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/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/blockdev.h"
22 #include "hw/virtio/virtio-blk.h"
23 #include "dataplane/virtio-blk.h"
24 #include "scsi/constants.h"
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/virtio/virtio-access.h"
31 /* We don't support discard yet, hide associated config fields. */
32 #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \
35 static void virtio_blk_init_request(VirtIOBlock
*s
, VirtQueue
*vq
,
46 static void virtio_blk_free_request(VirtIOBlockReq
*req
)
51 static void virtio_blk_req_complete(VirtIOBlockReq
*req
, unsigned char status
)
53 VirtIOBlock
*s
= req
->dev
;
54 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
56 trace_virtio_blk_req_complete(vdev
, req
, status
);
58 stb_p(&req
->in
->status
, status
);
59 virtqueue_push(req
->vq
, &req
->elem
, req
->in_len
);
60 if (s
->dataplane_started
&& !s
->dataplane_disabled
) {
61 virtio_blk_data_plane_notify(s
->dataplane
, req
->vq
);
63 virtio_notify(vdev
, req
->vq
);
67 static int virtio_blk_handle_rw_error(VirtIOBlockReq
*req
, int error
,
70 VirtIOBlock
*s
= req
->dev
;
71 BlockErrorAction action
= blk_get_error_action(s
->blk
, is_read
, error
);
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
;
92 VirtIOBlock
*s
= next
->dev
;
93 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
95 aio_context_acquire(blk_get_aio_context(s
->conf
.conf
.blk
));
97 VirtIOBlockReq
*req
= next
;
99 trace_virtio_blk_rw_complete(vdev
, req
, ret
);
101 if (req
->qiov
.nalloc
!= -1) {
102 /* If nalloc is != -1 req->qiov is a local copy of the original
103 * external iovec. It was allocated in submit_requests to be
104 * able to merge requests. */
105 qemu_iovec_destroy(&req
->qiov
);
109 int p
= virtio_ldl_p(VIRTIO_DEVICE(req
->dev
), &req
->out
.type
);
110 bool is_read
= !(p
& VIRTIO_BLK_T_OUT
);
111 /* Note that memory may be dirtied on read failure. If the
112 * virtio request is not completed here, as is the case for
113 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
114 * correctly during live migration. While this is ugly,
115 * it is acceptable because the device is free to write to
116 * the memory until the request is completed (which will
117 * happen on the other side of the migration).
119 if (virtio_blk_handle_rw_error(req
, -ret
, is_read
)) {
124 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
125 block_acct_done(blk_get_stats(req
->dev
->blk
), &req
->acct
);
126 virtio_blk_free_request(req
);
128 aio_context_release(blk_get_aio_context(s
->conf
.conf
.blk
));
131 static void virtio_blk_flush_complete(void *opaque
, int ret
)
133 VirtIOBlockReq
*req
= opaque
;
134 VirtIOBlock
*s
= req
->dev
;
136 aio_context_acquire(blk_get_aio_context(s
->conf
.conf
.blk
));
138 if (virtio_blk_handle_rw_error(req
, -ret
, 0)) {
143 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
144 block_acct_done(blk_get_stats(s
->blk
), &req
->acct
);
145 virtio_blk_free_request(req
);
148 aio_context_release(blk_get_aio_context(s
->conf
.conf
.blk
));
155 struct sg_io_hdr hdr
;
156 } VirtIOBlockIoctlReq
;
158 static void virtio_blk_ioctl_complete(void *opaque
, int status
)
160 VirtIOBlockIoctlReq
*ioctl_req
= opaque
;
161 VirtIOBlockReq
*req
= ioctl_req
->req
;
162 VirtIOBlock
*s
= req
->dev
;
163 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
164 struct virtio_scsi_inhdr
*scsi
;
165 struct sg_io_hdr
*hdr
;
167 scsi
= (void *)req
->elem
.in_sg
[req
->elem
.in_num
- 2].iov_base
;
170 status
= VIRTIO_BLK_S_UNSUPP
;
171 virtio_stl_p(vdev
, &scsi
->errors
, 255);
175 hdr
= &ioctl_req
->hdr
;
177 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
178 * clear the masked_status field [hence status gets cleared too, see
179 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
180 * status has occurred. However they do set DRIVER_SENSE in driver_status
181 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
183 if (hdr
->status
== 0 && hdr
->sb_len_wr
> 0) {
184 hdr
->status
= CHECK_CONDITION
;
187 virtio_stl_p(vdev
, &scsi
->errors
,
188 hdr
->status
| (hdr
->msg_status
<< 8) |
189 (hdr
->host_status
<< 16) | (hdr
->driver_status
<< 24));
190 virtio_stl_p(vdev
, &scsi
->residual
, hdr
->resid
);
191 virtio_stl_p(vdev
, &scsi
->sense_len
, hdr
->sb_len_wr
);
192 virtio_stl_p(vdev
, &scsi
->data_len
, hdr
->dxfer_len
);
195 aio_context_acquire(blk_get_aio_context(s
->conf
.conf
.blk
));
196 virtio_blk_req_complete(req
, status
);
197 virtio_blk_free_request(req
);
198 aio_context_release(blk_get_aio_context(s
->conf
.conf
.blk
));
204 static VirtIOBlockReq
*virtio_blk_get_request(VirtIOBlock
*s
, VirtQueue
*vq
)
206 VirtIOBlockReq
*req
= virtqueue_pop(vq
, sizeof(VirtIOBlockReq
));
209 virtio_blk_init_request(s
, vq
, req
);
214 static int virtio_blk_handle_scsi_req(VirtIOBlockReq
*req
)
216 int status
= VIRTIO_BLK_S_OK
;
217 struct virtio_scsi_inhdr
*scsi
= NULL
;
218 VirtIODevice
*vdev
= VIRTIO_DEVICE(req
->dev
);
219 VirtQueueElement
*elem
= &req
->elem
;
220 VirtIOBlock
*blk
= req
->dev
;
224 VirtIOBlockIoctlReq
*ioctl_req
;
229 * We require at least one output segment each for the virtio_blk_outhdr
230 * and the SCSI command block.
232 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
233 * and the sense buffer pointer in the input segments.
235 if (elem
->out_num
< 2 || elem
->in_num
< 3) {
236 status
= VIRTIO_BLK_S_IOERR
;
241 * The scsi inhdr is placed in the second-to-last input segment, just
242 * before the regular inhdr.
244 scsi
= (void *)elem
->in_sg
[elem
->in_num
- 2].iov_base
;
246 if (!blk
->conf
.scsi
) {
247 status
= VIRTIO_BLK_S_UNSUPP
;
252 * No support for bidirection commands yet.
254 if (elem
->out_num
> 2 && elem
->in_num
> 3) {
255 status
= VIRTIO_BLK_S_UNSUPP
;
260 ioctl_req
= g_new0(VirtIOBlockIoctlReq
, 1);
261 ioctl_req
->req
= req
;
262 ioctl_req
->hdr
.interface_id
= 'S';
263 ioctl_req
->hdr
.cmd_len
= elem
->out_sg
[1].iov_len
;
264 ioctl_req
->hdr
.cmdp
= elem
->out_sg
[1].iov_base
;
265 ioctl_req
->hdr
.dxfer_len
= 0;
267 if (elem
->out_num
> 2) {
269 * If there are more than the minimally required 2 output segments
270 * there is write payload starting from the third iovec.
272 ioctl_req
->hdr
.dxfer_direction
= SG_DXFER_TO_DEV
;
273 ioctl_req
->hdr
.iovec_count
= elem
->out_num
- 2;
275 for (i
= 0; i
< ioctl_req
->hdr
.iovec_count
; i
++) {
276 ioctl_req
->hdr
.dxfer_len
+= elem
->out_sg
[i
+ 2].iov_len
;
279 ioctl_req
->hdr
.dxferp
= elem
->out_sg
+ 2;
281 } else if (elem
->in_num
> 3) {
283 * If we have more than 3 input segments the guest wants to actually
286 ioctl_req
->hdr
.dxfer_direction
= SG_DXFER_FROM_DEV
;
287 ioctl_req
->hdr
.iovec_count
= elem
->in_num
- 3;
288 for (i
= 0; i
< ioctl_req
->hdr
.iovec_count
; i
++) {
289 ioctl_req
->hdr
.dxfer_len
+= elem
->in_sg
[i
].iov_len
;
292 ioctl_req
->hdr
.dxferp
= elem
->in_sg
;
295 * Some SCSI commands don't actually transfer any data.
297 ioctl_req
->hdr
.dxfer_direction
= SG_DXFER_NONE
;
300 ioctl_req
->hdr
.sbp
= elem
->in_sg
[elem
->in_num
- 3].iov_base
;
301 ioctl_req
->hdr
.mx_sb_len
= elem
->in_sg
[elem
->in_num
- 3].iov_len
;
303 acb
= blk_aio_ioctl(blk
->blk
, SG_IO
, &ioctl_req
->hdr
,
304 virtio_blk_ioctl_complete
, ioctl_req
);
307 status
= VIRTIO_BLK_S_UNSUPP
;
316 /* Just put anything nonzero so that the ioctl fails in the guest. */
318 virtio_stl_p(vdev
, &scsi
->errors
, 255);
323 static void virtio_blk_handle_scsi(VirtIOBlockReq
*req
)
327 status
= virtio_blk_handle_scsi_req(req
);
328 if (status
!= -EINPROGRESS
) {
329 virtio_blk_req_complete(req
, status
);
330 virtio_blk_free_request(req
);
334 static inline void submit_requests(BlockBackend
*blk
, MultiReqBuffer
*mrb
,
335 int start
, int num_reqs
, int niov
)
337 QEMUIOVector
*qiov
= &mrb
->reqs
[start
]->qiov
;
338 int64_t sector_num
= mrb
->reqs
[start
]->sector_num
;
339 bool is_write
= mrb
->is_write
;
343 struct iovec
*tmp_iov
= qiov
->iov
;
344 int tmp_niov
= qiov
->niov
;
346 /* mrb->reqs[start]->qiov was initialized from external so we can't
347 * modify it here. We need to initialize it locally and then add the
348 * external iovecs. */
349 qemu_iovec_init(qiov
, niov
);
351 for (i
= 0; i
< tmp_niov
; i
++) {
352 qemu_iovec_add(qiov
, tmp_iov
[i
].iov_base
, tmp_iov
[i
].iov_len
);
355 for (i
= start
+ 1; i
< start
+ num_reqs
; i
++) {
356 qemu_iovec_concat(qiov
, &mrb
->reqs
[i
]->qiov
, 0,
357 mrb
->reqs
[i
]->qiov
.size
);
358 mrb
->reqs
[i
- 1]->mr_next
= mrb
->reqs
[i
];
361 trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb
->reqs
[start
]->dev
),
362 mrb
, start
, num_reqs
,
363 sector_num
<< BDRV_SECTOR_BITS
,
364 qiov
->size
, is_write
);
365 block_acct_merge_done(blk_get_stats(blk
),
366 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
,
371 blk_aio_pwritev(blk
, sector_num
<< BDRV_SECTOR_BITS
, qiov
, 0,
372 virtio_blk_rw_complete
, mrb
->reqs
[start
]);
374 blk_aio_preadv(blk
, sector_num
<< BDRV_SECTOR_BITS
, qiov
, 0,
375 virtio_blk_rw_complete
, mrb
->reqs
[start
]);
379 static int multireq_compare(const void *a
, const void *b
)
381 const VirtIOBlockReq
*req1
= *(VirtIOBlockReq
**)a
,
382 *req2
= *(VirtIOBlockReq
**)b
;
385 * Note that we can't simply subtract sector_num1 from sector_num2
386 * here as that could overflow the return value.
388 if (req1
->sector_num
> req2
->sector_num
) {
390 } else if (req1
->sector_num
< req2
->sector_num
) {
397 static void virtio_blk_submit_multireq(BlockBackend
*blk
, MultiReqBuffer
*mrb
)
399 int i
= 0, start
= 0, num_reqs
= 0, niov
= 0, nb_sectors
= 0;
400 uint32_t max_transfer
;
401 int64_t sector_num
= 0;
403 if (mrb
->num_reqs
== 1) {
404 submit_requests(blk
, mrb
, 0, 1, -1);
409 max_transfer
= blk_get_max_transfer(mrb
->reqs
[0]->dev
->blk
);
411 qsort(mrb
->reqs
, mrb
->num_reqs
, sizeof(*mrb
->reqs
),
414 for (i
= 0; i
< mrb
->num_reqs
; i
++) {
415 VirtIOBlockReq
*req
= mrb
->reqs
[i
];
418 * NOTE: We cannot merge the requests in below situations:
419 * 1. requests are not sequential
420 * 2. merge would exceed maximum number of IOVs
421 * 3. merge would exceed maximum transfer length of backend device
423 if (sector_num
+ nb_sectors
!= req
->sector_num
||
424 niov
> blk_get_max_iov(blk
) - req
->qiov
.niov
||
425 req
->qiov
.size
> max_transfer
||
426 nb_sectors
> (max_transfer
-
427 req
->qiov
.size
) / BDRV_SECTOR_SIZE
) {
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 static int virtio_blk_handle_request(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
487 struct iovec
*in_iov
= req
->elem
.in_sg
;
488 struct iovec
*out_iov
= req
->elem
.out_sg
;
489 unsigned in_num
= req
->elem
.in_num
;
490 unsigned out_num
= req
->elem
.out_num
;
491 VirtIOBlock
*s
= req
->dev
;
492 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
494 if (req
->elem
.out_num
< 1 || req
->elem
.in_num
< 1) {
495 virtio_error(vdev
, "virtio-blk missing headers");
499 if (unlikely(iov_to_buf(out_iov
, out_num
, 0, &req
->out
,
500 sizeof(req
->out
)) != sizeof(req
->out
))) {
501 virtio_error(vdev
, "virtio-blk request outhdr too short");
505 iov_discard_front(&out_iov
, &out_num
, sizeof(req
->out
));
507 if (in_iov
[in_num
- 1].iov_len
< sizeof(struct virtio_blk_inhdr
)) {
508 virtio_error(vdev
, "virtio-blk request inhdr too short");
512 /* We always touch the last byte, so just see how big in_iov is. */
513 req
->in_len
= iov_size(in_iov
, in_num
);
514 req
->in
= (void *)in_iov
[in_num
- 1].iov_base
515 + in_iov
[in_num
- 1].iov_len
516 - sizeof(struct virtio_blk_inhdr
);
517 iov_discard_back(in_iov
, &in_num
, sizeof(struct virtio_blk_inhdr
));
519 type
= virtio_ldl_p(vdev
, &req
->out
.type
);
521 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
522 * is an optional flag. Although a guest should not send this flag if
523 * not negotiated we ignored it in the past. So keep ignoring it. */
524 switch (type
& ~(VIRTIO_BLK_T_OUT
| VIRTIO_BLK_T_BARRIER
)) {
525 case VIRTIO_BLK_T_IN
:
527 bool is_write
= type
& VIRTIO_BLK_T_OUT
;
528 req
->sector_num
= virtio_ldq_p(vdev
, &req
->out
.sector
);
531 qemu_iovec_init_external(&req
->qiov
, out_iov
, out_num
);
532 trace_virtio_blk_handle_write(vdev
, req
, req
->sector_num
,
533 req
->qiov
.size
/ BDRV_SECTOR_SIZE
);
535 qemu_iovec_init_external(&req
->qiov
, in_iov
, in_num
);
536 trace_virtio_blk_handle_read(vdev
, req
, req
->sector_num
,
537 req
->qiov
.size
/ BDRV_SECTOR_SIZE
);
540 if (!virtio_blk_sect_range_ok(s
, req
->sector_num
, req
->qiov
.size
)) {
541 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
542 block_acct_invalid(blk_get_stats(s
->blk
),
543 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
);
544 virtio_blk_free_request(req
);
548 block_acct_start(blk_get_stats(s
->blk
), &req
->acct
, req
->qiov
.size
,
549 is_write
? BLOCK_ACCT_WRITE
: BLOCK_ACCT_READ
);
551 /* merge would exceed maximum number of requests or IO direction
553 if (mrb
->num_reqs
> 0 && (mrb
->num_reqs
== VIRTIO_BLK_MAX_MERGE_REQS
||
554 is_write
!= mrb
->is_write
||
555 !s
->conf
.request_merging
)) {
556 virtio_blk_submit_multireq(s
->blk
, mrb
);
559 assert(mrb
->num_reqs
< VIRTIO_BLK_MAX_MERGE_REQS
);
560 mrb
->reqs
[mrb
->num_reqs
++] = req
;
561 mrb
->is_write
= is_write
;
564 case VIRTIO_BLK_T_FLUSH
:
565 virtio_blk_handle_flush(req
, mrb
);
567 case VIRTIO_BLK_T_SCSI_CMD
:
568 virtio_blk_handle_scsi(req
);
570 case VIRTIO_BLK_T_GET_ID
:
573 * NB: per existing s/n string convention the string is
574 * terminated by '\0' only when shorter than buffer.
576 const char *serial
= s
->conf
.serial
? s
->conf
.serial
: "";
577 size_t size
= MIN(strlen(serial
) + 1,
578 MIN(iov_size(in_iov
, in_num
),
579 VIRTIO_BLK_ID_BYTES
));
580 iov_from_buf(in_iov
, in_num
, 0, serial
, size
);
581 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
582 virtio_blk_free_request(req
);
586 virtio_blk_req_complete(req
, VIRTIO_BLK_S_UNSUPP
);
587 virtio_blk_free_request(req
);
592 bool virtio_blk_handle_vq(VirtIOBlock
*s
, VirtQueue
*vq
)
595 MultiReqBuffer mrb
= {};
596 bool progress
= false;
598 aio_context_acquire(blk_get_aio_context(s
->blk
));
602 virtio_queue_set_notification(vq
, 0);
604 while ((req
= virtio_blk_get_request(s
, vq
))) {
606 if (virtio_blk_handle_request(req
, &mrb
)) {
607 virtqueue_detach_element(req
->vq
, &req
->elem
, 0);
608 virtio_blk_free_request(req
);
613 virtio_queue_set_notification(vq
, 1);
614 } while (!virtio_queue_empty(vq
));
617 virtio_blk_submit_multireq(s
->blk
, &mrb
);
620 blk_io_unplug(s
->blk
);
621 aio_context_release(blk_get_aio_context(s
->blk
));
625 static void virtio_blk_handle_output_do(VirtIOBlock
*s
, VirtQueue
*vq
)
627 virtio_blk_handle_vq(s
, vq
);
630 static void virtio_blk_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
632 VirtIOBlock
*s
= (VirtIOBlock
*)vdev
;
635 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
636 * dataplane here instead of waiting for .set_status().
638 virtio_device_start_ioeventfd(vdev
);
639 if (!s
->dataplane_disabled
) {
643 virtio_blk_handle_output_do(s
, vq
);
646 static void virtio_blk_dma_restart_bh(void *opaque
)
648 VirtIOBlock
*s
= opaque
;
649 VirtIOBlockReq
*req
= s
->rq
;
650 MultiReqBuffer mrb
= {};
652 qemu_bh_delete(s
->bh
);
657 aio_context_acquire(blk_get_aio_context(s
->conf
.conf
.blk
));
659 VirtIOBlockReq
*next
= req
->next
;
660 if (virtio_blk_handle_request(req
, &mrb
)) {
661 /* Device is now broken and won't do any processing until it gets
662 * reset. Already queued requests will be lost: let's purge them.
666 virtqueue_detach_element(req
->vq
, &req
->elem
, 0);
667 virtio_blk_free_request(req
);
676 virtio_blk_submit_multireq(s
->blk
, &mrb
);
678 blk_dec_in_flight(s
->conf
.conf
.blk
);
679 aio_context_release(blk_get_aio_context(s
->conf
.conf
.blk
));
682 static void virtio_blk_dma_restart_cb(void *opaque
, int running
,
685 VirtIOBlock
*s
= opaque
;
692 /* FIXME The data plane is not started yet, so these requests are
693 * processed in the main thread. */
694 s
->bh
= aio_bh_new(blk_get_aio_context(s
->conf
.conf
.blk
),
695 virtio_blk_dma_restart_bh
, s
);
696 blk_inc_in_flight(s
->conf
.conf
.blk
);
697 qemu_bh_schedule(s
->bh
);
701 static void virtio_blk_reset(VirtIODevice
*vdev
)
703 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
707 ctx
= blk_get_aio_context(s
->blk
);
708 aio_context_acquire(ctx
);
711 /* We drop queued requests after blk_drain() because blk_drain() itself can
716 virtqueue_detach_element(req
->vq
, &req
->elem
, 0);
717 virtio_blk_free_request(req
);
720 aio_context_release(ctx
);
722 assert(!s
->dataplane_started
);
723 blk_set_enable_write_cache(s
->blk
, s
->original_wce
);
726 /* coalesce internal state, copy to pci i/o region 0
728 static void virtio_blk_update_config(VirtIODevice
*vdev
, uint8_t *config
)
730 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
731 BlockConf
*conf
= &s
->conf
.conf
;
732 struct virtio_blk_config blkcfg
;
735 int blk_size
= conf
->logical_block_size
;
737 blk_get_geometry(s
->blk
, &capacity
);
738 memset(&blkcfg
, 0, sizeof(blkcfg
));
739 virtio_stq_p(vdev
, &blkcfg
.capacity
, capacity
);
740 virtio_stl_p(vdev
, &blkcfg
.seg_max
, 128 - 2);
741 virtio_stw_p(vdev
, &blkcfg
.geometry
.cylinders
, conf
->cyls
);
742 virtio_stl_p(vdev
, &blkcfg
.blk_size
, blk_size
);
743 virtio_stw_p(vdev
, &blkcfg
.min_io_size
, conf
->min_io_size
/ blk_size
);
744 virtio_stw_p(vdev
, &blkcfg
.opt_io_size
, conf
->opt_io_size
/ blk_size
);
745 blkcfg
.geometry
.heads
= conf
->heads
;
747 * We must ensure that the block device capacity is a multiple of
748 * the logical block size. If that is not the case, let's use
749 * sector_mask to adopt the geometry to have a correct picture.
750 * For those devices where the capacity is ok for the given geometry
751 * we don't touch the sector value of the geometry, since some devices
752 * (like s390 dasd) need a specific value. Here the capacity is already
753 * cyls*heads*secs*blk_size and the sector value is not block size
754 * divided by 512 - instead it is the amount of blk_size blocks
755 * per track (cylinder).
757 length
= blk_getlength(s
->blk
);
758 if (length
> 0 && length
/ conf
->heads
/ conf
->secs
% blk_size
) {
759 blkcfg
.geometry
.sectors
= conf
->secs
& ~s
->sector_mask
;
761 blkcfg
.geometry
.sectors
= conf
->secs
;
764 blkcfg
.physical_block_exp
= get_physical_block_exp(conf
);
765 blkcfg
.alignment_offset
= 0;
766 blkcfg
.wce
= blk_enable_write_cache(s
->blk
);
767 virtio_stw_p(vdev
, &blkcfg
.num_queues
, s
->conf
.num_queues
);
768 memcpy(config
, &blkcfg
, VIRTIO_BLK_CFG_SIZE
);
769 QEMU_BUILD_BUG_ON(VIRTIO_BLK_CFG_SIZE
> sizeof(blkcfg
));
772 static void virtio_blk_set_config(VirtIODevice
*vdev
, const uint8_t *config
)
774 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
775 struct virtio_blk_config blkcfg
;
777 memcpy(&blkcfg
, config
, VIRTIO_BLK_CFG_SIZE
);
778 QEMU_BUILD_BUG_ON(VIRTIO_BLK_CFG_SIZE
> sizeof(blkcfg
));
780 aio_context_acquire(blk_get_aio_context(s
->blk
));
781 blk_set_enable_write_cache(s
->blk
, blkcfg
.wce
!= 0);
782 aio_context_release(blk_get_aio_context(s
->blk
));
785 static uint64_t virtio_blk_get_features(VirtIODevice
*vdev
, uint64_t features
,
788 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
790 virtio_add_feature(&features
, VIRTIO_BLK_F_SEG_MAX
);
791 virtio_add_feature(&features
, VIRTIO_BLK_F_GEOMETRY
);
792 virtio_add_feature(&features
, VIRTIO_BLK_F_TOPOLOGY
);
793 virtio_add_feature(&features
, VIRTIO_BLK_F_BLK_SIZE
);
794 if (virtio_has_feature(features
, VIRTIO_F_VERSION_1
)) {
796 error_setg(errp
, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
800 virtio_clear_feature(&features
, VIRTIO_F_ANY_LAYOUT
);
801 virtio_add_feature(&features
, VIRTIO_BLK_F_SCSI
);
804 if (s
->conf
.config_wce
) {
805 virtio_add_feature(&features
, VIRTIO_BLK_F_CONFIG_WCE
);
807 if (blk_enable_write_cache(s
->blk
)) {
808 virtio_add_feature(&features
, VIRTIO_BLK_F_WCE
);
810 if (blk_is_read_only(s
->blk
)) {
811 virtio_add_feature(&features
, VIRTIO_BLK_F_RO
);
813 if (s
->conf
.num_queues
> 1) {
814 virtio_add_feature(&features
, VIRTIO_BLK_F_MQ
);
820 static void virtio_blk_set_status(VirtIODevice
*vdev
, uint8_t status
)
822 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
824 if (!(status
& (VIRTIO_CONFIG_S_DRIVER
| VIRTIO_CONFIG_S_DRIVER_OK
))) {
825 assert(!s
->dataplane_started
);
828 if (!(status
& VIRTIO_CONFIG_S_DRIVER_OK
)) {
832 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
833 * cache flushes. Thus, the "auto writethrough" behavior is never
834 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
835 * Leaving it enabled would break the following sequence:
837 * Guest started with "-drive cache=writethrough"
838 * Guest sets status to 0
839 * Guest sets DRIVER bit in status field
840 * Guest reads host features (WCE=0, CONFIG_WCE=1)
841 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
842 * Guest writes 1 to the WCE configuration field (writeback mode)
843 * Guest sets DRIVER_OK bit in status field
845 * s->blk would erroneously be placed in writethrough mode.
847 if (!virtio_vdev_has_feature(vdev
, VIRTIO_BLK_F_CONFIG_WCE
)) {
848 aio_context_acquire(blk_get_aio_context(s
->blk
));
849 blk_set_enable_write_cache(s
->blk
,
850 virtio_vdev_has_feature(vdev
,
852 aio_context_release(blk_get_aio_context(s
->blk
));
856 static void virtio_blk_save_device(VirtIODevice
*vdev
, QEMUFile
*f
)
858 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
859 VirtIOBlockReq
*req
= s
->rq
;
862 qemu_put_sbyte(f
, 1);
864 if (s
->conf
.num_queues
> 1) {
865 qemu_put_be32(f
, virtio_get_queue_index(req
->vq
));
868 qemu_put_virtqueue_element(f
, &req
->elem
);
871 qemu_put_sbyte(f
, 0);
874 static int virtio_blk_load_device(VirtIODevice
*vdev
, QEMUFile
*f
,
877 VirtIOBlock
*s
= VIRTIO_BLK(vdev
);
879 while (qemu_get_sbyte(f
)) {
880 unsigned nvqs
= s
->conf
.num_queues
;
885 vq_idx
= qemu_get_be32(f
);
887 if (vq_idx
>= nvqs
) {
888 error_report("Invalid virtqueue index in request list: %#x",
894 req
= qemu_get_virtqueue_element(vdev
, f
, sizeof(VirtIOBlockReq
));
895 virtio_blk_init_request(s
, virtio_get_queue(vdev
, vq_idx
), req
);
903 static void virtio_blk_resize(void *opaque
)
905 VirtIODevice
*vdev
= VIRTIO_DEVICE(opaque
);
907 virtio_notify_config(vdev
);
910 static const BlockDevOps virtio_block_ops
= {
911 .resize_cb
= virtio_blk_resize
,
914 static void virtio_blk_device_realize(DeviceState
*dev
, Error
**errp
)
916 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
917 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
918 VirtIOBlkConf
*conf
= &s
->conf
;
922 if (!conf
->conf
.blk
) {
923 error_setg(errp
, "drive property not set");
926 if (!blk_is_inserted(conf
->conf
.blk
)) {
927 error_setg(errp
, "Device needs media, but drive is empty");
930 if (!conf
->num_queues
) {
931 error_setg(errp
, "num-queues property must be larger than 0");
934 if (!is_power_of_2(conf
->queue_size
) ||
935 conf
->queue_size
> VIRTQUEUE_MAX_SIZE
) {
936 error_setg(errp
, "invalid queue-size property (%" PRIu16
"), "
937 "must be a power of 2 (max %d)",
938 conf
->queue_size
, VIRTQUEUE_MAX_SIZE
);
942 if (!blkconf_apply_backend_options(&conf
->conf
,
943 blk_is_read_only(conf
->conf
.blk
), true,
947 s
->original_wce
= blk_enable_write_cache(conf
->conf
.blk
);
948 if (!blkconf_geometry(&conf
->conf
, NULL
, 65535, 255, 255, errp
)) {
952 blkconf_blocksizes(&conf
->conf
);
954 if (conf
->conf
.logical_block_size
>
955 conf
->conf
.physical_block_size
) {
957 "logical_block_size > physical_block_size not supported");
961 virtio_init(vdev
, "virtio-blk", VIRTIO_ID_BLOCK
, VIRTIO_BLK_CFG_SIZE
);
963 s
->blk
= conf
->conf
.blk
;
965 s
->sector_mask
= (s
->conf
.conf
.logical_block_size
/ BDRV_SECTOR_SIZE
) - 1;
967 for (i
= 0; i
< conf
->num_queues
; i
++) {
968 virtio_add_queue(vdev
, conf
->queue_size
, virtio_blk_handle_output
);
970 virtio_blk_data_plane_create(vdev
, conf
, &s
->dataplane
, &err
);
972 error_propagate(errp
, err
);
973 virtio_cleanup(vdev
);
977 s
->change
= qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb
, s
);
978 blk_set_dev_ops(s
->blk
, &virtio_block_ops
, s
);
979 blk_set_guest_block_size(s
->blk
, s
->conf
.conf
.logical_block_size
);
981 blk_iostatus_enable(s
->blk
);
984 static void virtio_blk_device_unrealize(DeviceState
*dev
, Error
**errp
)
986 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
987 VirtIOBlock
*s
= VIRTIO_BLK(dev
);
989 virtio_blk_data_plane_destroy(s
->dataplane
);
991 qemu_del_vm_change_state_handler(s
->change
);
992 blockdev_mark_auto_del(s
->blk
);
993 virtio_cleanup(vdev
);
996 static void virtio_blk_instance_init(Object
*obj
)
998 VirtIOBlock
*s
= VIRTIO_BLK(obj
);
1000 device_add_bootindex_property(obj
, &s
->conf
.conf
.bootindex
,
1001 "bootindex", "/disk@0,0",
1005 static const VMStateDescription vmstate_virtio_blk
= {
1006 .name
= "virtio-blk",
1007 .minimum_version_id
= 2,
1009 .fields
= (VMStateField
[]) {
1010 VMSTATE_VIRTIO_DEVICE
,
1011 VMSTATE_END_OF_LIST()
1015 static Property virtio_blk_properties
[] = {
1016 DEFINE_BLOCK_PROPERTIES(VirtIOBlock
, conf
.conf
),
1017 DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock
, conf
.conf
),
1018 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock
, conf
.conf
),
1019 DEFINE_PROP_STRING("serial", VirtIOBlock
, conf
.serial
),
1020 DEFINE_PROP_BIT("config-wce", VirtIOBlock
, conf
.config_wce
, 0, true),
1022 DEFINE_PROP_BIT("scsi", VirtIOBlock
, conf
.scsi
, 0, false),
1024 DEFINE_PROP_BIT("request-merging", VirtIOBlock
, conf
.request_merging
, 0,
1026 DEFINE_PROP_UINT16("num-queues", VirtIOBlock
, conf
.num_queues
, 1),
1027 DEFINE_PROP_UINT16("queue-size", VirtIOBlock
, conf
.queue_size
, 128),
1028 DEFINE_PROP_LINK("iothread", VirtIOBlock
, conf
.iothread
, TYPE_IOTHREAD
,
1030 DEFINE_PROP_END_OF_LIST(),
1033 static void virtio_blk_class_init(ObjectClass
*klass
, void *data
)
1035 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1036 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
1038 dc
->props
= virtio_blk_properties
;
1039 dc
->vmsd
= &vmstate_virtio_blk
;
1040 set_bit(DEVICE_CATEGORY_STORAGE
, dc
->categories
);
1041 vdc
->realize
= virtio_blk_device_realize
;
1042 vdc
->unrealize
= virtio_blk_device_unrealize
;
1043 vdc
->get_config
= virtio_blk_update_config
;
1044 vdc
->set_config
= virtio_blk_set_config
;
1045 vdc
->get_features
= virtio_blk_get_features
;
1046 vdc
->set_status
= virtio_blk_set_status
;
1047 vdc
->reset
= virtio_blk_reset
;
1048 vdc
->save
= virtio_blk_save_device
;
1049 vdc
->load
= virtio_blk_load_device
;
1050 vdc
->start_ioeventfd
= virtio_blk_data_plane_start
;
1051 vdc
->stop_ioeventfd
= virtio_blk_data_plane_stop
;
1054 static const TypeInfo virtio_blk_info
= {
1055 .name
= TYPE_VIRTIO_BLK
,
1056 .parent
= TYPE_VIRTIO_DEVICE
,
1057 .instance_size
= sizeof(VirtIOBlock
),
1058 .instance_init
= virtio_blk_instance_init
,
1059 .class_init
= virtio_blk_class_init
,
1062 static void virtio_register_types(void)
1064 type_register_static(&virtio_blk_info
);
1067 type_init(virtio_register_types
)