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"
15 #include "qemu-error.h"
17 #include "hw/block-common.h"
19 #include "virtio-blk.h"
20 #include "scsi-defs.h"
25 typedef struct VirtIOBlock
34 unsigned short sector_mask
;
38 static VirtIOBlock
*to_virtio_blk(VirtIODevice
*vdev
)
40 return (VirtIOBlock
*)vdev
;
43 typedef struct VirtIOBlockReq
46 VirtQueueElement elem
;
47 struct virtio_blk_inhdr
*in
;
48 struct virtio_blk_outhdr
*out
;
49 struct virtio_scsi_inhdr
*scsi
;
51 struct VirtIOBlockReq
*next
;
55 static void virtio_blk_req_complete(VirtIOBlockReq
*req
, int status
)
57 VirtIOBlock
*s
= req
->dev
;
59 trace_virtio_blk_req_complete(req
, status
);
61 stb_p(&req
->in
->status
, status
);
62 virtqueue_push(s
->vq
, &req
->elem
, req
->qiov
.size
+ sizeof(*req
->in
));
63 virtio_notify(&s
->vdev
, s
->vq
);
66 static int virtio_blk_handle_rw_error(VirtIOBlockReq
*req
, int error
,
69 BlockErrorAction action
= bdrv_get_on_error(req
->dev
->bs
, is_read
);
70 VirtIOBlock
*s
= req
->dev
;
72 if (action
== BLOCK_ERR_IGNORE
) {
73 bdrv_emit_qmp_error_event(s
->bs
, BDRV_ACTION_IGNORE
, is_read
);
77 if ((error
== ENOSPC
&& action
== BLOCK_ERR_STOP_ENOSPC
)
78 || action
== BLOCK_ERR_STOP_ANY
) {
81 bdrv_emit_qmp_error_event(s
->bs
, BDRV_ACTION_STOP
, is_read
);
82 vm_stop(RUN_STATE_IO_ERROR
);
83 bdrv_iostatus_set_err(s
->bs
, error
);
85 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
86 bdrv_acct_done(s
->bs
, &req
->acct
);
88 bdrv_emit_qmp_error_event(s
->bs
, BDRV_ACTION_REPORT
, is_read
);
94 static void virtio_blk_rw_complete(void *opaque
, int ret
)
96 VirtIOBlockReq
*req
= opaque
;
98 trace_virtio_blk_rw_complete(req
, ret
);
101 int is_read
= !(ldl_p(&req
->out
->type
) & VIRTIO_BLK_T_OUT
);
102 if (virtio_blk_handle_rw_error(req
, -ret
, is_read
))
106 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
107 bdrv_acct_done(req
->dev
->bs
, &req
->acct
);
111 static void virtio_blk_flush_complete(void *opaque
, int ret
)
113 VirtIOBlockReq
*req
= opaque
;
116 if (virtio_blk_handle_rw_error(req
, -ret
, 0)) {
121 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
122 bdrv_acct_done(req
->dev
->bs
, &req
->acct
);
126 static VirtIOBlockReq
*virtio_blk_alloc_request(VirtIOBlock
*s
)
128 VirtIOBlockReq
*req
= g_malloc(sizeof(*req
));
135 static VirtIOBlockReq
*virtio_blk_get_request(VirtIOBlock
*s
)
137 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
140 if (!virtqueue_pop(s
->vq
, &req
->elem
)) {
149 static void virtio_blk_handle_scsi(VirtIOBlockReq
*req
)
155 int status
= VIRTIO_BLK_S_OK
;
158 * We require at least one output segment each for the virtio_blk_outhdr
159 * and the SCSI command block.
161 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
162 * and the sense buffer pointer in the input segments.
164 if (req
->elem
.out_num
< 2 || req
->elem
.in_num
< 3) {
165 virtio_blk_req_complete(req
, VIRTIO_BLK_S_IOERR
);
171 * The scsi inhdr is placed in the second-to-last input segment, just
172 * before the regular inhdr.
174 req
->scsi
= (void *)req
->elem
.in_sg
[req
->elem
.in_num
- 2].iov_base
;
176 if (!req
->dev
->blk
->scsi
) {
177 status
= VIRTIO_BLK_S_UNSUPP
;
182 * No support for bidirection commands yet.
184 if (req
->elem
.out_num
> 2 && req
->elem
.in_num
> 3) {
185 status
= VIRTIO_BLK_S_UNSUPP
;
190 struct sg_io_hdr hdr
;
191 memset(&hdr
, 0, sizeof(struct sg_io_hdr
));
192 hdr
.interface_id
= 'S';
193 hdr
.cmd_len
= req
->elem
.out_sg
[1].iov_len
;
194 hdr
.cmdp
= req
->elem
.out_sg
[1].iov_base
;
197 if (req
->elem
.out_num
> 2) {
199 * If there are more than the minimally required 2 output segments
200 * there is write payload starting from the third iovec.
202 hdr
.dxfer_direction
= SG_DXFER_TO_DEV
;
203 hdr
.iovec_count
= req
->elem
.out_num
- 2;
205 for (i
= 0; i
< hdr
.iovec_count
; i
++)
206 hdr
.dxfer_len
+= req
->elem
.out_sg
[i
+ 2].iov_len
;
208 hdr
.dxferp
= req
->elem
.out_sg
+ 2;
210 } else if (req
->elem
.in_num
> 3) {
212 * If we have more than 3 input segments the guest wants to actually
215 hdr
.dxfer_direction
= SG_DXFER_FROM_DEV
;
216 hdr
.iovec_count
= req
->elem
.in_num
- 3;
217 for (i
= 0; i
< hdr
.iovec_count
; i
++)
218 hdr
.dxfer_len
+= req
->elem
.in_sg
[i
].iov_len
;
220 hdr
.dxferp
= req
->elem
.in_sg
;
223 * Some SCSI commands don't actually transfer any data.
225 hdr
.dxfer_direction
= SG_DXFER_NONE
;
228 hdr
.sbp
= req
->elem
.in_sg
[req
->elem
.in_num
- 3].iov_base
;
229 hdr
.mx_sb_len
= req
->elem
.in_sg
[req
->elem
.in_num
- 3].iov_len
;
231 ret
= bdrv_ioctl(req
->dev
->bs
, SG_IO
, &hdr
);
233 status
= VIRTIO_BLK_S_UNSUPP
;
238 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
239 * clear the masked_status field [hence status gets cleared too, see
240 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
241 * status has occurred. However they do set DRIVER_SENSE in driver_status
242 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
244 if (hdr
.status
== 0 && hdr
.sb_len_wr
> 0) {
245 hdr
.status
= CHECK_CONDITION
;
248 stl_p(&req
->scsi
->errors
,
249 hdr
.status
| (hdr
.msg_status
<< 8) |
250 (hdr
.host_status
<< 16) | (hdr
.driver_status
<< 24));
251 stl_p(&req
->scsi
->residual
, hdr
.resid
);
252 stl_p(&req
->scsi
->sense_len
, hdr
.sb_len_wr
);
253 stl_p(&req
->scsi
->data_len
, hdr
.dxfer_len
);
255 virtio_blk_req_complete(req
, status
);
262 /* Just put anything nonzero so that the ioctl fails in the guest. */
263 stl_p(&req
->scsi
->errors
, 255);
264 virtio_blk_req_complete(req
, status
);
268 typedef struct MultiReqBuffer
{
269 BlockRequest blkreq
[32];
270 unsigned int num_writes
;
273 static void virtio_submit_multiwrite(BlockDriverState
*bs
, MultiReqBuffer
*mrb
)
277 if (!mrb
->num_writes
) {
281 ret
= bdrv_aio_multiwrite(bs
, mrb
->blkreq
, mrb
->num_writes
);
283 for (i
= 0; i
< mrb
->num_writes
; i
++) {
284 if (mrb
->blkreq
[i
].error
) {
285 virtio_blk_rw_complete(mrb
->blkreq
[i
].opaque
, -EIO
);
293 static void virtio_blk_handle_flush(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
295 bdrv_acct_start(req
->dev
->bs
, &req
->acct
, 0, BDRV_ACCT_FLUSH
);
298 * Make sure all outstanding writes are posted to the backing device.
300 virtio_submit_multiwrite(req
->dev
->bs
, mrb
);
301 bdrv_aio_flush(req
->dev
->bs
, virtio_blk_flush_complete
, req
);
304 static void virtio_blk_handle_write(VirtIOBlockReq
*req
, MultiReqBuffer
*mrb
)
306 BlockRequest
*blkreq
;
309 sector
= ldq_p(&req
->out
->sector
);
311 bdrv_acct_start(req
->dev
->bs
, &req
->acct
, req
->qiov
.size
, BDRV_ACCT_WRITE
);
313 trace_virtio_blk_handle_write(req
, sector
, req
->qiov
.size
/ 512);
315 if (sector
& req
->dev
->sector_mask
) {
316 virtio_blk_rw_complete(req
, -EIO
);
319 if (req
->qiov
.size
% req
->dev
->conf
->logical_block_size
) {
320 virtio_blk_rw_complete(req
, -EIO
);
324 if (mrb
->num_writes
== 32) {
325 virtio_submit_multiwrite(req
->dev
->bs
, mrb
);
328 blkreq
= &mrb
->blkreq
[mrb
->num_writes
];
329 blkreq
->sector
= sector
;
330 blkreq
->nb_sectors
= req
->qiov
.size
/ BDRV_SECTOR_SIZE
;
331 blkreq
->qiov
= &req
->qiov
;
332 blkreq
->cb
= virtio_blk_rw_complete
;
333 blkreq
->opaque
= req
;
339 static void virtio_blk_handle_read(VirtIOBlockReq
*req
)
343 sector
= ldq_p(&req
->out
->sector
);
345 bdrv_acct_start(req
->dev
->bs
, &req
->acct
, req
->qiov
.size
, BDRV_ACCT_READ
);
347 trace_virtio_blk_handle_read(req
, sector
, req
->qiov
.size
/ 512);
349 if (sector
& req
->dev
->sector_mask
) {
350 virtio_blk_rw_complete(req
, -EIO
);
353 if (req
->qiov
.size
% req
->dev
->conf
->logical_block_size
) {
354 virtio_blk_rw_complete(req
, -EIO
);
357 bdrv_aio_readv(req
->dev
->bs
, sector
, &req
->qiov
,
358 req
->qiov
.size
/ BDRV_SECTOR_SIZE
,
359 virtio_blk_rw_complete
, req
);
362 static void virtio_blk_handle_request(VirtIOBlockReq
*req
,
367 if (req
->elem
.out_num
< 1 || req
->elem
.in_num
< 1) {
368 error_report("virtio-blk missing headers");
372 if (req
->elem
.out_sg
[0].iov_len
< sizeof(*req
->out
) ||
373 req
->elem
.in_sg
[req
->elem
.in_num
- 1].iov_len
< sizeof(*req
->in
)) {
374 error_report("virtio-blk header not in correct element");
378 req
->out
= (void *)req
->elem
.out_sg
[0].iov_base
;
379 req
->in
= (void *)req
->elem
.in_sg
[req
->elem
.in_num
- 1].iov_base
;
381 type
= ldl_p(&req
->out
->type
);
383 if (type
& VIRTIO_BLK_T_FLUSH
) {
384 virtio_blk_handle_flush(req
, mrb
);
385 } else if (type
& VIRTIO_BLK_T_SCSI_CMD
) {
386 virtio_blk_handle_scsi(req
);
387 } else if (type
& VIRTIO_BLK_T_GET_ID
) {
388 VirtIOBlock
*s
= req
->dev
;
391 * NB: per existing s/n string convention the string is
392 * terminated by '\0' only when shorter than buffer.
394 strncpy(req
->elem
.in_sg
[0].iov_base
,
395 s
->blk
->serial
? s
->blk
->serial
: "",
396 MIN(req
->elem
.in_sg
[0].iov_len
, VIRTIO_BLK_ID_BYTES
));
397 virtio_blk_req_complete(req
, VIRTIO_BLK_S_OK
);
399 } else if (type
& VIRTIO_BLK_T_OUT
) {
400 qemu_iovec_init_external(&req
->qiov
, &req
->elem
.out_sg
[1],
401 req
->elem
.out_num
- 1);
402 virtio_blk_handle_write(req
, mrb
);
404 qemu_iovec_init_external(&req
->qiov
, &req
->elem
.in_sg
[0],
405 req
->elem
.in_num
- 1);
406 virtio_blk_handle_read(req
);
410 static void virtio_blk_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
412 VirtIOBlock
*s
= to_virtio_blk(vdev
);
414 MultiReqBuffer mrb
= {
418 while ((req
= virtio_blk_get_request(s
))) {
419 virtio_blk_handle_request(req
, &mrb
);
422 virtio_submit_multiwrite(s
->bs
, &mrb
);
425 * FIXME: Want to check for completions before returning to guest mode,
426 * so cached reads and writes are reported as quickly as possible. But
427 * that should be done in the generic block layer.
431 static void virtio_blk_dma_restart_bh(void *opaque
)
433 VirtIOBlock
*s
= opaque
;
434 VirtIOBlockReq
*req
= s
->rq
;
435 MultiReqBuffer mrb
= {
439 qemu_bh_delete(s
->bh
);
445 virtio_blk_handle_request(req
, &mrb
);
449 virtio_submit_multiwrite(s
->bs
, &mrb
);
452 static void virtio_blk_dma_restart_cb(void *opaque
, int running
,
455 VirtIOBlock
*s
= opaque
;
461 s
->bh
= qemu_bh_new(virtio_blk_dma_restart_bh
, s
);
462 qemu_bh_schedule(s
->bh
);
466 static void virtio_blk_reset(VirtIODevice
*vdev
)
469 * This should cancel pending requests, but can't do nicely until there
470 * are per-device request lists.
475 /* coalesce internal state, copy to pci i/o region 0
477 static void virtio_blk_update_config(VirtIODevice
*vdev
, uint8_t *config
)
479 VirtIOBlock
*s
= to_virtio_blk(vdev
);
480 struct virtio_blk_config blkcfg
;
482 int blk_size
= s
->conf
->logical_block_size
;
484 bdrv_get_geometry(s
->bs
, &capacity
);
485 memset(&blkcfg
, 0, sizeof(blkcfg
));
486 stq_raw(&blkcfg
.capacity
, capacity
);
487 stl_raw(&blkcfg
.seg_max
, 128 - 2);
488 stw_raw(&blkcfg
.cylinders
, s
->conf
->cyls
);
489 stl_raw(&blkcfg
.blk_size
, blk_size
);
490 stw_raw(&blkcfg
.min_io_size
, s
->conf
->min_io_size
/ blk_size
);
491 stw_raw(&blkcfg
.opt_io_size
, s
->conf
->opt_io_size
/ blk_size
);
492 blkcfg
.heads
= s
->conf
->heads
;
494 * We must ensure that the block device capacity is a multiple of
495 * the logical block size. If that is not the case, lets use
496 * sector_mask to adopt the geometry to have a correct picture.
497 * For those devices where the capacity is ok for the given geometry
498 * we dont touch the sector value of the geometry, since some devices
499 * (like s390 dasd) need a specific value. Here the capacity is already
500 * cyls*heads*secs*blk_size and the sector value is not block size
501 * divided by 512 - instead it is the amount of blk_size blocks
502 * per track (cylinder).
504 if (bdrv_getlength(s
->bs
) / s
->conf
->heads
/ s
->conf
->secs
% blk_size
) {
505 blkcfg
.sectors
= s
->conf
->secs
& ~s
->sector_mask
;
507 blkcfg
.sectors
= s
->conf
->secs
;
510 blkcfg
.physical_block_exp
= get_physical_block_exp(s
->conf
);
511 blkcfg
.alignment_offset
= 0;
512 memcpy(config
, &blkcfg
, sizeof(struct virtio_blk_config
));
515 static uint32_t virtio_blk_get_features(VirtIODevice
*vdev
, uint32_t features
)
517 VirtIOBlock
*s
= to_virtio_blk(vdev
);
519 features
|= (1 << VIRTIO_BLK_F_SEG_MAX
);
520 features
|= (1 << VIRTIO_BLK_F_GEOMETRY
);
521 features
|= (1 << VIRTIO_BLK_F_TOPOLOGY
);
522 features
|= (1 << VIRTIO_BLK_F_BLK_SIZE
);
523 features
|= (1 << VIRTIO_BLK_F_SCSI
);
525 if (bdrv_enable_write_cache(s
->bs
))
526 features
|= (1 << VIRTIO_BLK_F_WCACHE
);
528 if (bdrv_is_read_only(s
->bs
))
529 features
|= 1 << VIRTIO_BLK_F_RO
;
534 static void virtio_blk_save(QEMUFile
*f
, void *opaque
)
536 VirtIOBlock
*s
= opaque
;
537 VirtIOBlockReq
*req
= s
->rq
;
539 virtio_save(&s
->vdev
, f
);
542 qemu_put_sbyte(f
, 1);
543 qemu_put_buffer(f
, (unsigned char*)&req
->elem
, sizeof(req
->elem
));
546 qemu_put_sbyte(f
, 0);
549 static int virtio_blk_load(QEMUFile
*f
, void *opaque
, int version_id
)
551 VirtIOBlock
*s
= opaque
;
557 ret
= virtio_load(&s
->vdev
, f
);
562 while (qemu_get_sbyte(f
)) {
563 VirtIOBlockReq
*req
= virtio_blk_alloc_request(s
);
564 qemu_get_buffer(f
, (unsigned char*)&req
->elem
, sizeof(req
->elem
));
568 virtqueue_map_sg(req
->elem
.in_sg
, req
->elem
.in_addr
,
569 req
->elem
.in_num
, 1);
570 virtqueue_map_sg(req
->elem
.out_sg
, req
->elem
.out_addr
,
571 req
->elem
.out_num
, 0);
577 static void virtio_blk_resize(void *opaque
)
579 VirtIOBlock
*s
= opaque
;
581 virtio_notify_config(&s
->vdev
);
584 static const BlockDevOps virtio_block_ops
= {
585 .resize_cb
= virtio_blk_resize
,
588 VirtIODevice
*virtio_blk_init(DeviceState
*dev
, VirtIOBlkConf
*blk
)
591 static int virtio_blk_id
;
594 error_report("drive property not set");
597 if (!bdrv_is_inserted(blk
->conf
.bs
)) {
598 error_report("Device needs media, but drive is empty");
602 blkconf_serial(&blk
->conf
, &blk
->serial
);
603 if (blkconf_geometry(&blk
->conf
, NULL
, 65535, 255, 255) < 0) {
607 s
= (VirtIOBlock
*)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK
,
608 sizeof(struct virtio_blk_config
),
609 sizeof(VirtIOBlock
));
611 s
->vdev
.get_config
= virtio_blk_update_config
;
612 s
->vdev
.get_features
= virtio_blk_get_features
;
613 s
->vdev
.reset
= virtio_blk_reset
;
614 s
->bs
= blk
->conf
.bs
;
615 s
->conf
= &blk
->conf
;
618 s
->sector_mask
= (s
->conf
->logical_block_size
/ BDRV_SECTOR_SIZE
) - 1;
620 s
->vq
= virtio_add_queue(&s
->vdev
, 128, virtio_blk_handle_output
);
622 qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb
, s
);
624 register_savevm(dev
, "virtio-blk", virtio_blk_id
++, 2,
625 virtio_blk_save
, virtio_blk_load
, s
);
626 bdrv_set_dev_ops(s
->bs
, &virtio_block_ops
, s
);
627 bdrv_set_buffer_alignment(s
->bs
, s
->conf
->logical_block_size
);
629 bdrv_iostatus_enable(s
->bs
);
630 add_boot_device_path(s
->conf
->bootindex
, dev
, "/disk@0,0");
635 void virtio_blk_exit(VirtIODevice
*vdev
)
637 VirtIOBlock
*s
= to_virtio_blk(vdev
);
638 unregister_savevm(s
->qdev
, "virtio-blk", s
);
639 blockdev_mark_auto_del(s
->bs
);
640 virtio_cleanup(vdev
);