4 * Copyright IBM, Corp. 2007
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"
22 typedef struct VirtIOBlock
28 char serial_str[BLOCK_SERIAL_STRLEN + 1];
33 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
35 return (VirtIOBlock *)vdev;
38 /* store identify data in little endian format
40 static inline void put_le16(uint16_t *p, unsigned int v)
45 /* copy to *dst from *src, nul pad dst tail as needed to len bytes
47 static inline void padstr(char *dst, const char *src, int len)
50 *dst++ = *src ? *src++ : '\0';
53 /* setup simulated identify data as appropriate for virtio block device
55 * ref: AT Attachment 8 - ATA/ATAPI Command Set (ATA8-ACS)
57 static inline void virtio_identify_template(struct virtio_blk_config *bc)
59 uint16_t *p = &bc->identify[0];
60 uint64_t lba_sectors = bc->capacity;
62 memset(p, 0, sizeof(bc->identify));
63 put_le16(p + 0, 0x0); /* ATA device */
64 padstr((char *)(p + 23), QEMU_VERSION, 8); /* firmware revision */
65 padstr((char *)(p + 27), "QEMU VIRT_BLK", 40); /* model# */
66 put_le16(p + 47, 0x80ff); /* max xfer 255 sectors */
67 put_le16(p + 49, 0x0b00); /* support IORDY/LBA/DMA */
68 put_le16(p + 59, 0x1ff); /* cur xfer 255 sectors */
69 put_le16(p + 80, 0x1f0); /* support ATA8/7/6/5/4 */
70 put_le16(p + 81, 0x16);
71 put_le16(p + 82, 0x400);
72 put_le16(p + 83, 0x400);
73 put_le16(p + 100, lba_sectors);
74 put_le16(p + 101, lba_sectors >> 16);
75 put_le16(p + 102, lba_sectors >> 32);
76 put_le16(p + 103, lba_sectors >> 48);
79 typedef struct VirtIOBlockReq
82 VirtQueueElement elem;
83 struct virtio_blk_inhdr *in;
84 struct virtio_blk_outhdr *out;
85 struct virtio_scsi_inhdr *scsi;
87 struct VirtIOBlockReq *next;
90 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
92 VirtIOBlock *s = req->dev;
94 req->in->status = status;
95 virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
96 virtio_notify(&s->vdev, s->vq);
101 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
104 BlockInterfaceErrorAction action =
105 drive_get_on_error(req->dev->bs, is_read);
106 VirtIOBlock *s = req->dev;
108 if (action == BLOCK_ERR_IGNORE)
111 if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
112 || action == BLOCK_ERR_STOP_ANY) {
117 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
123 static void virtio_blk_rw_complete(void *opaque, int ret)
125 VirtIOBlockReq *req = opaque;
128 int is_read = !(req->out->type & VIRTIO_BLK_T_OUT);
129 if (virtio_blk_handle_rw_error(req, -ret, is_read))
133 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
136 static void virtio_blk_flush_complete(void *opaque, int ret)
138 VirtIOBlockReq *req = opaque;
140 virtio_blk_req_complete(req, ret ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK);
143 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
145 VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
150 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
152 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
155 if (!virtqueue_pop(s->vq, &req->elem)) {
165 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
167 struct sg_io_hdr hdr;
173 * We require at least one output segment each for the virtio_blk_outhdr
174 * and the SCSI command block.
176 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
177 * and the sense buffer pointer in the input segments.
179 if (req->elem.out_num < 2 || req->elem.in_num < 3) {
180 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
185 * No support for bidirection commands yet.
187 if (req->elem.out_num > 2 && req->elem.in_num > 3) {
188 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
193 * The scsi inhdr is placed in the second-to-last input segment, just
194 * before the regular inhdr.
196 req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
198 memset(&hdr, 0, sizeof(struct sg_io_hdr));
199 hdr.interface_id = 'S';
200 hdr.cmd_len = req->elem.out_sg[1].iov_len;
201 hdr.cmdp = req->elem.out_sg[1].iov_base;
204 if (req->elem.out_num > 2) {
206 * If there are more than the minimally required 2 output segments
207 * there is write payload starting from the third iovec.
209 hdr.dxfer_direction = SG_DXFER_TO_DEV;
210 hdr.iovec_count = req->elem.out_num - 2;
212 for (i = 0; i < hdr.iovec_count; i++)
213 hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
215 hdr.dxferp = req->elem.out_sg + 2;
217 } else if (req->elem.in_num > 3) {
219 * If we have more than 3 input segments the guest wants to actually
222 hdr.dxfer_direction = SG_DXFER_FROM_DEV;
223 hdr.iovec_count = req->elem.in_num - 3;
224 for (i = 0; i < hdr.iovec_count; i++)
225 hdr.dxfer_len += req->elem.in_sg[i].iov_len;
227 hdr.dxferp = req->elem.in_sg;
230 * Some SCSI commands don't actually transfer any data.
232 hdr.dxfer_direction = SG_DXFER_NONE;
235 hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
236 hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
238 ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
240 status = VIRTIO_BLK_S_UNSUPP;
242 hdr.resid = hdr.dxfer_len;
243 } else if (hdr.status) {
244 status = VIRTIO_BLK_S_IOERR;
246 status = VIRTIO_BLK_S_OK;
249 req->scsi->errors = hdr.status;
250 req->scsi->residual = hdr.resid;
251 req->scsi->sense_len = hdr.sb_len_wr;
252 req->scsi->data_len = hdr.dxfer_len;
254 virtio_blk_req_complete(req, status);
257 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
259 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
261 #endif /* __linux__ */
263 static void do_multiwrite(BlockDriverState *bs, BlockRequest *blkreq,
267 ret = bdrv_aio_multiwrite(bs, blkreq, num_writes);
270 for (i = 0; i < num_writes; i++) {
271 if (blkreq[i].error) {
272 virtio_blk_req_complete(blkreq[i].opaque, VIRTIO_BLK_S_IOERR);
278 static void virtio_blk_handle_flush(VirtIOBlockReq *req)
280 BlockDriverAIOCB *acb;
282 acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
284 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
288 static void virtio_blk_handle_write(BlockRequest *blkreq, int *num_writes,
289 VirtIOBlockReq *req, BlockDriverState **old_bs)
291 if (req->dev->bs != *old_bs || *num_writes == 32) {
292 if (*old_bs != NULL) {
293 do_multiwrite(*old_bs, blkreq, *num_writes);
296 *old_bs = req->dev->bs;
299 blkreq[*num_writes].sector = req->out->sector;
300 blkreq[*num_writes].nb_sectors = req->qiov.size / 512;
301 blkreq[*num_writes].qiov = &req->qiov;
302 blkreq[*num_writes].cb = virtio_blk_rw_complete;
303 blkreq[*num_writes].opaque = req;
304 blkreq[*num_writes].error = 0;
309 static void virtio_blk_handle_read(VirtIOBlockReq *req)
311 BlockDriverAIOCB *acb;
313 acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
314 req->qiov.size / 512, virtio_blk_rw_complete, req);
316 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
320 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
322 VirtIOBlock *s = to_virtio_blk(vdev);
324 BlockRequest blkreq[32];
326 BlockDriverState *old_bs = NULL;
328 while ((req = virtio_blk_get_request(s))) {
329 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
330 fprintf(stderr, "virtio-blk missing headers\n");
334 if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
335 req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
336 fprintf(stderr, "virtio-blk header not in correct element\n");
340 req->out = (void *)req->elem.out_sg[0].iov_base;
341 req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
343 if (req->out->type & VIRTIO_BLK_T_FLUSH) {
344 virtio_blk_handle_flush(req);
345 } else if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
346 virtio_blk_handle_scsi(req);
347 } else if (req->out->type & VIRTIO_BLK_T_OUT) {
348 qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
349 req->elem.out_num - 1);
350 virtio_blk_handle_write(blkreq, &num_writes, req, &old_bs);
352 qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
353 req->elem.in_num - 1);
354 virtio_blk_handle_read(req);
358 if (num_writes > 0) {
359 do_multiwrite(old_bs, blkreq, num_writes);
363 * FIXME: Want to check for completions before returning to guest mode,
364 * so cached reads and writes are reported as quickly as possible. But
365 * that should be done in the generic block layer.
369 static void virtio_blk_dma_restart_bh(void *opaque)
371 VirtIOBlock *s = opaque;
372 VirtIOBlockReq *req = s->rq;
374 qemu_bh_delete(s->bh);
380 bdrv_aio_writev(req->dev->bs, req->out->sector, &req->qiov,
381 req->qiov.size / 512, virtio_blk_rw_complete, req);
386 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
388 VirtIOBlock *s = opaque;
394 s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
395 qemu_bh_schedule(s->bh);
399 static void virtio_blk_reset(VirtIODevice *vdev)
402 * This should cancel pending requests, but can't do nicely until there
403 * are per-device request lists.
408 /* coalesce internal state, copy to pci i/o region 0
410 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
412 VirtIOBlock *s = to_virtio_blk(vdev);
413 struct virtio_blk_config blkcfg;
415 int cylinders, heads, secs;
417 bdrv_get_geometry(s->bs, &capacity);
418 bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
419 memset(&blkcfg, 0, sizeof(blkcfg));
420 stq_raw(&blkcfg.capacity, capacity);
421 stl_raw(&blkcfg.seg_max, 128 - 2);
422 stw_raw(&blkcfg.cylinders, cylinders);
423 blkcfg.heads = heads;
424 blkcfg.sectors = secs;
426 virtio_identify_template(&blkcfg);
427 memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str,
428 VIRTIO_BLK_ID_SN_BYTES);
429 memcpy(config, &blkcfg, s->config_size);
432 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
434 VirtIOBlock *s = to_virtio_blk(vdev);
436 features |= (1 << VIRTIO_BLK_F_SEG_MAX);
437 features |= (1 << VIRTIO_BLK_F_GEOMETRY);
439 if (bdrv_enable_write_cache(s->bs))
440 features |= (1 << VIRTIO_BLK_F_WCACHE);
441 if (strcmp(s->serial_str, "0"))
442 features |= 1 << VIRTIO_BLK_F_IDENTIFY;
444 if (bdrv_is_read_only(s->bs))
445 features |= 1 << VIRTIO_BLK_F_RO;
450 static void virtio_blk_save(QEMUFile *f, void *opaque)
452 VirtIOBlock *s = opaque;
453 VirtIOBlockReq *req = s->rq;
455 virtio_save(&s->vdev, f);
458 qemu_put_sbyte(f, 1);
459 qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
462 qemu_put_sbyte(f, 0);
465 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
467 VirtIOBlock *s = opaque;
472 virtio_load(&s->vdev, f);
473 while (qemu_get_sbyte(f)) {
474 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
475 qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
483 VirtIODevice *virtio_blk_init(DeviceState *dev, DriveInfo *dinfo)
486 int cylinders, heads, secs;
487 static int virtio_blk_id;
488 char *ps = (char *)drive_get_serial(dinfo->bdrv);
489 size_t size = strlen(ps) ? sizeof(struct virtio_blk_config) :
490 offsetof(struct virtio_blk_config, _blk_size);
492 s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
494 sizeof(VirtIOBlock));
496 s->config_size = size;
497 s->vdev.get_config = virtio_blk_update_config;
498 s->vdev.get_features = virtio_blk_get_features;
499 s->vdev.reset = virtio_blk_reset;
503 strncpy(s->serial_str, ps, sizeof(s->serial_str));
505 snprintf(s->serial_str, sizeof(s->serial_str), "0");
506 bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
507 bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
509 s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
511 qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
512 register_savevm("virtio-blk", virtio_blk_id++, 2,
513 virtio_blk_save, virtio_blk_load, s);