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-report.h"
17
#include "hw/block-common.h"
18
#include "sysemu/blockdev.h"
19
#include "virtio-blk.h"
20
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
21
#include "hw/dataplane/virtio-blk.h"
23
#include "scsi-defs.h"
28
typedef struct VirtIOBlock
37
unsigned short sector_mask;
39
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
40
VirtIOBlockDataPlane *dataplane;
44
static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
46
return (VirtIOBlock *)vdev;
49
typedef struct VirtIOBlockReq
52
VirtQueueElement elem;
53
struct virtio_blk_inhdr *in;
54
struct virtio_blk_outhdr *out;
55
struct virtio_scsi_inhdr *scsi;
57
struct VirtIOBlockReq *next;
61
static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
63
VirtIOBlock *s = req->dev;
65
trace_virtio_blk_req_complete(req, status);
67
stb_p(&req->in->status, status);
68
virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
69
virtio_notify(&s->vdev, s->vq);
72
static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
75
BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
76
VirtIOBlock *s = req->dev;
78
if (action == BDRV_ACTION_STOP) {
81
} else if (action == BDRV_ACTION_REPORT) {
82
virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
83
bdrv_acct_done(s->bs, &req->acct);
87
bdrv_error_action(s->bs, action, is_read, error);
88
return action != BDRV_ACTION_IGNORE;
91
static void virtio_blk_rw_complete(void *opaque, int ret)
93
VirtIOBlockReq *req = opaque;
95
trace_virtio_blk_rw_complete(req, ret);
98
bool is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
99
if (virtio_blk_handle_rw_error(req, -ret, is_read))
103
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
104
bdrv_acct_done(req->dev->bs, &req->acct);
108
static void virtio_blk_flush_complete(void *opaque, int ret)
110
VirtIOBlockReq *req = opaque;
113
if (virtio_blk_handle_rw_error(req, -ret, 0)) {
118
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
119
bdrv_acct_done(req->dev->bs, &req->acct);
123
static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
125
VirtIOBlockReq *req = g_malloc(sizeof(*req));
132
static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
134
VirtIOBlockReq *req = virtio_blk_alloc_request(s);
137
if (!virtqueue_pop(s->vq, &req->elem)) {
146
static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
152
int status = VIRTIO_BLK_S_OK;
155
* We require at least one output segment each for the virtio_blk_outhdr
156
* and the SCSI command block.
158
* We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
159
* and the sense buffer pointer in the input segments.
161
if (req->elem.out_num < 2 || req->elem.in_num < 3) {
162
virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
168
* The scsi inhdr is placed in the second-to-last input segment, just
169
* before the regular inhdr.
171
req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
173
if (!req->dev->blk->scsi) {
174
status = VIRTIO_BLK_S_UNSUPP;
179
* No support for bidirection commands yet.
181
if (req->elem.out_num > 2 && req->elem.in_num > 3) {
182
status = VIRTIO_BLK_S_UNSUPP;
187
struct sg_io_hdr hdr;
188
memset(&hdr, 0, sizeof(struct sg_io_hdr));
189
hdr.interface_id = 'S';
190
hdr.cmd_len = req->elem.out_sg[1].iov_len;
191
hdr.cmdp = req->elem.out_sg[1].iov_base;
194
if (req->elem.out_num > 2) {
196
* If there are more than the minimally required 2 output segments
197
* there is write payload starting from the third iovec.
199
hdr.dxfer_direction = SG_DXFER_TO_DEV;
200
hdr.iovec_count = req->elem.out_num - 2;
202
for (i = 0; i < hdr.iovec_count; i++)
203
hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
205
hdr.dxferp = req->elem.out_sg + 2;
207
} else if (req->elem.in_num > 3) {
209
* If we have more than 3 input segments the guest wants to actually
212
hdr.dxfer_direction = SG_DXFER_FROM_DEV;
213
hdr.iovec_count = req->elem.in_num - 3;
214
for (i = 0; i < hdr.iovec_count; i++)
215
hdr.dxfer_len += req->elem.in_sg[i].iov_len;
217
hdr.dxferp = req->elem.in_sg;
220
* Some SCSI commands don't actually transfer any data.
222
hdr.dxfer_direction = SG_DXFER_NONE;
225
hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
226
hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
228
ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
230
status = VIRTIO_BLK_S_UNSUPP;
235
* From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
236
* clear the masked_status field [hence status gets cleared too, see
237
* block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
238
* status has occurred. However they do set DRIVER_SENSE in driver_status
239
* field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
241
if (hdr.status == 0 && hdr.sb_len_wr > 0) {
242
hdr.status = CHECK_CONDITION;
245
stl_p(&req->scsi->errors,
246
hdr.status | (hdr.msg_status << 8) |
247
(hdr.host_status << 16) | (hdr.driver_status << 24));
248
stl_p(&req->scsi->residual, hdr.resid);
249
stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
250
stl_p(&req->scsi->data_len, hdr.dxfer_len);
252
virtio_blk_req_complete(req, status);
260
/* Just put anything nonzero so that the ioctl fails in the guest. */
261
stl_p(&req->scsi->errors, 255);
262
virtio_blk_req_complete(req, status);
266
typedef struct MultiReqBuffer {
267
BlockRequest blkreq[32];
268
unsigned int num_writes;
271
static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
275
if (!mrb->num_writes) {
279
ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
281
for (i = 0; i < mrb->num_writes; i++) {
282
if (mrb->blkreq[i].error) {
283
virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
291
static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
293
bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
296
* Make sure all outstanding writes are posted to the backing device.
298
virtio_submit_multiwrite(req->dev->bs, mrb);
299
bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
302
static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
304
BlockRequest *blkreq;
307
sector = ldq_p(&req->out->sector);
309
bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
311
trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
313
if (sector & req->dev->sector_mask) {
314
virtio_blk_rw_complete(req, -EIO);
317
if (req->qiov.size % req->dev->conf->logical_block_size) {
318
virtio_blk_rw_complete(req, -EIO);
322
if (mrb->num_writes == 32) {
323
virtio_submit_multiwrite(req->dev->bs, mrb);
326
blkreq = &mrb->blkreq[mrb->num_writes];
327
blkreq->sector = sector;
328
blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
329
blkreq->qiov = &req->qiov;
330
blkreq->cb = virtio_blk_rw_complete;
331
blkreq->opaque = req;
337
static void virtio_blk_handle_read(VirtIOBlockReq *req)
341
sector = ldq_p(&req->out->sector);
343
bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
345
trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
347
if (sector & req->dev->sector_mask) {
348
virtio_blk_rw_complete(req, -EIO);
351
if (req->qiov.size % req->dev->conf->logical_block_size) {
352
virtio_blk_rw_complete(req, -EIO);
355
bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
356
req->qiov.size / BDRV_SECTOR_SIZE,
357
virtio_blk_rw_complete, req);
360
static void virtio_blk_handle_request(VirtIOBlockReq *req,
365
if (req->elem.out_num < 1 || req->elem.in_num < 1) {
366
error_report("virtio-blk missing headers");
370
if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
371
req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
372
error_report("virtio-blk header not in correct element");
376
req->out = (void *)req->elem.out_sg[0].iov_base;
377
req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
379
type = ldl_p(&req->out->type);
381
if (type & VIRTIO_BLK_T_FLUSH) {
382
virtio_blk_handle_flush(req, mrb);
383
} else if (type & VIRTIO_BLK_T_SCSI_CMD) {
384
virtio_blk_handle_scsi(req);
385
} else if (type & VIRTIO_BLK_T_GET_ID) {
386
VirtIOBlock *s = req->dev;
389
* NB: per existing s/n string convention the string is
390
* terminated by '\0' only when shorter than buffer.
392
strncpy(req->elem.in_sg[0].iov_base,
393
s->blk->serial ? s->blk->serial : "",
394
MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
395
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
397
} else if (type & VIRTIO_BLK_T_OUT) {
398
qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
399
req->elem.out_num - 1);
400
virtio_blk_handle_write(req, mrb);
401
} else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
402
/* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
403
qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
404
req->elem.in_num - 1);
405
virtio_blk_handle_read(req);
407
virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
412
static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
414
VirtIOBlock *s = to_virtio_blk(vdev);
416
MultiReqBuffer mrb = {
420
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
421
/* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
422
* dataplane here instead of waiting for .set_status().
425
virtio_blk_data_plane_start(s->dataplane);
430
while ((req = virtio_blk_get_request(s))) {
431
virtio_blk_handle_request(req, &mrb);
434
virtio_submit_multiwrite(s->bs, &mrb);
437
* FIXME: Want to check for completions before returning to guest mode,
438
* so cached reads and writes are reported as quickly as possible. But
439
* that should be done in the generic block layer.
443
static void virtio_blk_dma_restart_bh(void *opaque)
445
VirtIOBlock *s = opaque;
446
VirtIOBlockReq *req = s->rq;
447
MultiReqBuffer mrb = {
451
qemu_bh_delete(s->bh);
457
virtio_blk_handle_request(req, &mrb);
461
virtio_submit_multiwrite(s->bs, &mrb);
464
static void virtio_blk_dma_restart_cb(void *opaque, int running,
467
VirtIOBlock *s = opaque;
474
s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
475
qemu_bh_schedule(s->bh);
479
static void virtio_blk_reset(VirtIODevice *vdev)
481
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
482
VirtIOBlock *s = to_virtio_blk(vdev);
485
virtio_blk_data_plane_stop(s->dataplane);
490
* This should cancel pending requests, but can't do nicely until there
491
* are per-device request lists.
496
/* coalesce internal state, copy to pci i/o region 0
498
static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
500
VirtIOBlock *s = to_virtio_blk(vdev);
501
struct virtio_blk_config blkcfg;
503
int blk_size = s->conf->logical_block_size;
505
bdrv_get_geometry(s->bs, &capacity);
506
memset(&blkcfg, 0, sizeof(blkcfg));
507
stq_raw(&blkcfg.capacity, capacity);
508
stl_raw(&blkcfg.seg_max, 128 - 2);
509
stw_raw(&blkcfg.cylinders, s->conf->cyls);
510
stl_raw(&blkcfg.blk_size, blk_size);
511
stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
512
stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
513
blkcfg.heads = s->conf->heads;
515
* We must ensure that the block device capacity is a multiple of
516
* the logical block size. If that is not the case, lets use
517
* sector_mask to adopt the geometry to have a correct picture.
518
* For those devices where the capacity is ok for the given geometry
519
* we dont touch the sector value of the geometry, since some devices
520
* (like s390 dasd) need a specific value. Here the capacity is already
521
* cyls*heads*secs*blk_size and the sector value is not block size
522
* divided by 512 - instead it is the amount of blk_size blocks
523
* per track (cylinder).
525
if (bdrv_getlength(s->bs) / s->conf->heads / s->conf->secs % blk_size) {
526
blkcfg.sectors = s->conf->secs & ~s->sector_mask;
528
blkcfg.sectors = s->conf->secs;
531
blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
532
blkcfg.alignment_offset = 0;
533
blkcfg.wce = bdrv_enable_write_cache(s->bs);
534
memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
537
static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
539
VirtIOBlock *s = to_virtio_blk(vdev);
540
struct virtio_blk_config blkcfg;
542
memcpy(&blkcfg, config, sizeof(blkcfg));
543
bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
546
static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
548
VirtIOBlock *s = to_virtio_blk(vdev);
550
features |= (1 << VIRTIO_BLK_F_SEG_MAX);
551
features |= (1 << VIRTIO_BLK_F_GEOMETRY);
552
features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
553
features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
554
features |= (1 << VIRTIO_BLK_F_SCSI);
556
if (s->blk->config_wce) {
557
features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
559
if (bdrv_enable_write_cache(s->bs))
560
features |= (1 << VIRTIO_BLK_F_WCE);
562
if (bdrv_is_read_only(s->bs))
563
features |= 1 << VIRTIO_BLK_F_RO;
568
static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
570
VirtIOBlock *s = to_virtio_blk(vdev);
573
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
574
if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
575
VIRTIO_CONFIG_S_DRIVER_OK))) {
576
virtio_blk_data_plane_stop(s->dataplane);
580
if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
584
features = vdev->guest_features;
585
bdrv_set_enable_write_cache(s->bs, !!(features & (1 << VIRTIO_BLK_F_WCE)));
588
static void virtio_blk_save(QEMUFile *f, void *opaque)
590
VirtIOBlock *s = opaque;
591
VirtIOBlockReq *req = s->rq;
593
virtio_save(&s->vdev, f);
596
qemu_put_sbyte(f, 1);
597
qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
600
qemu_put_sbyte(f, 0);
603
static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
605
VirtIOBlock *s = opaque;
611
ret = virtio_load(&s->vdev, f);
616
while (qemu_get_sbyte(f)) {
617
VirtIOBlockReq *req = virtio_blk_alloc_request(s);
618
qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
622
virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
623
req->elem.in_num, 1);
624
virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
625
req->elem.out_num, 0);
631
static void virtio_blk_resize(void *opaque)
633
VirtIOBlock *s = opaque;
635
virtio_notify_config(&s->vdev);
638
static const BlockDevOps virtio_block_ops = {
639
.resize_cb = virtio_blk_resize,
642
VirtIODevice *virtio_blk_init(DeviceState *dev, VirtIOBlkConf *blk)
645
static int virtio_blk_id;
648
error_report("drive property not set");
651
if (!bdrv_is_inserted(blk->conf.bs)) {
652
error_report("Device needs media, but drive is empty");
656
blkconf_serial(&blk->conf, &blk->serial);
657
if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
661
s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
662
sizeof(struct virtio_blk_config),
663
sizeof(VirtIOBlock));
665
s->vdev.get_config = virtio_blk_update_config;
666
s->vdev.set_config = virtio_blk_set_config;
667
s->vdev.get_features = virtio_blk_get_features;
668
s->vdev.set_status = virtio_blk_set_status;
669
s->vdev.reset = virtio_blk_reset;
670
s->bs = blk->conf.bs;
671
s->conf = &blk->conf;
674
s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
676
s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
677
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
678
if (!virtio_blk_data_plane_create(&s->vdev, blk, &s->dataplane)) {
679
virtio_cleanup(&s->vdev);
684
qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
686
register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
687
virtio_blk_save, virtio_blk_load, s);
688
bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
689
bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size);
691
bdrv_iostatus_enable(s->bs);
692
add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
697
void virtio_blk_exit(VirtIODevice *vdev)
699
VirtIOBlock *s = to_virtio_blk(vdev);
701
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
702
virtio_blk_data_plane_destroy(s->dataplane);
705
unregister_savevm(s->qdev, "virtio-blk", s);
706
blockdev_mark_auto_del(s->bs);
707
virtio_cleanup(vdev);