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 "virtio-blk.h"
17
#include "block_int.h"
19
typedef struct VirtIOBlock
27
static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
29
return (VirtIOBlock *)vdev;
32
typedef struct VirtIOBlockReq
35
VirtQueueElement elem;
36
struct virtio_blk_inhdr *in;
37
struct virtio_blk_outhdr *out;
40
struct VirtIOBlockReq *next;
43
static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
45
VirtIOBlock *s = req->dev;
47
req->in->status = status;
48
virtqueue_push(s->vq, &req->elem, req->size + sizeof(*req->in));
49
virtio_notify(&s->vdev, s->vq);
51
qemu_free(req->buffer);
55
static int virtio_blk_handle_write_error(VirtIOBlockReq *req, int error)
57
BlockInterfaceErrorAction action = drive_get_onerror(req->dev->bs);
58
VirtIOBlock *s = req->dev;
60
if (action == BLOCK_ERR_IGNORE)
63
if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
64
|| action == BLOCK_ERR_STOP_ANY) {
69
virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
75
static void virtio_blk_rw_complete(void *opaque, int ret)
77
VirtIOBlockReq *req = opaque;
79
/* Copy read data to the guest */
80
if (!ret && !(req->out->type & VIRTIO_BLK_T_OUT)) {
84
for (i = 0; i < req->elem.in_num - 1; i++) {
87
/* Be pretty defensive wrt malicious guests */
88
len = MIN(req->elem.in_sg[i].iov_len,
91
memcpy(req->elem.in_sg[i].iov_base,
96
} else if (ret && (req->out->type & VIRTIO_BLK_T_OUT)) {
97
if (virtio_blk_handle_write_error(req, -ret))
101
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
104
static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
106
VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
111
static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
113
VirtIOBlockReq *req = virtio_blk_alloc_request(s);
116
if (!virtqueue_pop(s->vq, &req->elem)) {
125
static int virtio_blk_handle_write(VirtIOBlockReq *req)
131
for (i = 1; i < req->elem.out_num; i++)
132
req->size += req->elem.out_sg[i].iov_len;
134
req->buffer = qemu_memalign(512, req->size);
135
if (req->buffer == NULL) {
140
/* We copy the data from the SG list to avoid splitting up the request.
141
This helps performance a lot until we can pass full sg lists as AIO
143
for (i = 1; i < req->elem.out_num; i++) {
146
len = MIN(req->elem.out_sg[i].iov_len,
148
memcpy(req->buffer + offset,
149
req->elem.out_sg[i].iov_base,
155
bdrv_aio_write(req->dev->bs, req->out->sector, req->buffer, req->size / 512,
156
virtio_blk_rw_complete, req);
160
static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
162
VirtIOBlock *s = to_virtio_blk(vdev);
165
while ((req = virtio_blk_get_request(s))) {
168
if (req->elem.out_num < 1 || req->elem.in_num < 1) {
169
fprintf(stderr, "virtio-blk missing headers\n");
173
if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
174
req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
175
fprintf(stderr, "virtio-blk header not in correct element\n");
179
req->out = (void *)req->elem.out_sg[0].iov_base;
180
req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
182
if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
183
unsigned int len = sizeof(*req->in);
185
req->in->status = VIRTIO_BLK_S_UNSUPP;
186
virtqueue_push(vq, &req->elem, len);
187
virtio_notify(vdev, vq);
189
} else if (req->out->type & VIRTIO_BLK_T_OUT) {
190
if (virtio_blk_handle_write(req) < 0)
193
for (i = 0; i < req->elem.in_num - 1; i++)
194
req->size += req->elem.in_sg[i].iov_len;
196
req->buffer = qemu_memalign(512, req->size);
197
if (req->buffer == NULL) {
202
bdrv_aio_read(s->bs, req->out->sector,
205
virtio_blk_rw_complete,
210
* FIXME: Want to check for completions before returning to guest mode,
211
* so cached reads and writes are reported as quickly as possible. But
212
* that should be done in the generic block layer.
216
static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
218
VirtIOBlock *s = opaque;
219
VirtIOBlockReq *req = s->rq;
227
virtio_blk_handle_write(req);
232
static void virtio_blk_reset(VirtIODevice *vdev)
235
* This should cancel pending requests, but can't do nicely until there
236
* are per-device request lists.
241
static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
243
VirtIOBlock *s = to_virtio_blk(vdev);
244
struct virtio_blk_config blkcfg;
246
int cylinders, heads, secs;
248
bdrv_get_geometry(s->bs, &capacity);
249
bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
250
stq_raw(&blkcfg.capacity, capacity);
251
stl_raw(&blkcfg.seg_max, 128 - 2);
252
stw_raw(&blkcfg.cylinders, cylinders);
253
blkcfg.heads = heads;
254
blkcfg.sectors = secs;
255
memcpy(config, &blkcfg, sizeof(blkcfg));
258
static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
260
return (1 << VIRTIO_BLK_F_SEG_MAX | 1 << VIRTIO_BLK_F_GEOMETRY);
263
static void virtio_blk_save(QEMUFile *f, void *opaque)
265
VirtIOBlock *s = opaque;
266
VirtIOBlockReq *req = s->rq;
268
virtio_save(&s->vdev, f);
271
qemu_put_sbyte(f, 1);
272
qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
275
qemu_put_sbyte(f, 0);
278
static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
280
VirtIOBlock *s = opaque;
285
virtio_load(&s->vdev, f);
286
while (qemu_get_sbyte(f)) {
287
VirtIOBlockReq *req = virtio_blk_alloc_request(s);
288
qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
296
void *virtio_blk_init(PCIBus *bus, BlockDriverState *bs)
299
int cylinders, heads, secs;
300
static int virtio_blk_id;
302
s = (VirtIOBlock *)virtio_init_pci(bus, "virtio-blk",
303
PCI_VENDOR_ID_REDHAT_QUMRANET,
304
PCI_DEVICE_ID_VIRTIO_BLOCK,
305
PCI_VENDOR_ID_REDHAT_QUMRANET,
307
PCI_CLASS_STORAGE_OTHER, 0x00,
308
sizeof(struct virtio_blk_config), sizeof(VirtIOBlock));
312
s->vdev.get_config = virtio_blk_update_config;
313
s->vdev.get_features = virtio_blk_get_features;
314
s->vdev.reset = virtio_blk_reset;
317
bs->private = &s->vdev.pci_dev;
318
bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
319
bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
321
s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
323
qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
324
register_savevm("virtio-blk", virtio_blk_id++, 2,
325
virtio_blk_save, virtio_blk_load, s);