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_slice_new(VirtIOBlockReq
);
40 void virtio_blk_free_request(VirtIOBlockReq
*req
)
43 g_slice_free(VirtIOBlockReq
, req
);
47 static void virtio_blk_complete_request(VirtIOBlockReq
*req
,
50 VirtIOBlock
*s
= req
->dev
;
51 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
53 trace_virtio_blk_req_complete(req
, status
);
55 stb_p(&req
->in
->status
, status
);
56 virtqueue_push(s
->vq
, &req
->elem
, req
->qiov
.size
+ sizeof(*req
->in
));
57 virtio_notify(vdev
, s
->vq
);
60 static void virtio_blk_req_complete(VirtIOBlockReq
*req
, unsigned char status
)
62 req
->dev
->complete_request(req
, status
);
65 static int virtio_blk_handle_rw_error(VirtIOBlockReq
*req
, int error
,
68 BlockErrorAction action
= blk_get_error_action(req
->dev
->blk
,
70 VirtIOBlock
*s
= req
->dev
;
72 if (action
== BLOCK_ERROR_ACTION_STOP
) {
75 } else if (action
== BLOCK_ERROR_ACTION_REPORT
) {
76 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
77 block_acct_done(blk_get_stats(s
->blk
), &req
->acct
);
78 virtio_blk_free_request(req
);
81 blk_error_action(s
->blk
, action
, is_read
, error
);
82 return action
!= BLOCK_ERROR_ACTION_IGNORE
;
85 static void virtio_blk_rw_complete(void *opaque
, int ret
)
87 VirtIOBlockReq
*req
= opaque
;
89 trace_virtio_blk_rw_complete(req
, ret
);
92 int p
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
93 bool is_read
= !(p
& VIRTIO_BLK_T_OUT
);
94 if (virtio_blk_handle_rw_error(req
, -ret
, is_read
))
98 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
99 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
100 virtio_blk_free_request(req
);
103 static void virtio_blk_flush_complete(void *opaque
, int ret
)
105 VirtIOBlockReq
*req
= opaque
;
108 if (virtio_blk_handle_rw_error(req
, -ret
, 0)) {
113 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
114 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
115 virtio_blk_free_request(req
);
118 static VirtIOBlockReq
*virtio_blk_get_request(VirtIOBlock
*s
)
120 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
122 if (!virtqueue_pop(s
->vq
, &req
->elem
)) {
123 virtio_blk_free_request(req
);
130 int virtio_blk_handle_scsi_req(VirtIOBlock
*blk
,
131 VirtQueueElement
*elem
)
133 int status
= VIRTIO_BLK_S_OK
;
134 struct virtio_scsi_inhdr
*scsi
= NULL
;
135 VirtIODevice
*vdev
= VIRTIO_DEVICE(blk
);
139 struct sg_io_hdr hdr
;
143 * We require at least one output segment each for the virtio_blk_outhdr
144 * and the SCSI command block.
146 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
147 * and the sense buffer pointer in the input segments.
149 if (elem
->out_num
< 2 || elem
->in_num
< 3) {
150 status
= VIRTIO_BLK_S_IOERR
;
155 * The scsi inhdr is placed in the second-to-last input segment, just
156 * before the regular inhdr.
158 scsi
= (void *)elem
->in_sg
[elem
->in_num
- 2].iov_base
;
160 if (!blk
->conf
.scsi
) {
161 status
= VIRTIO_BLK_S_UNSUPP
;
166 * No support for bidirection commands yet.
168 if (elem
->out_num
> 2 && elem
->in_num
> 3) {
169 status
= VIRTIO_BLK_S_UNSUPP
;
174 memset(&hdr
, 0, sizeof(struct sg_io_hdr
));
175 hdr
.interface_id
= 'S';
176 hdr
.cmd_len
= elem
->out_sg
[1].iov_len
;
177 hdr
.cmdp
= elem
->out_sg
[1].iov_base
;
180 if (elem
->out_num
> 2) {
182 * If there are more than the minimally required 2 output segments
183 * there is write payload starting from the third iovec.
185 hdr
.dxfer_direction
= SG_DXFER_TO_DEV
;
186 hdr
.iovec_count
= elem
->out_num
- 2;
188 for (i
= 0; i
< hdr
.iovec_count
; i
++)
189 hdr
.dxfer_len
+= elem
->out_sg
[i
+ 2].iov_len
;
191 hdr
.dxferp
= elem
->out_sg
+ 2;
193 } else if (elem
->in_num
> 3) {
195 * If we have more than 3 input segments the guest wants to actually
198 hdr
.dxfer_direction
= SG_DXFER_FROM_DEV
;
199 hdr
.iovec_count
= elem
->in_num
- 3;
200 for (i
= 0; i
< hdr
.iovec_count
; i
++)
201 hdr
.dxfer_len
+= elem
->in_sg
[i
].iov_len
;
203 hdr
.dxferp
= elem
->in_sg
;
206 * Some SCSI commands don't actually transfer any data.
208 hdr
.dxfer_direction
= SG_DXFER_NONE
;
211 hdr
.sbp
= elem
->in_sg
[elem
->in_num
- 3].iov_base
;
212 hdr
.mx_sb_len
= elem
->in_sg
[elem
->in_num
- 3].iov_len
;
214 status
= blk_ioctl(blk
->blk
, SG_IO
, &hdr
);
216 status
= VIRTIO_BLK_S_UNSUPP
;
221 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
222 * clear the masked_status field [hence status gets cleared too, see
223 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
224 * status has occurred. However they do set DRIVER_SENSE in driver_status
225 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
227 if (hdr
.status
== 0 && hdr
.sb_len_wr
> 0) {
228 hdr
.status
= CHECK_CONDITION
;
231 virtio_stl_p(vdev
, &scsi
->errors
,
232 hdr
.status
| (hdr
.msg_status
<< 8) |
233 (hdr
.host_status
<< 16) | (hdr
.driver_status
<< 24));
234 virtio_stl_p(vdev
, &scsi
->residual
, hdr
.resid
);
235 virtio_stl_p(vdev
, &scsi
->sense_len
, hdr
.sb_len_wr
);
236 virtio_stl_p(vdev
, &scsi
->data_len
, hdr
.dxfer_len
);
244 /* Just put anything nonzero so that the ioctl fails in the guest. */
246 virtio_stl_p(vdev
, &scsi
->errors
, 255);
251 static void virtio_blk_handle_scsi(VirtIOBlockReq
*req
)
255 status
= virtio_blk_handle_scsi_req(req
->dev
, &req
->elem
);
256 virtio_blk_req_complete(req
, status
);
257 virtio_blk_free_request(req
);
260 void virtio_submit_multiwrite(BlockBackend
*blk
, MultiReqBuffer
*mrb
)
264 if (!mrb
->num_writes
) {
268 ret
= blk_aio_multiwrite(blk
, mrb
->blkreq
, mrb
->num_writes
);
270 for (i
= 0; i
< mrb
->num_writes
; i
++) {
271 if (mrb
->blkreq
[i
].error
) {
272 virtio_blk_rw_complete(mrb
->blkreq
[i
].opaque
, -EIO
);
280 static void virtio_blk_handle_flush(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
282 block_acct_start(blk_get_stats(req
->dev
->blk
), &req
->acct
, 0,
286 * Make sure all outstanding writes are posted to the backing device.
288 virtio_submit_multiwrite(req
->dev
->blk
, mrb
);
289 blk_aio_flush(req
->dev
->blk
, virtio_blk_flush_complete
, req
);
292 static bool virtio_blk_sect_range_ok(VirtIOBlock
*dev
,
293 uint64_t sector
, size_t size
)
295 uint64_t nb_sectors
= size
>> BDRV_SECTOR_BITS
;
296 uint64_t total_sectors
;
298 if (sector
& dev
->sector_mask
) {
301 if (size
% dev
->conf
.conf
.logical_block_size
) {
304 blk_get_geometry(dev
->blk
, &total_sectors
);
305 if (sector
> total_sectors
|| nb_sectors
> total_sectors
- sector
) {
311 static void virtio_blk_handle_write(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
313 BlockRequest
*blkreq
;
316 sector
= virtio_ldq_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.sector
);
318 trace_virtio_blk_handle_write(req
, sector
, req
->qiov
.size
/ 512);
320 if (!virtio_blk_sect_range_ok(req
->dev
, sector
, req
->qiov
.size
)) {
321 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
322 virtio_blk_free_request(req
);
326 block_acct_start(blk_get_stats(req
->dev
->blk
), &req
->acct
, req
->qiov
.size
,
329 if (mrb
->num_writes
== 32) {
330 virtio_submit_multiwrite(req
->dev
->blk
, mrb
);
333 blkreq
= &mrb
->blkreq
[mrb
->num_writes
];
334 blkreq
->sector
= sector
;
335 blkreq
->nb_sectors
= req
->qiov
.size
/ BDRV_SECTOR_SIZE
;
336 blkreq
->qiov
= &req
->qiov
;
337 blkreq
->cb
= virtio_blk_rw_complete
;
338 blkreq
->opaque
= req
;
344 static void virtio_blk_handle_read(VirtIOBlockReq
*req
)
348 sector
= virtio_ldq_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.sector
);
350 trace_virtio_blk_handle_read(req
, sector
, req
->qiov
.size
/ 512);
352 if (!virtio_blk_sect_range_ok(req
->dev
, sector
, req
->qiov
.size
)) {
353 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
354 virtio_blk_free_request(req
);
358 block_acct_start(blk_get_stats(req
->dev
->blk
), &req
->acct
, req
->qiov
.size
,
360 blk_aio_readv(req
->dev
->blk
, sector
, &req
->qiov
,
361 req
->qiov
.size
/ BDRV_SECTOR_SIZE
,
362 virtio_blk_rw_complete
, req
);
365 void virtio_blk_handle_request(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
368 struct iovec
*in_iov
= req
->elem
.in_sg
;
369 struct iovec
*iov
= req
->elem
.out_sg
;
370 unsigned in_num
= req
->elem
.in_num
;
371 unsigned out_num
= req
->elem
.out_num
;
373 if (req
->elem
.out_num
< 1 || req
->elem
.in_num
< 1) {
374 error_report("virtio-blk missing headers");
378 if (unlikely(iov_to_buf(iov
, out_num
, 0, &req
->out
,
379 sizeof(req
->out
)) != sizeof(req
->out
))) {
380 error_report("virtio-blk request outhdr too short");
384 iov_discard_front(&iov
, &out_num
, sizeof(req
->out
));
387 in_iov
[in_num
- 1].iov_len
< sizeof(struct virtio_blk_inhdr
)) {
388 error_report("virtio-blk request inhdr too short");
392 req
->in
= (void *)in_iov
[in_num
- 1].iov_base
393 + in_iov
[in_num
- 1].iov_len
394 - sizeof(struct virtio_blk_inhdr
);
395 iov_discard_back(in_iov
, &in_num
, sizeof(struct virtio_blk_inhdr
));
397 type
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
399 if (type
& VIRTIO_BLK_T_FLUSH
) {
400 virtio_blk_handle_flush(req
, mrb
);
401 } else if (type
& VIRTIO_BLK_T_SCSI_CMD
) {
402 virtio_blk_handle_scsi(req
);
403 } else if (type
& VIRTIO_BLK_T_GET_ID
) {
404 VirtIOBlock
*s
= req
->dev
;
407 * NB: per existing s/n string convention the string is
408 * terminated by '\0' only when shorter than buffer.
410 const char *serial
= s
->conf
.serial
? s
->conf
.serial
: "";
411 size_t size
= MIN(strlen(serial
) + 1,
412 MIN(iov_size(in_iov
, in_num
),
413 VIRTIO_BLK_ID_BYTES
));
414 iov_from_buf(in_iov
, in_num
, 0, serial
, size
);
415 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
416 virtio_blk_free_request(req
);
417 } else if (type
& VIRTIO_BLK_T_OUT
) {
418 qemu_iovec_init_external(&req
->qiov
, iov
, out_num
);
419 virtio_blk_handle_write(req
, mrb
);
420 } else if (type
== VIRTIO_BLK_T_IN
|| type
== VIRTIO_BLK_T_BARRIER
) {
421 /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
422 qemu_iovec_init_external(&req
->qiov
, in_iov
, in_num
);
423 virtio_blk_handle_read(req
);
425 virtio_blk_req_complete(req
, VIRTIO_BLK_S_UNSUPP
);
426 virtio_blk_free_request(req
);
430 static void virtio_blk_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
432 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
434 MultiReqBuffer mrb
= {
438 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
439 * dataplane here instead of waiting for .set_status().
442 virtio_blk_data_plane_start(s
->dataplane
);
446 while ((req
= virtio_blk_get_request(s
))) {
447 virtio_blk_handle_request(req
, &mrb
);
450 virtio_submit_multiwrite(s
->blk
, &mrb
);
453 * FIXME: Want to check for completions before returning to guest mode,
454 * so cached reads and writes are reported as quickly as possible. But
455 * that should be done in the generic block layer.
459 static void virtio_blk_dma_restart_bh(void *opaque
)
461 VirtIOBlock
*s
= opaque
;
462 VirtIOBlockReq
*req
= s
->rq
;
463 MultiReqBuffer mrb
= {
467 qemu_bh_delete(s
->bh
);
473 VirtIOBlockReq
*next
= req
->next
;
474 virtio_blk_handle_request(req
, &mrb
);
478 virtio_submit_multiwrite(s
->blk
, &mrb
);
481 static void virtio_blk_dma_restart_cb(void *opaque
, int running
,
484 VirtIOBlock
*s
= opaque
;
491 s
->bh
= aio_bh_new(blk_get_aio_context(s
->conf
.conf
.blk
),
492 virtio_blk_dma_restart_bh
, s
);
493 qemu_bh_schedule(s
->bh
);
497 static void virtio_blk_reset(VirtIODevice
*vdev
)
499 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
502 virtio_blk_data_plane_stop(s
->dataplane
);
506 * This should cancel pending requests, but can't do nicely until there
507 * are per-device request lists.
510 blk_set_enable_write_cache(s
->blk
, s
->original_wce
);
513 /* coalesce internal state, copy to pci i/o region 0
515 static void virtio_blk_update_config(VirtIODevice
*vdev
, uint8_t *config
)
517 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
518 BlockConf
*conf
= &s
->conf
.conf
;
519 struct virtio_blk_config blkcfg
;
521 int blk_size
= conf
->logical_block_size
;
523 blk_get_geometry(s
->blk
, &capacity
);
524 memset(&blkcfg
, 0, sizeof(blkcfg
));
525 virtio_stq_p(vdev
, &blkcfg
.capacity
, capacity
);
526 virtio_stl_p(vdev
, &blkcfg
.seg_max
, 128 - 2);
527 virtio_stw_p(vdev
, &blkcfg
.cylinders
, conf
->cyls
);
528 virtio_stl_p(vdev
, &blkcfg
.blk_size
, blk_size
);
529 virtio_stw_p(vdev
, &blkcfg
.min_io_size
, conf
->min_io_size
/ blk_size
);
530 virtio_stw_p(vdev
, &blkcfg
.opt_io_size
, conf
->opt_io_size
/ blk_size
);
531 blkcfg
.heads
= conf
->heads
;
533 * We must ensure that the block device capacity is a multiple of
534 * the logical block size. If that is not the case, let's use
535 * sector_mask to adopt the geometry to have a correct picture.
536 * For those devices where the capacity is ok for the given geometry
537 * we don't touch the sector value of the geometry, since some devices
538 * (like s390 dasd) need a specific value. Here the capacity is already
539 * cyls*heads*secs*blk_size and the sector value is not block size
540 * divided by 512 - instead it is the amount of blk_size blocks
541 * per track (cylinder).
543 if (blk_getlength(s
->blk
) / conf
->heads
/ conf
->secs
% blk_size
) {
544 blkcfg
.sectors
= conf
->secs
& ~s
->sector_mask
;
546 blkcfg
.sectors
= conf
->secs
;
549 blkcfg
.physical_block_exp
= get_physical_block_exp(conf
);
550 blkcfg
.alignment_offset
= 0;
551 blkcfg
.wce
= blk_enable_write_cache(s
->blk
);
552 memcpy(config
, &blkcfg
, sizeof(struct virtio_blk_config
));
555 static void virtio_blk_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
557 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
558 struct virtio_blk_config blkcfg
;
560 memcpy(&blkcfg
, config
, sizeof(blkcfg
));
562 aio_context_acquire(blk_get_aio_context(s
->blk
));
563 blk_set_enable_write_cache(s
->blk
, blkcfg
.wce
!= 0);
564 aio_context_release(blk_get_aio_context(s
->blk
));
567 static uint32_t virtio_blk_get_features(VirtIODevice
*vdev
, uint32_t features
)
569 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
571 features
|= (1 << VIRTIO_BLK_F_SEG_MAX
);
572 features
|= (1 << VIRTIO_BLK_F_GEOMETRY
);
573 features
|= (1 << VIRTIO_BLK_F_TOPOLOGY
);
574 features
|= (1 << VIRTIO_BLK_F_BLK_SIZE
);
575 features
|= (1 << VIRTIO_BLK_F_SCSI
);
577 if (s
->conf
.config_wce
) {
578 features
|= (1 << VIRTIO_BLK_F_CONFIG_WCE
);
580 if (blk_enable_write_cache(s
->blk
)) {
581 features
|= (1 << VIRTIO_BLK_F_WCE
);
583 if (blk_is_read_only(s
->blk
)) {
584 features
|= 1 << VIRTIO_BLK_F_RO
;
590 static void virtio_blk_set_status(VirtIODevice
*vdev
, uint8_t status
)
592 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
595 if (s
->dataplane
&& !(status
& (VIRTIO_CONFIG_S_DRIVER
|
596 VIRTIO_CONFIG_S_DRIVER_OK
))) {
597 virtio_blk_data_plane_stop(s
->dataplane
);
600 if (!(status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
604 features
= vdev
->guest_features
;
606 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
607 * cache flushes. Thus, the "auto writethrough" behavior is never
608 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
609 * Leaving it enabled would break the following sequence:
611 * Guest started with "-drive cache=writethrough"
612 * Guest sets status to 0
613 * Guest sets DRIVER bit in status field
614 * Guest reads host features (WCE=0, CONFIG_WCE=1)
615 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
616 * Guest writes 1 to the WCE configuration field (writeback mode)
617 * Guest sets DRIVER_OK bit in status field
619 * s->blk would erroneously be placed in writethrough mode.
621 if (!(features
& (1 << VIRTIO_BLK_F_CONFIG_WCE
))) {
622 aio_context_acquire(blk_get_aio_context(s
->blk
));
623 blk_set_enable_write_cache(s
->blk
,
624 !!(features
& (1 << VIRTIO_BLK_F_WCE
)));
625 aio_context_release(blk_get_aio_context(s
->blk
));
629 static void virtio_blk_save(QEMUFile
*f
, void *opaque
)
631 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
633 virtio_save(vdev
, f
);
636 static void virtio_blk_save_device(VirtIODevice
*vdev
, QEMUFile
*f
)
638 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
639 VirtIOBlockReq
*req
= s
->rq
;
642 qemu_put_sbyte(f
, 1);
643 qemu_put_buffer(f
, (unsigned char *)&req
->elem
,
644 sizeof(VirtQueueElement
));
647 qemu_put_sbyte(f
, 0);
650 static int virtio_blk_load(QEMUFile
*f
, void *opaque
, int version_id
)
652 VirtIOBlock
*s
= opaque
;
653 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
658 return virtio_load(vdev
, f
, version_id
);
661 static int virtio_blk_load_device(VirtIODevice
*vdev
, QEMUFile
*f
,
664 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
666 while (qemu_get_sbyte(f
)) {
667 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
668 qemu_get_buffer(f
, (unsigned char *)&req
->elem
,
669 sizeof(VirtQueueElement
));
673 virtqueue_map_sg(req
->elem
.in_sg
, req
->elem
.in_addr
,
674 req
->elem
.in_num
, 1);
675 virtqueue_map_sg(req
->elem
.out_sg
, req
->elem
.out_addr
,
676 req
->elem
.out_num
, 0);
682 static void virtio_blk_resize(void *opaque
)
684 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
686 virtio_notify_config(vdev
);
689 static const BlockDevOps virtio_block_ops
= {
690 .resize_cb
= virtio_blk_resize
,
693 /* Disable dataplane thread during live migration since it does not
694 * update the dirty memory bitmap yet.
696 static void virtio_blk_migration_state_changed(Notifier
*notifier
, void *data
)
698 VirtIOBlock
*s
= container_of(notifier
, VirtIOBlock
,
699 migration_state_notifier
);
700 MigrationState
*mig
= data
;
703 if (migration_in_setup(mig
)) {
707 virtio_blk_data_plane_destroy(s
->dataplane
);
709 } else if (migration_has_finished(mig
) ||
710 migration_has_failed(mig
)) {
714 blk_drain_all(); /* complete in-flight non-dataplane requests */
715 virtio_blk_data_plane_create(VIRTIO_DEVICE(s
), &s
->conf
,
716 &s
->dataplane
, &err
);
718 error_report("%s", error_get_pretty(err
));
724 static void virtio_blk_device_realize(DeviceState
*dev
, Error
**errp
)
726 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
727 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
728 VirtIOBlkConf
*conf
= &s
->conf
;
730 static int virtio_blk_id
;
732 if (!conf
->conf
.blk
) {
733 error_setg(errp
, "drive property not set");
736 if (!blk_is_inserted(conf
->conf
.blk
)) {
737 error_setg(errp
, "Device needs media, but drive is empty");
741 blkconf_serial(&conf
->conf
, &conf
->serial
);
742 s
->original_wce
= blk_enable_write_cache(conf
->conf
.blk
);
743 blkconf_geometry(&conf
->conf
, NULL
, 65535, 255, 255, &err
);
745 error_propagate(errp
, err
);
749 virtio_init(vdev
, "virtio-blk", VIRTIO_ID_BLOCK
,
750 sizeof(struct virtio_blk_config
));
752 s
->blk
= conf
->conf
.blk
;
754 s
->sector_mask
= (s
->conf
.conf
.logical_block_size
/ BDRV_SECTOR_SIZE
) - 1;
756 s
->vq
= virtio_add_queue(vdev
, 128, virtio_blk_handle_output
);
757 s
->complete_request
= virtio_blk_complete_request
;
758 virtio_blk_data_plane_create(vdev
, conf
, &s
->dataplane
, &err
);
760 error_propagate(errp
, err
);
761 virtio_cleanup(vdev
);
764 s
->migration_state_notifier
.notify
= virtio_blk_migration_state_changed
;
765 add_migration_state_change_notifier(&s
->migration_state_notifier
);
767 s
->change
= qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb
, s
);
768 register_savevm(dev
, "virtio-blk", virtio_blk_id
++, 2,
769 virtio_blk_save
, virtio_blk_load
, s
);
770 blk_set_dev_ops(s
->blk
, &virtio_block_ops
, s
);
771 blk_set_guest_block_size(s
->blk
, s
->conf
.conf
.logical_block_size
);
773 blk_iostatus_enable(s
->blk
);
776 static void virtio_blk_device_unrealize(DeviceState
*dev
, Error
**errp
)
778 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
779 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
781 remove_migration_state_change_notifier(&s
->migration_state_notifier
);
782 virtio_blk_data_plane_destroy(s
->dataplane
);
784 qemu_del_vm_change_state_handler(s
->change
);
785 unregister_savevm(dev
, "virtio-blk", s
);
786 blockdev_mark_auto_del(s
->blk
);
787 virtio_cleanup(vdev
);
790 static void virtio_blk_instance_init(Object
*obj
)
792 VirtIOBlock
*s
= VIRTIO_BLK(obj
);
794 object_property_add_link(obj
, "iothread", TYPE_IOTHREAD
,
795 (Object
**)&s
->conf
.iothread
,
796 qdev_prop_allow_set_link_before_realize
,
797 OBJ_PROP_LINK_UNREF_ON_RELEASE
, NULL
);
798 device_add_bootindex_property(obj
, &s
->conf
.conf
.bootindex
,
799 "bootindex", "/disk@0,0",
803 static Property virtio_blk_properties
[] = {
804 DEFINE_BLOCK_PROPERTIES(VirtIOBlock
, conf
.conf
),
805 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock
, conf
.conf
),
806 DEFINE_PROP_STRING("serial", VirtIOBlock
, conf
.serial
),
807 DEFINE_PROP_BIT("config-wce", VirtIOBlock
, conf
.config_wce
, 0, true),
809 DEFINE_PROP_BIT("scsi", VirtIOBlock
, conf
.scsi
, 0, true),
811 DEFINE_PROP_BIT("x-data-plane", VirtIOBlock
, conf
.data_plane
, 0, false),
812 DEFINE_PROP_END_OF_LIST(),
815 static void virtio_blk_class_init(ObjectClass
*klass
, void *data
)
817 DeviceClass
*dc
= DEVICE_CLASS(klass
);
818 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
820 dc
->props
= virtio_blk_properties
;
821 set_bit(DEVICE_CATEGORY_STORAGE
, dc
->categories
);
822 vdc
->realize
= virtio_blk_device_realize
;
823 vdc
->unrealize
= virtio_blk_device_unrealize
;
824 vdc
->get_config
= virtio_blk_update_config
;
825 vdc
->set_config
= virtio_blk_set_config
;
826 vdc
->get_features
= virtio_blk_get_features
;
827 vdc
->set_status
= virtio_blk_set_status
;
828 vdc
->reset
= virtio_blk_reset
;
829 vdc
->save
= virtio_blk_save_device
;
830 vdc
->load
= virtio_blk_load_device
;
833 static const TypeInfo virtio_device_info
= {
834 .name
= TYPE_VIRTIO_BLK
,
835 .parent
= TYPE_VIRTIO_DEVICE
,
836 .instance_size
= sizeof(VirtIOBlock
),
837 .instance_init
= virtio_blk_instance_init
,
838 .class_init
= virtio_blk_class_init
,
841 static void virtio_register_types(void)
843 type_register_static(&virtio_device_info
);
846 type_init(virtio_register_types
)