4 * Copyright IBM, Corp. 2007
7 * Anthony Liguori <aliguori@us.ibm.com>
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-common.h"
16 #include "qemu/error-report.h"
18 #include "hw/block/block.h"
19 #include "sysemu/block-backend.h"
20 #include "sysemu/blockdev.h"
21 #include "hw/virtio/virtio-blk.h"
22 #include "dataplane/virtio-blk.h"
23 #include "migration/migration.h"
24 #include "block/scsi.h"
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/virtio/virtio-access.h"
31 VirtIOBlockReq
*virtio_blk_alloc_request(VirtIOBlock
*s
)
33 VirtIOBlockReq
*req
= g_new(VirtIOBlockReq
, 1);
42 void virtio_blk_free_request(VirtIOBlockReq
*req
)
49 static void virtio_blk_complete_request(VirtIOBlockReq
*req
,
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(s
->vq
, &req
->elem
, req
->in_len
);
59 virtio_notify(vdev
, s
->vq
);
62 static void virtio_blk_req_complete(VirtIOBlockReq
*req
, unsigned char status
)
64 req
->dev
->complete_request(req
, status
);
67 static int virtio_blk_handle_rw_error(VirtIOBlockReq
*req
, int error
,
70 BlockErrorAction action
= blk_get_error_action(req
->dev
->blk
,
72 VirtIOBlock
*s
= req
->dev
;
74 if (action
== BLOCK_ERROR_ACTION_STOP
) {
75 /* Break the link as the next request is going to be parsed from the
76 * ring again. Otherwise we may end up doing a double completion! */
80 } else if (action
== BLOCK_ERROR_ACTION_REPORT
) {
81 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
82 block_acct_failed(blk_get_stats(s
->blk
), &req
->acct
);
83 virtio_blk_free_request(req
);
86 blk_error_action(s
->blk
, action
, is_read
, error
);
87 return action
!= BLOCK_ERROR_ACTION_IGNORE
;
90 static void virtio_blk_rw_complete(void *opaque
, int ret
)
92 VirtIOBlockReq
*next
= opaque
;
95 VirtIOBlockReq
*req
= next
;
97 trace_virtio_blk_rw_complete(req
, ret
);
99 if (req
->qiov
.nalloc
!= -1) {
100 /* If nalloc is != 1 req->qiov is a local copy of the original
101 * external iovec. It was allocated in submit_merged_requests
102 * to be able to merge requests. */
103 qemu_iovec_destroy(&req
->qiov
);
107 int p
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
108 bool is_read
= !(p
& VIRTIO_BLK_T_OUT
);
109 /* Note that memory may be dirtied on read failure. If the
110 * virtio request is not completed here, as is the case for
111 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
112 * correctly during live migration. While this is ugly,
113 * it is acceptable because the device is free to write to
114 * the memory until the request is completed (which will
115 * happen on the other side of the migration).
117 if (virtio_blk_handle_rw_error(req
, -ret
, is_read
)) {
122 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
123 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
124 virtio_blk_free_request(req
);
128 static void virtio_blk_flush_complete(void *opaque
, int ret
)
130 VirtIOBlockReq
*req
= opaque
;
133 if (virtio_blk_handle_rw_error(req
, -ret
, 0)) {
138 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
139 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
140 virtio_blk_free_request(req
);
147 struct sg_io_hdr hdr
;
148 } VirtIOBlockIoctlReq
;
150 static void virtio_blk_ioctl_complete(void *opaque
, int status
)
152 VirtIOBlockIoctlReq
*ioctl_req
= opaque
;
153 VirtIOBlockReq
*req
= ioctl_req
->req
;
154 VirtIODevice
*vdev
= VIRTIO_DEVICE(req
->dev
);
155 struct virtio_scsi_inhdr
*scsi
;
156 struct sg_io_hdr
*hdr
;
158 scsi
= (void *)req
->elem
.in_sg
[req
->elem
.in_num
- 2].iov_base
;
161 status
= VIRTIO_BLK_S_UNSUPP
;
162 virtio_stl_p(vdev
, &scsi
->errors
, 255);
166 hdr
= &ioctl_req
->hdr
;
168 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
169 * clear the masked_status field [hence status gets cleared too, see
170 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
171 * status has occurred. However they do set DRIVER_SENSE in driver_status
172 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
174 if (hdr
->status
== 0 && hdr
->sb_len_wr
> 0) {
175 hdr
->status
= CHECK_CONDITION
;
178 virtio_stl_p(vdev
, &scsi
->errors
,
179 hdr
->status
| (hdr
->msg_status
<< 8) |
180 (hdr
->host_status
<< 16) | (hdr
->driver_status
<< 24));
181 virtio_stl_p(vdev
, &scsi
->residual
, hdr
->resid
);
182 virtio_stl_p(vdev
, &scsi
->sense_len
, hdr
->sb_len_wr
);
183 virtio_stl_p(vdev
, &scsi
->data_len
, hdr
->dxfer_len
);
186 virtio_blk_req_complete(req
, status
);
187 virtio_blk_free_request(req
);
193 static VirtIOBlockReq
*virtio_blk_get_request(VirtIOBlock
*s
)
195 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
197 if (!virtqueue_pop(s
->vq
, &req
->elem
)) {
198 virtio_blk_free_request(req
);
205 static int virtio_blk_handle_scsi_req(VirtIOBlockReq
*req
)
207 int status
= VIRTIO_BLK_S_OK
;
208 struct virtio_scsi_inhdr
*scsi
= NULL
;
209 VirtIODevice
*vdev
= VIRTIO_DEVICE(req
->dev
);
210 VirtQueueElement
*elem
= &req
->elem
;
211 VirtIOBlock
*blk
= req
->dev
;
215 VirtIOBlockIoctlReq
*ioctl_req
;
220 * We require at least one output segment each for the virtio_blk_outhdr
221 * and the SCSI command block.
223 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
224 * and the sense buffer pointer in the input segments.
226 if (elem
->out_num
< 2 || elem
->in_num
< 3) {
227 status
= VIRTIO_BLK_S_IOERR
;
232 * The scsi inhdr is placed in the second-to-last input segment, just
233 * before the regular inhdr.
235 scsi
= (void *)elem
->in_sg
[elem
->in_num
- 2].iov_base
;
237 if (!blk
->conf
.scsi
) {
238 status
= VIRTIO_BLK_S_UNSUPP
;
243 * No support for bidirection commands yet.
245 if (elem
->out_num
> 2 && elem
->in_num
> 3) {
246 status
= VIRTIO_BLK_S_UNSUPP
;
251 ioctl_req
= g_new0(VirtIOBlockIoctlReq
, 1);
252 ioctl_req
->req
= req
;
253 ioctl_req
->hdr
.interface_id
= 'S';
254 ioctl_req
->hdr
.cmd_len
= elem
->out_sg
[1].iov_len
;
255 ioctl_req
->hdr
.cmdp
= elem
->out_sg
[1].iov_base
;
256 ioctl_req
->hdr
.dxfer_len
= 0;
258 if (elem
->out_num
> 2) {
260 * If there are more than the minimally required 2 output segments
261 * there is write payload starting from the third iovec.
263 ioctl_req
->hdr
.dxfer_direction
= SG_DXFER_TO_DEV
;
264 ioctl_req
->hdr
.iovec_count
= elem
->out_num
- 2;
266 for (i
= 0; i
< ioctl_req
->hdr
.iovec_count
; i
++) {
267 ioctl_req
->hdr
.dxfer_len
+= elem
->out_sg
[i
+ 2].iov_len
;
270 ioctl_req
->hdr
.dxferp
= elem
->out_sg
+ 2;
272 } else if (elem
->in_num
> 3) {
274 * If we have more than 3 input segments the guest wants to actually
277 ioctl_req
->hdr
.dxfer_direction
= SG_DXFER_FROM_DEV
;
278 ioctl_req
->hdr
.iovec_count
= elem
->in_num
- 3;
279 for (i
= 0; i
< ioctl_req
->hdr
.iovec_count
; i
++) {
280 ioctl_req
->hdr
.dxfer_len
+= elem
->in_sg
[i
].iov_len
;
283 ioctl_req
->hdr
.dxferp
= elem
->in_sg
;
286 * Some SCSI commands don't actually transfer any data.
288 ioctl_req
->hdr
.dxfer_direction
= SG_DXFER_NONE
;
291 ioctl_req
->hdr
.sbp
= elem
->in_sg
[elem
->in_num
- 3].iov_base
;
292 ioctl_req
->hdr
.mx_sb_len
= elem
->in_sg
[elem
->in_num
- 3].iov_len
;
294 acb
= blk_aio_ioctl(blk
->blk
, SG_IO
, &ioctl_req
->hdr
,
295 virtio_blk_ioctl_complete
, ioctl_req
);
298 status
= VIRTIO_BLK_S_UNSUPP
;
307 /* Just put anything nonzero so that the ioctl fails in the guest. */
309 virtio_stl_p(vdev
, &scsi
->errors
, 255);
314 static void virtio_blk_handle_scsi(VirtIOBlockReq
*req
)
318 status
= virtio_blk_handle_scsi_req(req
);
319 if (status
!= -EINPROGRESS
) {
320 virtio_blk_req_complete(req
, status
);
321 virtio_blk_free_request(req
);
325 static inline void submit_requests(BlockBackend
*blk
, MultiReqBuffer
*mrb
,
326 int start
, int num_reqs
, int niov
)
328 QEMUIOVector
*qiov
= &mrb
->reqs
[start
]->qiov
;
329 int64_t sector_num
= mrb
->reqs
[start
]->sector_num
;
330 int nb_sectors
= mrb
->reqs
[start
]->qiov
.size
/ BDRV_SECTOR_SIZE
;
331 bool is_write
= mrb
->is_write
;
335 struct iovec
*tmp_iov
= qiov
->iov
;
336 int tmp_niov
= qiov
->niov
;
338 /* mrb->reqs[start]->qiov was initialized from external so we can't
339 * modifiy it here. We need to initialize it locally and then add the
340 * external iovecs. */
341 qemu_iovec_init(qiov
, niov
);
343 for (i
= 0; i
< tmp_niov
; i
++) {
344 qemu_iovec_add(qiov
, tmp_iov
[i
].iov_base
, tmp_iov
[i
].iov_len
);
347 for (i
= start
+ 1; i
< start
+ num_reqs
; i
++) {
348 qemu_iovec_concat(qiov
, &mrb
->reqs
[i
]->qiov
, 0,
349 mrb
->reqs
[i
]->qiov
.size
);
350 mrb
->reqs
[i
- 1]->mr_next
= mrb
->reqs
[i
];
351 nb_sectors
+= mrb
->reqs
[i
]->qiov
.size
/ BDRV_SECTOR_SIZE
;
353 assert(nb_sectors
== qiov
->size
/ BDRV_SECTOR_SIZE
);
355 trace_virtio_blk_submit_multireq(mrb
, start
, num_reqs
, sector_num
,
356 nb_sectors
, is_write
);
357 block_acct_merge_done(blk_get_stats(blk
),
358 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
,
363 blk_aio_writev(blk
, sector_num
, qiov
, nb_sectors
,
364 virtio_blk_rw_complete
, mrb
->reqs
[start
]);
366 blk_aio_readv(blk
, sector_num
, qiov
, nb_sectors
,
367 virtio_blk_rw_complete
, mrb
->reqs
[start
]);
371 static int multireq_compare(const void *a
, const void *b
)
373 const VirtIOBlockReq
*req1
= *(VirtIOBlockReq
**)a
,
374 *req2
= *(VirtIOBlockReq
**)b
;
377 * Note that we can't simply subtract sector_num1 from sector_num2
378 * here as that could overflow the return value.
380 if (req1
->sector_num
> req2
->sector_num
) {
382 } else if (req1
->sector_num
< req2
->sector_num
) {
389 void virtio_blk_submit_multireq(BlockBackend
*blk
, MultiReqBuffer
*mrb
)
391 int i
= 0, start
= 0, num_reqs
= 0, niov
= 0, nb_sectors
= 0;
392 int max_xfer_len
= 0;
393 int64_t sector_num
= 0;
395 if (mrb
->num_reqs
== 1) {
396 submit_requests(blk
, mrb
, 0, 1, -1);
401 max_xfer_len
= blk_get_max_transfer_length(mrb
->reqs
[0]->dev
->blk
);
402 max_xfer_len
= MIN_NON_ZERO(max_xfer_len
, BDRV_REQUEST_MAX_SECTORS
);
404 qsort(mrb
->reqs
, mrb
->num_reqs
, sizeof(*mrb
->reqs
),
407 for (i
= 0; i
< mrb
->num_reqs
; i
++) {
408 VirtIOBlockReq
*req
= mrb
->reqs
[i
];
412 /* merge would exceed maximum number of IOVs */
413 if (niov
+ req
->qiov
.niov
> IOV_MAX
) {
417 /* merge would exceed maximum transfer length of backend device */
418 if (req
->qiov
.size
/ BDRV_SECTOR_SIZE
+ nb_sectors
> max_xfer_len
) {
422 /* requests are not sequential */
423 if (sector_num
+ nb_sectors
!= req
->sector_num
) {
428 submit_requests(blk
, mrb
, start
, num_reqs
, niov
);
434 sector_num
= req
->sector_num
;
435 nb_sectors
= niov
= 0;
439 nb_sectors
+= req
->qiov
.size
/ BDRV_SECTOR_SIZE
;
440 niov
+= req
->qiov
.niov
;
444 submit_requests(blk
, mrb
, start
, num_reqs
, niov
);
448 static void virtio_blk_handle_flush(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
450 block_acct_start(blk_get_stats(req
->dev
->blk
), &req
->acct
, 0,
454 * Make sure all outstanding writes are posted to the backing device.
456 if (mrb
->is_write
&& mrb
->num_reqs
> 0) {
457 virtio_blk_submit_multireq(req
->dev
->blk
, mrb
);
459 blk_aio_flush(req
->dev
->blk
, virtio_blk_flush_complete
, req
);
462 static bool virtio_blk_sect_range_ok(VirtIOBlock
*dev
,
463 uint64_t sector
, size_t size
)
465 uint64_t nb_sectors
= size
>> BDRV_SECTOR_BITS
;
466 uint64_t total_sectors
;
468 if (nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
471 if (sector
& dev
->sector_mask
) {
474 if (size
% dev
->conf
.conf
.logical_block_size
) {
477 blk_get_geometry(dev
->blk
, &total_sectors
);
478 if (sector
> total_sectors
|| nb_sectors
> total_sectors
- sector
) {
484 void virtio_blk_handle_request(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
487 struct iovec
*in_iov
= req
->elem
.in_sg
;
488 struct iovec
*iov
= req
->elem
.out_sg
;
489 unsigned in_num
= req
->elem
.in_num
;
490 unsigned out_num
= req
->elem
.out_num
;
492 if (req
->elem
.out_num
< 1 || req
->elem
.in_num
< 1) {
493 error_report("virtio-blk missing headers");
497 if (unlikely(iov_to_buf(iov
, out_num
, 0, &req
->out
,
498 sizeof(req
->out
)) != sizeof(req
->out
))) {
499 error_report("virtio-blk request outhdr too short");
503 iov_discard_front(&iov
, &out_num
, sizeof(req
->out
));
505 if (in_iov
[in_num
- 1].iov_len
< sizeof(struct virtio_blk_inhdr
)) {
506 error_report("virtio-blk request inhdr too short");
510 /* We always touch the last byte, so just see how big in_iov is. */
511 req
->in_len
= iov_size(in_iov
, in_num
);
512 req
->in
= (void *)in_iov
[in_num
- 1].iov_base
513 + in_iov
[in_num
- 1].iov_len
514 - sizeof(struct virtio_blk_inhdr
);
515 iov_discard_back(in_iov
, &in_num
, sizeof(struct virtio_blk_inhdr
));
517 type
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
519 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
520 * is an optional flag. Although a guest should not send this flag if
521 * not negotiated we ignored it in the past. So keep ignoring it. */
522 switch (type
& ~(VIRTIO_BLK_T_OUT
| VIRTIO_BLK_T_BARRIER
)) {
523 case VIRTIO_BLK_T_IN
:
525 bool is_write
= type
& VIRTIO_BLK_T_OUT
;
526 req
->sector_num
= virtio_ldq_p(VIRTIO_DEVICE(req
->dev
),
530 qemu_iovec_init_external(&req
->qiov
, iov
, out_num
);
531 trace_virtio_blk_handle_write(req
, req
->sector_num
,
532 req
->qiov
.size
/ BDRV_SECTOR_SIZE
);
534 qemu_iovec_init_external(&req
->qiov
, in_iov
, in_num
);
535 trace_virtio_blk_handle_read(req
, req
->sector_num
,
536 req
->qiov
.size
/ BDRV_SECTOR_SIZE
);
539 if (!virtio_blk_sect_range_ok(req
->dev
, req
->sector_num
,
541 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
542 block_acct_invalid(blk_get_stats(req
->dev
->blk
),
543 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
);
544 virtio_blk_free_request(req
);
548 block_acct_start(blk_get_stats(req
->dev
->blk
),
549 &req
->acct
, req
->qiov
.size
,
550 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
);
552 /* merge would exceed maximum number of requests or IO direction
554 if (mrb
->num_reqs
> 0 && (mrb
->num_reqs
== VIRTIO_BLK_MAX_MERGE_REQS
||
555 is_write
!= mrb
->is_write
||
556 !req
->dev
->conf
.request_merging
)) {
557 virtio_blk_submit_multireq(req
->dev
->blk
, mrb
);
560 assert(mrb
->num_reqs
< VIRTIO_BLK_MAX_MERGE_REQS
);
561 mrb
->reqs
[mrb
->num_reqs
++] = req
;
562 mrb
->is_write
= is_write
;
565 case VIRTIO_BLK_T_FLUSH
:
566 virtio_blk_handle_flush(req
, mrb
);
568 case VIRTIO_BLK_T_SCSI_CMD
:
569 virtio_blk_handle_scsi(req
);
571 case VIRTIO_BLK_T_GET_ID
:
573 VirtIOBlock
*s
= req
->dev
;
576 * NB: per existing s/n string convention the string is
577 * terminated by '\0' only when shorter than buffer.
579 const char *serial
= s
->conf
.serial
? s
->conf
.serial
: "";
580 size_t size
= MIN(strlen(serial
) + 1,
581 MIN(iov_size(in_iov
, in_num
),
582 VIRTIO_BLK_ID_BYTES
));
583 iov_from_buf(in_iov
, in_num
, 0, serial
, size
);
584 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
585 virtio_blk_free_request(req
);
589 virtio_blk_req_complete(req
, VIRTIO_BLK_S_UNSUPP
);
590 virtio_blk_free_request(req
);
594 static void virtio_blk_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
596 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
598 MultiReqBuffer mrb
= {};
600 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
601 * dataplane here instead of waiting for .set_status().
604 virtio_blk_data_plane_start(s
->dataplane
);
610 while ((req
= virtio_blk_get_request(s
))) {
611 virtio_blk_handle_request(req
, &mrb
);
615 virtio_blk_submit_multireq(s
->blk
, &mrb
);
618 blk_io_unplug(s
->blk
);
621 static void virtio_blk_dma_restart_bh(void *opaque
)
623 VirtIOBlock
*s
= opaque
;
624 VirtIOBlockReq
*req
= s
->rq
;
625 MultiReqBuffer mrb
= {};
627 qemu_bh_delete(s
->bh
);
633 VirtIOBlockReq
*next
= req
->next
;
634 virtio_blk_handle_request(req
, &mrb
);
639 virtio_blk_submit_multireq(s
->blk
, &mrb
);
643 static void virtio_blk_dma_restart_cb(void *opaque
, int running
,
646 VirtIOBlock
*s
= opaque
;
653 s
->bh
= aio_bh_new(blk_get_aio_context(s
->conf
.conf
.blk
),
654 virtio_blk_dma_restart_bh
, s
);
655 qemu_bh_schedule(s
->bh
);
659 static void virtio_blk_reset(VirtIODevice
*vdev
)
661 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
665 * This should cancel pending requests, but can't do nicely until there
666 * are per-device request lists.
668 ctx
= blk_get_aio_context(s
->blk
);
669 aio_context_acquire(ctx
);
673 virtio_blk_data_plane_stop(s
->dataplane
);
675 aio_context_release(ctx
);
677 blk_set_enable_write_cache(s
->blk
, s
->original_wce
);
680 /* coalesce internal state, copy to pci i/o region 0
682 static void virtio_blk_update_config(VirtIODevice
*vdev
, uint8_t *config
)
684 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
685 BlockConf
*conf
= &s
->conf
.conf
;
686 struct virtio_blk_config blkcfg
;
688 int blk_size
= conf
->logical_block_size
;
690 blk_get_geometry(s
->blk
, &capacity
);
691 memset(&blkcfg
, 0, sizeof(blkcfg
));
692 virtio_stq_p(vdev
, &blkcfg
.capacity
, capacity
);
693 virtio_stl_p(vdev
, &blkcfg
.seg_max
, 128 - 2);
694 virtio_stw_p(vdev
, &blkcfg
.geometry
.cylinders
, conf
->cyls
);
695 virtio_stl_p(vdev
, &blkcfg
.blk_size
, blk_size
);
696 virtio_stw_p(vdev
, &blkcfg
.min_io_size
, conf
->min_io_size
/ blk_size
);
697 virtio_stw_p(vdev
, &blkcfg
.opt_io_size
, conf
->opt_io_size
/ blk_size
);
698 blkcfg
.geometry
.heads
= conf
->heads
;
700 * We must ensure that the block device capacity is a multiple of
701 * the logical block size. If that is not the case, let's use
702 * sector_mask to adopt the geometry to have a correct picture.
703 * For those devices where the capacity is ok for the given geometry
704 * we don't touch the sector value of the geometry, since some devices
705 * (like s390 dasd) need a specific value. Here the capacity is already
706 * cyls*heads*secs*blk_size and the sector value is not block size
707 * divided by 512 - instead it is the amount of blk_size blocks
708 * per track (cylinder).
710 if (blk_getlength(s
->blk
) / conf
->heads
/ conf
->secs
% blk_size
) {
711 blkcfg
.geometry
.sectors
= conf
->secs
& ~s
->sector_mask
;
713 blkcfg
.geometry
.sectors
= conf
->secs
;
716 blkcfg
.physical_block_exp
= get_physical_block_exp(conf
);
717 blkcfg
.alignment_offset
= 0;
718 blkcfg
.wce
= blk_enable_write_cache(s
->blk
);
719 memcpy(config
, &blkcfg
, sizeof(struct virtio_blk_config
));
722 static void virtio_blk_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
724 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
725 struct virtio_blk_config blkcfg
;
727 memcpy(&blkcfg
, config
, sizeof(blkcfg
));
729 aio_context_acquire(blk_get_aio_context(s
->blk
));
730 blk_set_enable_write_cache(s
->blk
, blkcfg
.wce
!= 0);
731 aio_context_release(blk_get_aio_context(s
->blk
));
734 static uint64_t virtio_blk_get_features(VirtIODevice
*vdev
, uint64_t features
,
737 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
739 virtio_add_feature(&features
, VIRTIO_BLK_F_SEG_MAX
);
740 virtio_add_feature(&features
, VIRTIO_BLK_F_GEOMETRY
);
741 virtio_add_feature(&features
, VIRTIO_BLK_F_TOPOLOGY
);
742 virtio_add_feature(&features
, VIRTIO_BLK_F_BLK_SIZE
);
743 if (virtio_has_feature(features
, VIRTIO_F_VERSION_1
)) {
745 error_setg(errp
, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
749 virtio_clear_feature(&features
, VIRTIO_F_ANY_LAYOUT
);
750 virtio_add_feature(&features
, VIRTIO_BLK_F_SCSI
);
753 if (s
->conf
.config_wce
) {
754 virtio_add_feature(&features
, VIRTIO_BLK_F_CONFIG_WCE
);
756 if (blk_enable_write_cache(s
->blk
)) {
757 virtio_add_feature(&features
, VIRTIO_BLK_F_WCE
);
759 if (blk_is_read_only(s
->blk
)) {
760 virtio_add_feature(&features
, VIRTIO_BLK_F_RO
);
766 static void virtio_blk_set_status(VirtIODevice
*vdev
, uint8_t status
)
768 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
770 if (s
->dataplane
&& !(status
& (VIRTIO_CONFIG_S_DRIVER
|
771 VIRTIO_CONFIG_S_DRIVER_OK
))) {
772 virtio_blk_data_plane_stop(s
->dataplane
);
775 if (!(status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
779 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
780 * cache flushes. Thus, the "auto writethrough" behavior is never
781 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
782 * Leaving it enabled would break the following sequence:
784 * Guest started with "-drive cache=writethrough"
785 * Guest sets status to 0
786 * Guest sets DRIVER bit in status field
787 * Guest reads host features (WCE=0, CONFIG_WCE=1)
788 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
789 * Guest writes 1 to the WCE configuration field (writeback mode)
790 * Guest sets DRIVER_OK bit in status field
792 * s->blk would erroneously be placed in writethrough mode.
794 if (!virtio_vdev_has_feature(vdev
, VIRTIO_BLK_F_CONFIG_WCE
)) {
795 aio_context_acquire(blk_get_aio_context(s
->blk
));
796 blk_set_enable_write_cache(s
->blk
,
797 virtio_vdev_has_feature(vdev
,
799 aio_context_release(blk_get_aio_context(s
->blk
));
803 static void virtio_blk_save(QEMUFile
*f
, void *opaque
)
805 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
806 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
809 virtio_blk_data_plane_stop(s
->dataplane
);
812 virtio_save(vdev
, f
);
815 static void virtio_blk_save_device(VirtIODevice
*vdev
, QEMUFile
*f
)
817 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
818 VirtIOBlockReq
*req
= s
->rq
;
821 qemu_put_sbyte(f
, 1);
822 qemu_put_buffer(f
, (unsigned char *)&req
->elem
,
823 sizeof(VirtQueueElement
));
826 qemu_put_sbyte(f
, 0);
829 static int virtio_blk_load(QEMUFile
*f
, void *opaque
, int version_id
)
831 VirtIOBlock
*s
= opaque
;
832 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
837 return virtio_load(vdev
, f
, version_id
);
840 static int virtio_blk_load_device(VirtIODevice
*vdev
, QEMUFile
*f
,
843 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
845 while (qemu_get_sbyte(f
)) {
846 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
847 qemu_get_buffer(f
, (unsigned char *)&req
->elem
,
848 sizeof(VirtQueueElement
));
852 virtqueue_map(&req
->elem
);
858 static void virtio_blk_resize(void *opaque
)
860 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
862 virtio_notify_config(vdev
);
865 static const BlockDevOps virtio_block_ops
= {
866 .resize_cb
= virtio_blk_resize
,
869 /* Disable dataplane thread during live migration since it does not
870 * update the dirty memory bitmap yet.
872 static void virtio_blk_migration_state_changed(Notifier
*notifier
, void *data
)
874 VirtIOBlock
*s
= container_of(notifier
, VirtIOBlock
,
875 migration_state_notifier
);
876 MigrationState
*mig
= data
;
879 if (migration_in_setup(mig
)) {
883 virtio_blk_data_plane_destroy(s
->dataplane
);
885 } else if (migration_has_finished(mig
) ||
886 migration_has_failed(mig
)) {
890 blk_drain_all(); /* complete in-flight non-dataplane requests */
891 virtio_blk_data_plane_create(VIRTIO_DEVICE(s
), &s
->conf
,
892 &s
->dataplane
, &err
);
894 error_report_err(err
);
899 static void virtio_blk_device_realize(DeviceState
*dev
, Error
**errp
)
901 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
902 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
903 VirtIOBlkConf
*conf
= &s
->conf
;
905 static int virtio_blk_id
;
907 if (!conf
->conf
.blk
) {
908 error_setg(errp
, "drive property not set");
911 if (!blk_is_inserted(conf
->conf
.blk
)) {
912 error_setg(errp
, "Device needs media, but drive is empty");
916 blkconf_serial(&conf
->conf
, &conf
->serial
);
917 s
->original_wce
= blk_enable_write_cache(conf
->conf
.blk
);
918 blkconf_geometry(&conf
->conf
, NULL
, 65535, 255, 255, &err
);
920 error_propagate(errp
, err
);
923 blkconf_blocksizes(&conf
->conf
);
925 virtio_init(vdev
, "virtio-blk", VIRTIO_ID_BLOCK
,
926 sizeof(struct virtio_blk_config
));
928 s
->blk
= conf
->conf
.blk
;
930 s
->sector_mask
= (s
->conf
.conf
.logical_block_size
/ BDRV_SECTOR_SIZE
) - 1;
932 s
->vq
= virtio_add_queue(vdev
, 128, virtio_blk_handle_output
);
933 s
->complete_request
= virtio_blk_complete_request
;
934 virtio_blk_data_plane_create(vdev
, conf
, &s
->dataplane
, &err
);
936 error_propagate(errp
, err
);
937 virtio_cleanup(vdev
);
940 s
->migration_state_notifier
.notify
= virtio_blk_migration_state_changed
;
941 add_migration_state_change_notifier(&s
->migration_state_notifier
);
943 s
->change
= qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb
, s
);
944 register_savevm(dev
, "virtio-blk", virtio_blk_id
++, 2,
945 virtio_blk_save
, virtio_blk_load
, s
);
946 blk_set_dev_ops(s
->blk
, &virtio_block_ops
, s
);
947 blk_set_guest_block_size(s
->blk
, s
->conf
.conf
.logical_block_size
);
949 blk_iostatus_enable(s
->blk
);
952 static void virtio_blk_device_unrealize(DeviceState
*dev
, Error
**errp
)
954 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
955 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
957 remove_migration_state_change_notifier(&s
->migration_state_notifier
);
958 virtio_blk_data_plane_destroy(s
->dataplane
);
960 qemu_del_vm_change_state_handler(s
->change
);
961 unregister_savevm(dev
, "virtio-blk", s
);
962 blockdev_mark_auto_del(s
->blk
);
963 virtio_cleanup(vdev
);
966 static void virtio_blk_instance_init(Object
*obj
)
968 VirtIOBlock
*s
= VIRTIO_BLK(obj
);
970 object_property_add_link(obj
, "iothread", TYPE_IOTHREAD
,
971 (Object
**)&s
->conf
.iothread
,
972 qdev_prop_allow_set_link_before_realize
,
973 OBJ_PROP_LINK_UNREF_ON_RELEASE
, NULL
);
974 device_add_bootindex_property(obj
, &s
->conf
.conf
.bootindex
,
975 "bootindex", "/disk@0,0",
979 static Property virtio_blk_properties
[] = {
980 DEFINE_BLOCK_PROPERTIES(VirtIOBlock
, conf
.conf
),
981 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock
, conf
.conf
),
982 DEFINE_PROP_STRING("serial", VirtIOBlock
, conf
.serial
),
983 DEFINE_PROP_BIT("config-wce", VirtIOBlock
, conf
.config_wce
, 0, true),
985 DEFINE_PROP_BIT("scsi", VirtIOBlock
, conf
.scsi
, 0, false),
987 DEFINE_PROP_BIT("request-merging", VirtIOBlock
, conf
.request_merging
, 0,
989 DEFINE_PROP_BIT("x-data-plane", VirtIOBlock
, conf
.data_plane
, 0, false),
990 DEFINE_PROP_END_OF_LIST(),
993 static void virtio_blk_class_init(ObjectClass
*klass
, void *data
)
995 DeviceClass
*dc
= DEVICE_CLASS(klass
);
996 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
998 dc
->props
= virtio_blk_properties
;
999 set_bit(DEVICE_CATEGORY_STORAGE
, dc
->categories
);
1000 vdc
->realize
= virtio_blk_device_realize
;
1001 vdc
->unrealize
= virtio_blk_device_unrealize
;
1002 vdc
->get_config
= virtio_blk_update_config
;
1003 vdc
->set_config
= virtio_blk_set_config
;
1004 vdc
->get_features
= virtio_blk_get_features
;
1005 vdc
->set_status
= virtio_blk_set_status
;
1006 vdc
->reset
= virtio_blk_reset
;
1007 vdc
->save
= virtio_blk_save_device
;
1008 vdc
->load
= virtio_blk_load_device
;
1011 static const TypeInfo virtio_device_info
= {
1012 .name
= TYPE_VIRTIO_BLK
,
1013 .parent
= TYPE_VIRTIO_DEVICE
,
1014 .instance_size
= sizeof(VirtIOBlock
),
1015 .instance_init
= virtio_blk_instance_init
,
1016 .class_init
= virtio_blk_class_init
,
1019 static void virtio_register_types(void)
1021 type_register_static(&virtio_device_info
);
1024 type_init(virtio_register_types
)