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/block.h"
18
#include "sysemu/blockdev.h"
19
#include "hw/virtio/virtio-blk.h"
20
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
21
# include "dataplane/virtio-blk.h"
22
# include "migration/migration.h"
24
#include "block/scsi.h"
28
#include "hw/virtio/virtio-bus.h"
30
typedef struct VirtIOBlockReq
33
VirtQueueElement elem;
34
struct virtio_blk_inhdr *in;
35
struct virtio_blk_outhdr *out;
36
struct virtio_scsi_inhdr *scsi;
38
struct VirtIOBlockReq *next;
42
static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
44
VirtIOBlock *s = req->dev;
45
VirtIODevice *vdev = VIRTIO_DEVICE(s);
47
trace_virtio_blk_req_complete(req, status);
49
stb_p(&req->in->status, status);
50
virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
51
virtio_notify(vdev, s->vq);
54
static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
57
BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
58
VirtIOBlock *s = req->dev;
60
if (action == BDRV_ACTION_STOP) {
63
} else if (action == BDRV_ACTION_REPORT) {
64
virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
65
bdrv_acct_done(s->bs, &req->acct);
69
bdrv_error_action(s->bs, action, is_read, error);
70
return action != BDRV_ACTION_IGNORE;
73
static void virtio_blk_rw_complete(void *opaque, int ret)
75
VirtIOBlockReq *req = opaque;
77
trace_virtio_blk_rw_complete(req, ret);
80
bool is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
81
if (virtio_blk_handle_rw_error(req, -ret, is_read))
85
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
86
bdrv_acct_done(req->dev->bs, &req->acct);
90
static void virtio_blk_flush_complete(void *opaque, int ret)
92
VirtIOBlockReq *req = opaque;
95
if (virtio_blk_handle_rw_error(req, -ret, 0)) {
100
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
101
bdrv_acct_done(req->dev->bs, &req->acct);
105
static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
107
VirtIOBlockReq *req = g_malloc(sizeof(*req));
114
static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
116
VirtIOBlockReq *req = virtio_blk_alloc_request(s);
119
if (!virtqueue_pop(s->vq, &req->elem)) {
128
static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
134
int status = VIRTIO_BLK_S_OK;
137
* We require at least one output segment each for the virtio_blk_outhdr
138
* and the SCSI command block.
140
* We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
141
* and the sense buffer pointer in the input segments.
143
if (req->elem.out_num < 2 || req->elem.in_num < 3) {
144
virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
150
* The scsi inhdr is placed in the second-to-last input segment, just
151
* before the regular inhdr.
153
req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
155
if (!req->dev->blk.scsi) {
156
status = VIRTIO_BLK_S_UNSUPP;
161
* No support for bidirection commands yet.
163
if (req->elem.out_num > 2 && req->elem.in_num > 3) {
164
status = VIRTIO_BLK_S_UNSUPP;
169
struct sg_io_hdr hdr;
170
memset(&hdr, 0, sizeof(struct sg_io_hdr));
171
hdr.interface_id = 'S';
172
hdr.cmd_len = req->elem.out_sg[1].iov_len;
173
hdr.cmdp = req->elem.out_sg[1].iov_base;
176
if (req->elem.out_num > 2) {
178
* If there are more than the minimally required 2 output segments
179
* there is write payload starting from the third iovec.
181
hdr.dxfer_direction = SG_DXFER_TO_DEV;
182
hdr.iovec_count = req->elem.out_num - 2;
184
for (i = 0; i < hdr.iovec_count; i++)
185
hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
187
hdr.dxferp = req->elem.out_sg + 2;
189
} else if (req->elem.in_num > 3) {
191
* If we have more than 3 input segments the guest wants to actually
194
hdr.dxfer_direction = SG_DXFER_FROM_DEV;
195
hdr.iovec_count = req->elem.in_num - 3;
196
for (i = 0; i < hdr.iovec_count; i++)
197
hdr.dxfer_len += req->elem.in_sg[i].iov_len;
199
hdr.dxferp = req->elem.in_sg;
202
* Some SCSI commands don't actually transfer any data.
204
hdr.dxfer_direction = SG_DXFER_NONE;
207
hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
208
hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
210
ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
212
status = VIRTIO_BLK_S_UNSUPP;
217
* From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
218
* clear the masked_status field [hence status gets cleared too, see
219
* block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
220
* status has occurred. However they do set DRIVER_SENSE in driver_status
221
* field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
223
if (hdr.status == 0 && hdr.sb_len_wr > 0) {
224
hdr.status = CHECK_CONDITION;
227
stl_p(&req->scsi->errors,
228
hdr.status | (hdr.msg_status << 8) |
229
(hdr.host_status << 16) | (hdr.driver_status << 24));
230
stl_p(&req->scsi->residual, hdr.resid);
231
stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
232
stl_p(&req->scsi->data_len, hdr.dxfer_len);
234
virtio_blk_req_complete(req, status);
242
/* Just put anything nonzero so that the ioctl fails in the guest. */
243
stl_p(&req->scsi->errors, 255);
244
virtio_blk_req_complete(req, status);
248
typedef struct MultiReqBuffer {
249
BlockRequest blkreq[32];
250
unsigned int num_writes;
253
static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
257
if (!mrb->num_writes) {
261
ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
263
for (i = 0; i < mrb->num_writes; i++) {
264
if (mrb->blkreq[i].error) {
265
virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
273
static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
275
bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
278
* Make sure all outstanding writes are posted to the backing device.
280
virtio_submit_multiwrite(req->dev->bs, mrb);
281
bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
284
static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
286
BlockRequest *blkreq;
289
sector = ldq_p(&req->out->sector);
291
bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
293
trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
295
if (sector & req->dev->sector_mask) {
296
virtio_blk_rw_complete(req, -EIO);
299
if (req->qiov.size % req->dev->conf->logical_block_size) {
300
virtio_blk_rw_complete(req, -EIO);
304
if (mrb->num_writes == 32) {
305
virtio_submit_multiwrite(req->dev->bs, mrb);
308
blkreq = &mrb->blkreq[mrb->num_writes];
309
blkreq->sector = sector;
310
blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
311
blkreq->qiov = &req->qiov;
312
blkreq->cb = virtio_blk_rw_complete;
313
blkreq->opaque = req;
319
static void virtio_blk_handle_read(VirtIOBlockReq *req)
323
sector = ldq_p(&req->out->sector);
325
bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
327
trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
329
if (sector & req->dev->sector_mask) {
330
virtio_blk_rw_complete(req, -EIO);
333
if (req->qiov.size % req->dev->conf->logical_block_size) {
334
virtio_blk_rw_complete(req, -EIO);
337
bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
338
req->qiov.size / BDRV_SECTOR_SIZE,
339
virtio_blk_rw_complete, req);
342
static void virtio_blk_handle_request(VirtIOBlockReq *req,
347
if (req->elem.out_num < 1 || req->elem.in_num < 1) {
348
error_report("virtio-blk missing headers");
352
if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
353
req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
354
error_report("virtio-blk header not in correct element");
358
req->out = (void *)req->elem.out_sg[0].iov_base;
359
req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
361
type = ldl_p(&req->out->type);
363
if (type & VIRTIO_BLK_T_FLUSH) {
364
virtio_blk_handle_flush(req, mrb);
365
} else if (type & VIRTIO_BLK_T_SCSI_CMD) {
366
virtio_blk_handle_scsi(req);
367
} else if (type & VIRTIO_BLK_T_GET_ID) {
368
VirtIOBlock *s = req->dev;
371
* NB: per existing s/n string convention the string is
372
* terminated by '\0' only when shorter than buffer.
374
strncpy(req->elem.in_sg[0].iov_base,
375
s->blk.serial ? s->blk.serial : "",
376
MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
377
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
379
} else if (type & VIRTIO_BLK_T_OUT) {
380
qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
381
req->elem.out_num - 1);
382
virtio_blk_handle_write(req, mrb);
383
} else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
384
/* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
385
qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
386
req->elem.in_num - 1);
387
virtio_blk_handle_read(req);
389
virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
394
static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
396
VirtIOBlock *s = VIRTIO_BLK(vdev);
398
MultiReqBuffer mrb = {
402
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
403
/* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
404
* dataplane here instead of waiting for .set_status().
407
virtio_blk_data_plane_start(s->dataplane);
412
while ((req = virtio_blk_get_request(s))) {
413
virtio_blk_handle_request(req, &mrb);
416
virtio_submit_multiwrite(s->bs, &mrb);
419
* FIXME: Want to check for completions before returning to guest mode,
420
* so cached reads and writes are reported as quickly as possible. But
421
* that should be done in the generic block layer.
425
static void virtio_blk_dma_restart_bh(void *opaque)
427
VirtIOBlock *s = opaque;
428
VirtIOBlockReq *req = s->rq;
429
MultiReqBuffer mrb = {
433
qemu_bh_delete(s->bh);
439
virtio_blk_handle_request(req, &mrb);
443
virtio_submit_multiwrite(s->bs, &mrb);
446
static void virtio_blk_dma_restart_cb(void *opaque, int running,
449
VirtIOBlock *s = opaque;
456
s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
457
qemu_bh_schedule(s->bh);
461
static void virtio_blk_reset(VirtIODevice *vdev)
463
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
464
VirtIOBlock *s = VIRTIO_BLK(vdev);
467
virtio_blk_data_plane_stop(s->dataplane);
472
* This should cancel pending requests, but can't do nicely until there
473
* are per-device request lists.
478
/* coalesce internal state, copy to pci i/o region 0
480
static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
482
VirtIOBlock *s = VIRTIO_BLK(vdev);
483
struct virtio_blk_config blkcfg;
485
int blk_size = s->conf->logical_block_size;
487
bdrv_get_geometry(s->bs, &capacity);
488
memset(&blkcfg, 0, sizeof(blkcfg));
489
stq_raw(&blkcfg.capacity, capacity);
490
stl_raw(&blkcfg.seg_max, 128 - 2);
491
stw_raw(&blkcfg.cylinders, s->conf->cyls);
492
stl_raw(&blkcfg.blk_size, blk_size);
493
stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
494
stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
495
blkcfg.heads = s->conf->heads;
497
* We must ensure that the block device capacity is a multiple of
498
* the logical block size. If that is not the case, let's use
499
* sector_mask to adopt the geometry to have a correct picture.
500
* For those devices where the capacity is ok for the given geometry
501
* we don't touch the sector value of the geometry, since some devices
502
* (like s390 dasd) need a specific value. Here the capacity is already
503
* cyls*heads*secs*blk_size and the sector value is not block size
504
* divided by 512 - instead it is the amount of blk_size blocks
505
* per track (cylinder).
507
if (bdrv_getlength(s->bs) / s->conf->heads / s->conf->secs % blk_size) {
508
blkcfg.sectors = s->conf->secs & ~s->sector_mask;
510
blkcfg.sectors = s->conf->secs;
513
blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
514
blkcfg.alignment_offset = 0;
515
blkcfg.wce = bdrv_enable_write_cache(s->bs);
516
memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
519
static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
521
VirtIOBlock *s = VIRTIO_BLK(vdev);
522
struct virtio_blk_config blkcfg;
524
memcpy(&blkcfg, config, sizeof(blkcfg));
525
bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
528
static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
530
VirtIOBlock *s = VIRTIO_BLK(vdev);
532
features |= (1 << VIRTIO_BLK_F_SEG_MAX);
533
features |= (1 << VIRTIO_BLK_F_GEOMETRY);
534
features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
535
features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
536
features |= (1 << VIRTIO_BLK_F_SCSI);
538
if (s->blk.config_wce) {
539
features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
541
if (bdrv_enable_write_cache(s->bs))
542
features |= (1 << VIRTIO_BLK_F_WCE);
544
if (bdrv_is_read_only(s->bs))
545
features |= 1 << VIRTIO_BLK_F_RO;
550
static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
552
VirtIOBlock *s = VIRTIO_BLK(vdev);
555
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
556
if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
557
VIRTIO_CONFIG_S_DRIVER_OK))) {
558
virtio_blk_data_plane_stop(s->dataplane);
562
if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
566
features = vdev->guest_features;
567
bdrv_set_enable_write_cache(s->bs, !!(features & (1 << VIRTIO_BLK_F_WCE)));
570
static void virtio_blk_save(QEMUFile *f, void *opaque)
572
VirtIOBlock *s = opaque;
573
VirtIODevice *vdev = VIRTIO_DEVICE(s);
574
VirtIOBlockReq *req = s->rq;
576
virtio_save(vdev, f);
579
qemu_put_sbyte(f, 1);
580
qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
583
qemu_put_sbyte(f, 0);
586
static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
588
VirtIOBlock *s = opaque;
589
VirtIODevice *vdev = VIRTIO_DEVICE(s);
595
ret = virtio_load(vdev, f);
600
while (qemu_get_sbyte(f)) {
601
VirtIOBlockReq *req = virtio_blk_alloc_request(s);
602
qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
606
virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
607
req->elem.in_num, 1);
608
virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
609
req->elem.out_num, 0);
615
static void virtio_blk_resize(void *opaque)
617
VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
619
virtio_notify_config(vdev);
622
static const BlockDevOps virtio_block_ops = {
623
.resize_cb = virtio_blk_resize,
626
void virtio_blk_set_conf(DeviceState *dev, VirtIOBlkConf *blk)
628
VirtIOBlock *s = VIRTIO_BLK(dev);
629
memcpy(&(s->blk), blk, sizeof(struct VirtIOBlkConf));
632
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
633
/* Disable dataplane thread during live migration since it does not
634
* update the dirty memory bitmap yet.
636
static void virtio_blk_migration_state_changed(Notifier *notifier, void *data)
638
VirtIOBlock *s = container_of(notifier, VirtIOBlock,
639
migration_state_notifier);
640
MigrationState *mig = data;
642
if (migration_in_setup(mig)) {
646
virtio_blk_data_plane_destroy(s->dataplane);
648
} else if (migration_has_finished(mig) ||
649
migration_has_failed(mig)) {
653
bdrv_drain_all(); /* complete in-flight non-dataplane requests */
654
virtio_blk_data_plane_create(VIRTIO_DEVICE(s), &s->blk,
658
#endif /* CONFIG_VIRTIO_BLK_DATA_PLANE */
660
static int virtio_blk_device_init(VirtIODevice *vdev)
662
DeviceState *qdev = DEVICE(vdev);
663
VirtIOBlock *s = VIRTIO_BLK(vdev);
664
VirtIOBlkConf *blk = &(s->blk);
665
static int virtio_blk_id;
668
error_report("drive property not set");
671
if (!bdrv_is_inserted(blk->conf.bs)) {
672
error_report("Device needs media, but drive is empty");
676
blkconf_serial(&blk->conf, &blk->serial);
677
if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
681
virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
682
sizeof(struct virtio_blk_config));
684
s->bs = blk->conf.bs;
685
s->conf = &blk->conf;
686
memcpy(&(s->blk), blk, sizeof(struct VirtIOBlkConf));
688
s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
690
s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
691
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
692
if (!virtio_blk_data_plane_create(vdev, blk, &s->dataplane)) {
693
virtio_cleanup(vdev);
696
s->migration_state_notifier.notify = virtio_blk_migration_state_changed;
697
add_migration_state_change_notifier(&s->migration_state_notifier);
700
s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
701
register_savevm(qdev, "virtio-blk", virtio_blk_id++, 2,
702
virtio_blk_save, virtio_blk_load, s);
703
bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
704
bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size);
706
bdrv_iostatus_enable(s->bs);
708
add_boot_device_path(s->conf->bootindex, qdev, "/disk@0,0");
712
static int virtio_blk_device_exit(DeviceState *dev)
714
VirtIODevice *vdev = VIRTIO_DEVICE(dev);
715
VirtIOBlock *s = VIRTIO_BLK(dev);
716
#ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
717
remove_migration_state_change_notifier(&s->migration_state_notifier);
718
virtio_blk_data_plane_destroy(s->dataplane);
721
qemu_del_vm_change_state_handler(s->change);
722
unregister_savevm(dev, "virtio-blk", s);
723
blockdev_mark_auto_del(s->bs);
724
virtio_cleanup(vdev);
728
static Property virtio_blk_properties[] = {
729
DEFINE_VIRTIO_BLK_PROPERTIES(VirtIOBlock, blk),
730
DEFINE_PROP_END_OF_LIST(),
733
static void virtio_blk_class_init(ObjectClass *klass, void *data)
735
DeviceClass *dc = DEVICE_CLASS(klass);
736
VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
737
dc->exit = virtio_blk_device_exit;
738
dc->props = virtio_blk_properties;
739
set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
740
vdc->init = virtio_blk_device_init;
741
vdc->get_config = virtio_blk_update_config;
742
vdc->set_config = virtio_blk_set_config;
743
vdc->get_features = virtio_blk_get_features;
744
vdc->set_status = virtio_blk_set_status;
745
vdc->reset = virtio_blk_reset;
748
static const TypeInfo virtio_device_info = {
749
.name = TYPE_VIRTIO_BLK,
750
.parent = TYPE_VIRTIO_DEVICE,
751
.instance_size = sizeof(VirtIOBlock),
752
.class_init = virtio_blk_class_init,
755
static void virtio_register_types(void)
757
type_register_static(&virtio_device_info);
760
type_init(virtio_register_types)