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
);
42 void virtio_blk_free_request(VirtIOBlockReq
*req
)
45 g_slice_free(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
) {
77 } else if (action
== BLOCK_ERROR_ACTION_REPORT
) {
78 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
79 block_acct_done(blk_get_stats(s
->blk
), &req
->acct
);
80 virtio_blk_free_request(req
);
83 blk_error_action(s
->blk
, action
, is_read
, error
);
84 return action
!= BLOCK_ERROR_ACTION_IGNORE
;
87 static void virtio_blk_rw_complete(void *opaque
, int ret
)
89 VirtIOBlockReq
*next
= opaque
;
92 VirtIOBlockReq
*req
= next
;
94 trace_virtio_blk_rw_complete(req
, ret
);
96 if (req
->qiov
.nalloc
!= -1) {
97 /* If nalloc is != 1 req->qiov is a local copy of the original
98 * external iovec. It was allocated in submit_merged_requests
99 * to be able to merge requests. */
100 qemu_iovec_destroy(&req
->qiov
);
104 int p
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
105 bool is_read
= !(p
& VIRTIO_BLK_T_OUT
);
106 /* Note that memory may be dirtied on read failure. If the
107 * virtio request is not completed here, as is the case for
108 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
109 * correctly during live migration. While this is ugly,
110 * it is acceptable because the device is free to write to
111 * the memory until the request is completed (which will
112 * happen on the other side of the migration).
114 if (virtio_blk_handle_rw_error(req
, -ret
, is_read
)) {
119 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
120 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
121 virtio_blk_free_request(req
);
125 static void virtio_blk_flush_complete(void *opaque
, int ret
)
127 VirtIOBlockReq
*req
= opaque
;
130 if (virtio_blk_handle_rw_error(req
, -ret
, 0)) {
135 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
136 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
137 virtio_blk_free_request(req
);
144 struct sg_io_hdr hdr
;
145 } VirtIOBlockIoctlReq
;
147 static void virtio_blk_ioctl_complete(void *opaque
, int status
)
149 VirtIOBlockIoctlReq
*ioctl_req
= opaque
;
150 VirtIOBlockReq
*req
= ioctl_req
->req
;
151 VirtIODevice
*vdev
= VIRTIO_DEVICE(req
->dev
);
152 struct virtio_scsi_inhdr
*scsi
;
153 struct sg_io_hdr
*hdr
;
155 scsi
= (void *)req
->elem
.in_sg
[req
->elem
.in_num
- 2].iov_base
;
158 status
= VIRTIO_BLK_S_UNSUPP
;
159 virtio_stl_p(vdev
, &scsi
->errors
, 255);
163 hdr
= &ioctl_req
->hdr
;
165 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
166 * clear the masked_status field [hence status gets cleared too, see
167 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
168 * status has occurred. However they do set DRIVER_SENSE in driver_status
169 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
171 if (hdr
->status
== 0 && hdr
->sb_len_wr
> 0) {
172 hdr
->status
= CHECK_CONDITION
;
175 virtio_stl_p(vdev
, &scsi
->errors
,
176 hdr
->status
| (hdr
->msg_status
<< 8) |
177 (hdr
->host_status
<< 16) | (hdr
->driver_status
<< 24));
178 virtio_stl_p(vdev
, &scsi
->residual
, hdr
->resid
);
179 virtio_stl_p(vdev
, &scsi
->sense_len
, hdr
->sb_len_wr
);
180 virtio_stl_p(vdev
, &scsi
->data_len
, hdr
->dxfer_len
);
183 virtio_blk_req_complete(req
, status
);
184 virtio_blk_free_request(req
);
190 static VirtIOBlockReq
*virtio_blk_get_request(VirtIOBlock
*s
)
192 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
194 if (!virtqueue_pop(s
->vq
, &req
->elem
)) {
195 virtio_blk_free_request(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 int nb_sectors
= mrb
->reqs
[start
]->qiov
.size
/ BDRV_SECTOR_SIZE
;
328 bool is_write
= mrb
->is_write
;
332 struct iovec
*tmp_iov
= qiov
->iov
;
333 int tmp_niov
= qiov
->niov
;
335 /* mrb->reqs[start]->qiov was initialized from external so we can't
336 * modifiy it here. We need to initialize it locally and then add the
337 * external iovecs. */
338 qemu_iovec_init(qiov
, niov
);
340 for (i
= 0; i
< tmp_niov
; i
++) {
341 qemu_iovec_add(qiov
, tmp_iov
[i
].iov_base
, tmp_iov
[i
].iov_len
);
344 for (i
= start
+ 1; i
< start
+ num_reqs
; i
++) {
345 qemu_iovec_concat(qiov
, &mrb
->reqs
[i
]->qiov
, 0,
346 mrb
->reqs
[i
]->qiov
.size
);
347 mrb
->reqs
[i
- 1]->mr_next
= mrb
->reqs
[i
];
348 nb_sectors
+= mrb
->reqs
[i
]->qiov
.size
/ BDRV_SECTOR_SIZE
;
350 assert(nb_sectors
== qiov
->size
/ BDRV_SECTOR_SIZE
);
352 trace_virtio_blk_submit_multireq(mrb
, start
, num_reqs
, sector_num
,
353 nb_sectors
, is_write
);
354 block_acct_merge_done(blk_get_stats(blk
),
355 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
,
360 blk_aio_writev(blk
, sector_num
, qiov
, nb_sectors
,
361 virtio_blk_rw_complete
, mrb
->reqs
[start
]);
363 blk_aio_readv(blk
, sector_num
, qiov
, nb_sectors
,
364 virtio_blk_rw_complete
, mrb
->reqs
[start
]);
368 static int multireq_compare(const void *a
, const void *b
)
370 const VirtIOBlockReq
*req1
= *(VirtIOBlockReq
**)a
,
371 *req2
= *(VirtIOBlockReq
**)b
;
374 * Note that we can't simply subtract sector_num1 from sector_num2
375 * here as that could overflow the return value.
377 if (req1
->sector_num
> req2
->sector_num
) {
379 } else if (req1
->sector_num
< req2
->sector_num
) {
386 void virtio_blk_submit_multireq(BlockBackend
*blk
, MultiReqBuffer
*mrb
)
388 int i
= 0, start
= 0, num_reqs
= 0, niov
= 0, nb_sectors
= 0;
389 int max_xfer_len
= 0;
390 int64_t sector_num
= 0;
392 if (mrb
->num_reqs
== 1) {
393 submit_requests(blk
, mrb
, 0, 1, -1);
398 max_xfer_len
= blk_get_max_transfer_length(mrb
->reqs
[0]->dev
->blk
);
399 max_xfer_len
= MIN_NON_ZERO(max_xfer_len
, BDRV_REQUEST_MAX_SECTORS
);
401 qsort(mrb
->reqs
, mrb
->num_reqs
, sizeof(*mrb
->reqs
),
404 for (i
= 0; i
< mrb
->num_reqs
; i
++) {
405 VirtIOBlockReq
*req
= mrb
->reqs
[i
];
409 /* merge would exceed maximum number of IOVs */
410 if (niov
+ req
->qiov
.niov
> IOV_MAX
) {
414 /* merge would exceed maximum transfer length of backend device */
415 if (req
->qiov
.size
/ BDRV_SECTOR_SIZE
+ nb_sectors
> max_xfer_len
) {
419 /* requests are not sequential */
420 if (sector_num
+ nb_sectors
!= req
->sector_num
) {
425 submit_requests(blk
, mrb
, start
, num_reqs
, niov
);
431 sector_num
= req
->sector_num
;
432 nb_sectors
= niov
= 0;
436 nb_sectors
+= req
->qiov
.size
/ BDRV_SECTOR_SIZE
;
437 niov
+= req
->qiov
.niov
;
441 submit_requests(blk
, mrb
, start
, num_reqs
, niov
);
445 static void virtio_blk_handle_flush(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
447 block_acct_start(blk_get_stats(req
->dev
->blk
), &req
->acct
, 0,
451 * Make sure all outstanding writes are posted to the backing device.
453 if (mrb
->is_write
&& mrb
->num_reqs
> 0) {
454 virtio_blk_submit_multireq(req
->dev
->blk
, mrb
);
456 blk_aio_flush(req
->dev
->blk
, virtio_blk_flush_complete
, req
);
459 static bool virtio_blk_sect_range_ok(VirtIOBlock
*dev
,
460 uint64_t sector
, size_t size
)
462 uint64_t nb_sectors
= size
>> BDRV_SECTOR_BITS
;
463 uint64_t total_sectors
;
465 if (nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
468 if (sector
& dev
->sector_mask
) {
471 if (size
% dev
->conf
.conf
.logical_block_size
) {
474 blk_get_geometry(dev
->blk
, &total_sectors
);
475 if (sector
> total_sectors
|| nb_sectors
> total_sectors
- sector
) {
481 void virtio_blk_handle_request(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
484 struct iovec
*in_iov
= req
->elem
.in_sg
;
485 struct iovec
*iov
= req
->elem
.out_sg
;
486 unsigned in_num
= req
->elem
.in_num
;
487 unsigned out_num
= req
->elem
.out_num
;
489 if (req
->elem
.out_num
< 1 || req
->elem
.in_num
< 1) {
490 error_report("virtio-blk missing headers");
494 if (unlikely(iov_to_buf(iov
, out_num
, 0, &req
->out
,
495 sizeof(req
->out
)) != sizeof(req
->out
))) {
496 error_report("virtio-blk request outhdr too short");
500 iov_discard_front(&iov
, &out_num
, sizeof(req
->out
));
503 in_iov
[in_num
- 1].iov_len
< sizeof(struct virtio_blk_inhdr
)) {
504 error_report("virtio-blk request inhdr too short");
508 /* We always touch the last byte, so just see how big in_iov is. */
509 req
->in_len
= iov_size(in_iov
, in_num
);
510 req
->in
= (void *)in_iov
[in_num
- 1].iov_base
511 + in_iov
[in_num
- 1].iov_len
512 - sizeof(struct virtio_blk_inhdr
);
513 iov_discard_back(in_iov
, &in_num
, sizeof(struct virtio_blk_inhdr
));
515 type
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
517 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
518 * is an optional flag. Although a guest should not send this flag if
519 * not negotiated we ignored it in the past. So keep ignoring it. */
520 switch (type
& ~(VIRTIO_BLK_T_OUT
| VIRTIO_BLK_T_BARRIER
)) {
521 case VIRTIO_BLK_T_IN
:
523 bool is_write
= type
& VIRTIO_BLK_T_OUT
;
524 req
->sector_num
= virtio_ldq_p(VIRTIO_DEVICE(req
->dev
),
528 qemu_iovec_init_external(&req
->qiov
, iov
, out_num
);
529 trace_virtio_blk_handle_write(req
, req
->sector_num
,
530 req
->qiov
.size
/ BDRV_SECTOR_SIZE
);
532 qemu_iovec_init_external(&req
->qiov
, in_iov
, in_num
);
533 trace_virtio_blk_handle_read(req
, req
->sector_num
,
534 req
->qiov
.size
/ BDRV_SECTOR_SIZE
);
537 if (!virtio_blk_sect_range_ok(req
->dev
, req
->sector_num
,
539 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
540 virtio_blk_free_request(req
);
544 block_acct_start(blk_get_stats(req
->dev
->blk
),
545 &req
->acct
, req
->qiov
.size
,
546 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
);
548 /* merge would exceed maximum number of requests or IO direction
550 if (mrb
->num_reqs
> 0 && (mrb
->num_reqs
== VIRTIO_BLK_MAX_MERGE_REQS
||
551 is_write
!= mrb
->is_write
||
552 !req
->dev
->conf
.request_merging
)) {
553 virtio_blk_submit_multireq(req
->dev
->blk
, mrb
);
556 assert(mrb
->num_reqs
< VIRTIO_BLK_MAX_MERGE_REQS
);
557 mrb
->reqs
[mrb
->num_reqs
++] = req
;
558 mrb
->is_write
= is_write
;
561 case VIRTIO_BLK_T_FLUSH
:
562 virtio_blk_handle_flush(req
, mrb
);
564 case VIRTIO_BLK_T_SCSI_CMD
:
565 virtio_blk_handle_scsi(req
);
567 case VIRTIO_BLK_T_GET_ID
:
569 VirtIOBlock
*s
= req
->dev
;
572 * NB: per existing s/n string convention the string is
573 * terminated by '\0' only when shorter than buffer.
575 const char *serial
= s
->conf
.serial
? s
->conf
.serial
: "";
576 size_t size
= MIN(strlen(serial
) + 1,
577 MIN(iov_size(in_iov
, in_num
),
578 VIRTIO_BLK_ID_BYTES
));
579 iov_from_buf(in_iov
, in_num
, 0, serial
, size
);
580 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
581 virtio_blk_free_request(req
);
585 virtio_blk_req_complete(req
, VIRTIO_BLK_S_UNSUPP
);
586 virtio_blk_free_request(req
);
590 static void virtio_blk_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
592 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
594 MultiReqBuffer mrb
= {};
596 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
597 * dataplane here instead of waiting for .set_status().
600 virtio_blk_data_plane_start(s
->dataplane
);
604 while ((req
= virtio_blk_get_request(s
))) {
605 virtio_blk_handle_request(req
, &mrb
);
609 virtio_blk_submit_multireq(s
->blk
, &mrb
);
613 static void virtio_blk_dma_restart_bh(void *opaque
)
615 VirtIOBlock
*s
= opaque
;
616 VirtIOBlockReq
*req
= s
->rq
;
617 MultiReqBuffer mrb
= {};
619 qemu_bh_delete(s
->bh
);
625 VirtIOBlockReq
*next
= req
->next
;
626 virtio_blk_handle_request(req
, &mrb
);
631 virtio_blk_submit_multireq(s
->blk
, &mrb
);
635 static void virtio_blk_dma_restart_cb(void *opaque
, int running
,
638 VirtIOBlock
*s
= opaque
;
645 s
->bh
= aio_bh_new(blk_get_aio_context(s
->conf
.conf
.blk
),
646 virtio_blk_dma_restart_bh
, s
);
647 qemu_bh_schedule(s
->bh
);
651 static void virtio_blk_reset(VirtIODevice
*vdev
)
653 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
656 virtio_blk_data_plane_stop(s
->dataplane
);
660 * This should cancel pending requests, but can't do nicely until there
661 * are per-device request lists.
664 blk_set_enable_write_cache(s
->blk
, s
->original_wce
);
667 /* coalesce internal state, copy to pci i/o region 0
669 static void virtio_blk_update_config(VirtIODevice
*vdev
, uint8_t *config
)
671 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
672 BlockConf
*conf
= &s
->conf
.conf
;
673 struct virtio_blk_config blkcfg
;
675 int blk_size
= conf
->logical_block_size
;
677 blk_get_geometry(s
->blk
, &capacity
);
678 memset(&blkcfg
, 0, sizeof(blkcfg
));
679 virtio_stq_p(vdev
, &blkcfg
.capacity
, capacity
);
680 virtio_stl_p(vdev
, &blkcfg
.seg_max
, 128 - 2);
681 virtio_stw_p(vdev
, &blkcfg
.geometry
.cylinders
, conf
->cyls
);
682 virtio_stl_p(vdev
, &blkcfg
.blk_size
, blk_size
);
683 virtio_stw_p(vdev
, &blkcfg
.min_io_size
, conf
->min_io_size
/ blk_size
);
684 virtio_stw_p(vdev
, &blkcfg
.opt_io_size
, conf
->opt_io_size
/ blk_size
);
685 blkcfg
.geometry
.heads
= conf
->heads
;
687 * We must ensure that the block device capacity is a multiple of
688 * the logical block size. If that is not the case, let's use
689 * sector_mask to adopt the geometry to have a correct picture.
690 * For those devices where the capacity is ok for the given geometry
691 * we don't touch the sector value of the geometry, since some devices
692 * (like s390 dasd) need a specific value. Here the capacity is already
693 * cyls*heads*secs*blk_size and the sector value is not block size
694 * divided by 512 - instead it is the amount of blk_size blocks
695 * per track (cylinder).
697 if (blk_getlength(s
->blk
) / conf
->heads
/ conf
->secs
% blk_size
) {
698 blkcfg
.geometry
.sectors
= conf
->secs
& ~s
->sector_mask
;
700 blkcfg
.geometry
.sectors
= conf
->secs
;
703 blkcfg
.physical_block_exp
= get_physical_block_exp(conf
);
704 blkcfg
.alignment_offset
= 0;
705 blkcfg
.wce
= blk_enable_write_cache(s
->blk
);
706 memcpy(config
, &blkcfg
, sizeof(struct virtio_blk_config
));
709 static void virtio_blk_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
711 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
712 struct virtio_blk_config blkcfg
;
714 memcpy(&blkcfg
, config
, sizeof(blkcfg
));
716 aio_context_acquire(blk_get_aio_context(s
->blk
));
717 blk_set_enable_write_cache(s
->blk
, blkcfg
.wce
!= 0);
718 aio_context_release(blk_get_aio_context(s
->blk
));
721 static uint64_t virtio_blk_get_features(VirtIODevice
*vdev
, uint64_t features
)
723 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
725 virtio_add_feature(&features
, VIRTIO_BLK_F_SEG_MAX
);
726 virtio_add_feature(&features
, VIRTIO_BLK_F_GEOMETRY
);
727 virtio_add_feature(&features
, VIRTIO_BLK_F_TOPOLOGY
);
728 virtio_add_feature(&features
, VIRTIO_BLK_F_BLK_SIZE
);
729 virtio_add_feature(&features
, VIRTIO_BLK_F_SCSI
);
731 if (s
->conf
.config_wce
) {
732 virtio_add_feature(&features
, VIRTIO_BLK_F_CONFIG_WCE
);
734 if (blk_enable_write_cache(s
->blk
)) {
735 virtio_add_feature(&features
, VIRTIO_BLK_F_WCE
);
737 if (blk_is_read_only(s
->blk
)) {
738 virtio_add_feature(&features
, VIRTIO_BLK_F_RO
);
744 static void virtio_blk_set_status(VirtIODevice
*vdev
, uint8_t status
)
746 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
748 if (s
->dataplane
&& !(status
& (VIRTIO_CONFIG_S_DRIVER
|
749 VIRTIO_CONFIG_S_DRIVER_OK
))) {
750 virtio_blk_data_plane_stop(s
->dataplane
);
753 if (!(status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
757 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
758 * cache flushes. Thus, the "auto writethrough" behavior is never
759 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
760 * Leaving it enabled would break the following sequence:
762 * Guest started with "-drive cache=writethrough"
763 * Guest sets status to 0
764 * Guest sets DRIVER bit in status field
765 * Guest reads host features (WCE=0, CONFIG_WCE=1)
766 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
767 * Guest writes 1 to the WCE configuration field (writeback mode)
768 * Guest sets DRIVER_OK bit in status field
770 * s->blk would erroneously be placed in writethrough mode.
772 if (!virtio_has_feature(vdev
, VIRTIO_BLK_F_CONFIG_WCE
)) {
773 aio_context_acquire(blk_get_aio_context(s
->blk
));
774 blk_set_enable_write_cache(s
->blk
,
775 virtio_has_feature(vdev
, VIRTIO_BLK_F_WCE
));
776 aio_context_release(blk_get_aio_context(s
->blk
));
780 static void virtio_blk_save(QEMUFile
*f
, void *opaque
)
782 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
784 virtio_save(vdev
, f
);
787 static void virtio_blk_save_device(VirtIODevice
*vdev
, QEMUFile
*f
)
789 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
790 VirtIOBlockReq
*req
= s
->rq
;
793 qemu_put_sbyte(f
, 1);
794 qemu_put_buffer(f
, (unsigned char *)&req
->elem
,
795 sizeof(VirtQueueElement
));
798 qemu_put_sbyte(f
, 0);
801 static int virtio_blk_load(QEMUFile
*f
, void *opaque
, int version_id
)
803 VirtIOBlock
*s
= opaque
;
804 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
809 return virtio_load(vdev
, f
, version_id
);
812 static int virtio_blk_load_device(VirtIODevice
*vdev
, QEMUFile
*f
,
815 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
817 while (qemu_get_sbyte(f
)) {
818 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
819 qemu_get_buffer(f
, (unsigned char *)&req
->elem
,
820 sizeof(VirtQueueElement
));
824 virtqueue_map_sg(req
->elem
.in_sg
, req
->elem
.in_addr
,
825 req
->elem
.in_num
, 1);
826 virtqueue_map_sg(req
->elem
.out_sg
, req
->elem
.out_addr
,
827 req
->elem
.out_num
, 0);
833 static void virtio_blk_resize(void *opaque
)
835 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
837 virtio_notify_config(vdev
);
840 static const BlockDevOps virtio_block_ops
= {
841 .resize_cb
= virtio_blk_resize
,
844 /* Disable dataplane thread during live migration since it does not
845 * update the dirty memory bitmap yet.
847 static void virtio_blk_migration_state_changed(Notifier
*notifier
, void *data
)
849 VirtIOBlock
*s
= container_of(notifier
, VirtIOBlock
,
850 migration_state_notifier
);
851 MigrationState
*mig
= data
;
854 if (migration_in_setup(mig
)) {
858 virtio_blk_data_plane_destroy(s
->dataplane
);
860 } else if (migration_has_finished(mig
) ||
861 migration_has_failed(mig
)) {
865 blk_drain_all(); /* complete in-flight non-dataplane requests */
866 virtio_blk_data_plane_create(VIRTIO_DEVICE(s
), &s
->conf
,
867 &s
->dataplane
, &err
);
869 error_report_err(err
);
874 static void virtio_blk_device_realize(DeviceState
*dev
, Error
**errp
)
876 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
877 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
878 VirtIOBlkConf
*conf
= &s
->conf
;
880 static int virtio_blk_id
;
882 if (!conf
->conf
.blk
) {
883 error_setg(errp
, "drive property not set");
886 if (!blk_is_inserted(conf
->conf
.blk
)) {
887 error_setg(errp
, "Device needs media, but drive is empty");
891 blkconf_serial(&conf
->conf
, &conf
->serial
);
892 s
->original_wce
= blk_enable_write_cache(conf
->conf
.blk
);
893 blkconf_geometry(&conf
->conf
, NULL
, 65535, 255, 255, &err
);
895 error_propagate(errp
, err
);
898 blkconf_blocksizes(&conf
->conf
);
900 virtio_init(vdev
, "virtio-blk", VIRTIO_ID_BLOCK
,
901 sizeof(struct virtio_blk_config
));
903 s
->blk
= conf
->conf
.blk
;
905 s
->sector_mask
= (s
->conf
.conf
.logical_block_size
/ BDRV_SECTOR_SIZE
) - 1;
907 s
->vq
= virtio_add_queue(vdev
, 128, virtio_blk_handle_output
);
908 s
->complete_request
= virtio_blk_complete_request
;
909 virtio_blk_data_plane_create(vdev
, conf
, &s
->dataplane
, &err
);
911 error_propagate(errp
, err
);
912 virtio_cleanup(vdev
);
915 s
->migration_state_notifier
.notify
= virtio_blk_migration_state_changed
;
916 add_migration_state_change_notifier(&s
->migration_state_notifier
);
918 s
->change
= qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb
, s
);
919 register_savevm(dev
, "virtio-blk", virtio_blk_id
++, 2,
920 virtio_blk_save
, virtio_blk_load
, s
);
921 blk_set_dev_ops(s
->blk
, &virtio_block_ops
, s
);
922 blk_set_guest_block_size(s
->blk
, s
->conf
.conf
.logical_block_size
);
924 blk_iostatus_enable(s
->blk
);
927 static void virtio_blk_device_unrealize(DeviceState
*dev
, Error
**errp
)
929 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
930 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
932 remove_migration_state_change_notifier(&s
->migration_state_notifier
);
933 virtio_blk_data_plane_destroy(s
->dataplane
);
935 qemu_del_vm_change_state_handler(s
->change
);
936 unregister_savevm(dev
, "virtio-blk", s
);
937 blockdev_mark_auto_del(s
->blk
);
938 virtio_cleanup(vdev
);
941 static void virtio_blk_instance_init(Object
*obj
)
943 VirtIOBlock
*s
= VIRTIO_BLK(obj
);
945 object_property_add_link(obj
, "iothread", TYPE_IOTHREAD
,
946 (Object
**)&s
->conf
.iothread
,
947 qdev_prop_allow_set_link_before_realize
,
948 OBJ_PROP_LINK_UNREF_ON_RELEASE
, NULL
);
949 device_add_bootindex_property(obj
, &s
->conf
.conf
.bootindex
,
950 "bootindex", "/disk@0,0",
954 static Property virtio_blk_properties
[] = {
955 DEFINE_BLOCK_PROPERTIES(VirtIOBlock
, conf
.conf
),
956 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock
, conf
.conf
),
957 DEFINE_PROP_STRING("serial", VirtIOBlock
, conf
.serial
),
958 DEFINE_PROP_BIT("config-wce", VirtIOBlock
, conf
.config_wce
, 0, true),
960 DEFINE_PROP_BIT("scsi", VirtIOBlock
, conf
.scsi
, 0, true),
962 DEFINE_PROP_BIT("request-merging", VirtIOBlock
, conf
.request_merging
, 0,
964 DEFINE_PROP_BIT("x-data-plane", VirtIOBlock
, conf
.data_plane
, 0, false),
965 DEFINE_PROP_END_OF_LIST(),
968 static void virtio_blk_class_init(ObjectClass
*klass
, void *data
)
970 DeviceClass
*dc
= DEVICE_CLASS(klass
);
971 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
973 dc
->props
= virtio_blk_properties
;
974 set_bit(DEVICE_CATEGORY_STORAGE
, dc
->categories
);
975 vdc
->realize
= virtio_blk_device_realize
;
976 vdc
->unrealize
= virtio_blk_device_unrealize
;
977 vdc
->get_config
= virtio_blk_update_config
;
978 vdc
->set_config
= virtio_blk_set_config
;
979 vdc
->get_features
= virtio_blk_get_features
;
980 vdc
->set_status
= virtio_blk_set_status
;
981 vdc
->reset
= virtio_blk_reset
;
982 vdc
->save
= virtio_blk_save_device
;
983 vdc
->load
= virtio_blk_load_device
;
986 static const TypeInfo virtio_device_info
= {
987 .name
= TYPE_VIRTIO_BLK
,
988 .parent
= TYPE_VIRTIO_DEVICE
,
989 .instance_size
= sizeof(VirtIOBlock
),
990 .instance_init
= virtio_blk_instance_init
,
991 .class_init
= virtio_blk_class_init
,
994 static void virtio_register_types(void)
996 type_register_static(&virtio_device_info
);
999 type_init(virtio_register_types
)