK(vdev);
679 BlockConf *conf = &s->conf.conf;
680 struct virtio_blk_config blkcfg;
682 int blk_size = conf->logical_block_size;
684 blk_get_geometry(s->blk, &capacity);
685 memset(&blkcfg, 0, sizeof(blkcfg));
686 virtio_stq_p(vdev, &blkcfg.capacity, capacity);
687 virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2);
688 virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
689 virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
690 virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
691 virtio_stw_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
692 blkcfg.geometry.heads = conf->heads;
694 * We must ensure that the block device capacity is a multiple of
695 * the logical block size. If that is not the case, let's use
696 * sector_mask to adopt the geometry to have a correct picture.
697 * For those devices where the capacity is ok for the given geometry
698 * we don't touch the sector value of the geometry, since some devices
699 * (like s390 dasd) need a specific value. Here the capacity is already
700 * cyls*heads*secs*blk_size and the sector value is not block size
701 * divided by 512 - instead it is the amount of blk_size blocks
702 * per track (cylinder).
704 if (blk_getlength(s->blk) / conf->heads / conf->secs % blk_size) {
705 blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
707 blkcfg.geometry.sectors = conf->secs;
710 blkcfg.physical_block_exp = get_physical_block_exp(conf);
711 blkcfg.alignment_offset = 0;
712 blkcfg.wce = blk_enable_write_cache(s->blk);
713 virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
714 memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
717 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
719 VirtIOBlock *s = VIRTIO_BLK(vdev);
720 struct virtio_blk_config blkcfg;
722 memcpy(&blkcfg, config, sizeof(blkcfg));
724 aio_context_acquire(blk_get_aio_context(s->blk));
725 blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
726 aio_context_release(blk_get_aio_context(s->blk));
729 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
732 VirtIOBlock *s = VIRTIO_BLK(vdev);
734 virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
735 virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
736 virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
737 virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
738 if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
740 error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
744 virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
745 virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
748 if (s->conf.config_wce) {
749 virtio_add_feature(&features, VIRTIO_BLK_F_CONFIG_WCE);
751 if (blk_enable_write_cache(s->blk)) {
752 virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
754 if (blk_is_read_only(s->blk)) {
755 virtio_add_feature(&features, VIRTIO_BLK_F_RO);
757 if (s->conf.num_queues > 1) {
758 virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
764 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
766 VirtIOBlock *s = VIRTIO_BLK(vdev);
768 if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
769 VIRTIO_CONFIG_S_DRIVER_OK))) {
770 virtio_blk_data_plane_stop(s->dataplane);
773 if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
777 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
778 * cache flushes. Thus, the "auto writethrough" behavior is never
779 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
780 * Leaving it enabled would break the following sequence:
782 * Guest started with "-drive cache=writethrough"
783 * Guest sets status to 0
784 * Guest sets DRIVER bit in status field
785 * Guest reads host features (WCE=0, CONFIG_WCE=1)
786 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
787 * Guest writes 1 to the WCE configuration field (writeback mode)
788 * Guest sets DRIVER_OK bit in status field
790 * s->blk would erroneously be placed in writethrough mode.
792 if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
793 aio_context_acquire(blk_get_aio_context(s->blk));
794 blk_set_enable_write_cache(s->blk,
795 virtio_vdev_has_feature(vdev,
797 aio_context_release(blk_get_aio_context(s->blk));
801 static void virtio_blk_save(QEMUFile *f, void *opaque)
803 VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
805 virtio_save(vdev, f);
808 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
810 VirtIOBlock *s = VIRTIO_BLK(vdev);
811 VirtIOBlockReq *req = s->rq;
814 qemu_put_sbyte(f, 1);
816 if (s->conf.num_queues > 1) {
817 qemu_put_be32(f, virtio_get_queue_index(req->vq));
820 qemu_put_virtqueue_element(f, &req->elem);
823 qemu_put_sbyte(f, 0);
826 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
828 VirtIOBlock *s = opaque;
829 VirtIODevice *vdev = VIRTIO_DEVICE(s);
834 return virtio_load(vdev, f, version_id);
837 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
840 VirtIOBlock *s = VIRTIO_BLK(vdev);
842 while (qemu_get_sbyte(f)) {
843 unsigned nvqs = s->conf.num_queues;
848 vq_idx = qemu_get_be32(f);
850 if (vq_idx >= nvqs) {
851 error_report("Invalid virtqueue index in request list: %#x",
857 req = qemu_get_virtqueue_element(f, sizeof(VirtIOBlockReq));
858 virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
866 static void virtio_blk_resize(void *opaque)
868 VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
870 virtio_notify_config(vdev);
873 static const BlockDevOps virtio_block_ops = {
874 .resize_cb = virtio_blk_resize,
877 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
879 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
880 VirtIOBlock *s = VIRTIO_BLK(dev);
881 VirtIOBlkConf *conf = &s->conf;
883 static int virtio_blk_id;
886 if (!conf->conf.blk) {
887 error_setg(errp, "drive property not set");
890 if (!blk_is_inserted(conf->conf.blk)) {
891 error_setg(errp, "Device needs media, but drive is empty");
894 if (!conf->num_queues) {
895 error_setg(errp, "num-queues property must be larger than 0");
899 blkconf_serial(&conf->conf, &conf->serial);
900 s->original_wce = blk_enable_write_cache(conf->conf.blk);
901 blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, &err);
903 error_propagate(errp, err);
906 blkconf_blocksizes(&conf->conf);
908 virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
909 sizeof(struct virtio_blk_config));
911 s->blk = conf->conf.blk;
913 s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
915 for (i = 0; i < conf->num_queues; i++) {
916 virtio_add_queue(vdev, 128, virtio_blk_handle_output);
918 virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
920 error_propagate(errp, err);
921 virtio_cleanup(vdev);
925 s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
926 register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
927 virtio_blk_save, virtio_blk_load, s);
928 blk_set_dev_ops(s->blk, &virtio_block_ops, s);
929 blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
931 blk_iostatus_enable(s->blk);
934 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
936 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
937 VirtIOBlock *s = VIRTIO_BLK(dev);
939 virtio_blk_data_plane_destroy(s->dataplane);
941 qemu_del_vm_change_state_handler(s->change);
942 unregister_savevm(dev, "virtio-blk", s);
943 blockdev_mark_auto_del(s->blk);
944 virtio_cleanup(vdev);
947 static void virtio_blk_instance_init(Object *obj)
949 VirtIOBlock *s = VIRTIO_BLK(obj);
951 object_property_add_link(obj, "iothread", TYPE_IOTHREAD,
952 (Object **)&s->conf.iothread,
953 qdev_prop_allow_set_link_before_realize,
954 OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL);
955 device_add_bootindex_property(obj, &s->conf.conf.bootindex,
956 "bootindex", "/disk@0,0",
960 static Property virtio_blk_properties[] = {
961 DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
962 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
963 DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
964 DEFINE_PROP_BIT("config-wce", VirtIOBlock, conf.config_wce, 0, true),
966 DEFINE_PROP_BIT("scsi", VirtIOBlock, conf.scsi, 0, false),
968 DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
970 DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
971 DEFINE_PROP_END_OF_LIST(),
974 static void virtio_blk_class_init(ObjectClass *klass, void *data)
976 DeviceClass *dc = DEVICE_CLASS(klass);
977 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
979 dc->props = virtio_blk_properties;
980 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
981 vdc->realize = virtio_blk_device_realize;
982 vdc->unrealize = virtio_blk_device_unrealize;
983 vdc->get_config = virtio_blk_update_config;
984 vdc->set_config = virtio_blk_set_config;
985 vdc->get_features = virtio_blk_get_features;
986 vdc->set_status = virtio_blk_set_status;
987 vdc->reset = virtio_blk_reset;
988 vdc->save = virtio_blk_save_device;
989 vdc->load = virtio_blk_load_device;
992 static const TypeInfo virtio_device_info = {
993 .name = TYPE_VIRTIO_BLK,
994 .parent = TYPE_VIRTIO_DEVICE,
995 .instance_size = sizeof(VirtIOBlock),
996 .instance_init = virtio_blk_instance_init,
997 .class_init = virtio_blk_class_init,
1000 static void virtio_register_types(void)
1002 type_register_static(&virtio_device_info);
1005 type_init(virtio_register_types)