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-transport.h"
20 #include "virtio-pci.h"
21 #include "virtio-blk.h"
22 #include "scsi-defs.h"
27 typedef struct VirtIOBlock
36 unsigned short sector_mask;
40 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
42 return (VirtIOBlock *)vdev;
45 typedef struct VirtIOBlockReq
48 VirtQueueElement elem;
49 struct virtio_blk_inhdr *in;
50 struct virtio_blk_outhdr *out;
51 struct virtio_scsi_inhdr *scsi;
53 struct VirtIOBlockReq *next;
57 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
59 VirtIOBlock *s = req->dev;
61 trace_virtio_blk_req_complete(req, status);
63 stb_p(&req->in->status, status);
64 virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
65 virtio_notify(&s->vdev, s->vq);
68 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
71 BlockErrorAction action = bdrv_get_on_error(req->dev->bs, is_read);
72 VirtIOBlock *s = req->dev;
74 if (action == BLOCK_ERR_IGNORE) {
75 bdrv_emit_qmp_error_event(s->bs, BDRV_ACTION_IGNORE, is_read);
79 if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
80 || action == BLOCK_ERR_STOP_ANY) {
83 bdrv_emit_qmp_error_event(s->bs, BDRV_ACTION_STOP, is_read);
84 vm_stop(RUN_STATE_IO_ERROR);
85 bdrv_iostatus_set_err(s->bs, error);
87 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
88 bdrv_acct_done(s->bs, &req->acct);
90 bdrv_emit_qmp_error_event(s->bs, BDRV_ACTION_REPORT, is_read);
96 static void virtio_blk_rw_complete(void *opaque, int ret)
98 VirtIOBlockReq *req = opaque;
100 trace_virtio_blk_rw_complete(req, ret);
103 int is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
104 if (virtio_blk_handle_rw_error(req, -ret, is_read))
108 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
109 bdrv_acct_done(req->dev->bs, &req->acct);
113 static void virtio_blk_flush_complete(void *opaque, int ret)
115 VirtIOBlockReq *req = opaque;
118 if (virtio_blk_handle_rw_error(req, -ret, 0)) {
123 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
124 bdrv_acct_done(req->dev->bs, &req->acct);
128 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
130 VirtIOBlockReq *req = g_malloc(sizeof(*req));
137 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
139 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
142 if (!virtqueue_pop(s->vq, &req->elem)) {
151 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
157 int status = VIRTIO_BLK_S_OK;
160 * We require at least one output segment each for the virtio_blk_outhdr
161 * and the SCSI command block.
163 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
164 * and the sense buffer pointer in the input segments.
166 if (req->elem.out_num < 2 || req->elem.in_num < 3) {
167 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
173 * The scsi inhdr is placed in the second-to-last input segment, just
174 * before the regular inhdr.
176 req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
178 if (!req->dev->blk->scsi) {
179 status = VIRTIO_BLK_S_UNSUPP;
184 * No support for bidirection commands yet.
186 if (req->elem.out_num > 2 && req->elem.in_num > 3) {
187 status = VIRTIO_BLK_S_UNSUPP;
192 struct sg_io_hdr hdr;
193 memset(&hdr, 0, sizeof(struct sg_io_hdr));
194 hdr.interface_id = 'S';
195 hdr.cmd_len = req->elem.out_sg[1].iov_len;
196 hdr.cmdp = req->elem.out_sg[1].iov_base;
199 if (req->elem.out_num > 2) {
201 * If there are more than the minimally required 2 output segments
202 * there is write payload starting from the third iovec.
204 hdr.dxfer_direction = SG_DXFER_TO_DEV;
205 hdr.iovec_count = req->elem.out_num - 2;
207 for (i = 0; i < hdr.iovec_count; i++)
208 hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
210 hdr.dxferp = req->elem.out_sg + 2;
212 } else if (req->elem.in_num > 3) {
214 * If we have more than 3 input segments the guest wants to actually
217 hdr.dxfer_direction = SG_DXFER_FROM_DEV;
218 hdr.iovec_count = req->elem.in_num - 3;
219 for (i = 0; i < hdr.iovec_count; i++)
220 hdr.dxfer_len += req->elem.in_sg[i].iov_len;
222 hdr.dxferp = req->elem.in_sg;
225 * Some SCSI commands don't actually transfer any data.
227 hdr.dxfer_direction = SG_DXFER_NONE;
230 hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
231 hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
233 ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
235 status = VIRTIO_BLK_S_UNSUPP;
240 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
241 * clear the masked_status field [hence status gets cleared too, see
242 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
243 * status has occurred. However they do set DRIVER_SENSE in driver_status
244 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
246 if (hdr.status == 0 && hdr.sb_len_wr > 0) {
247 hdr.status = CHECK_CONDITION;
250 stl_p(&req->scsi->errors,
251 hdr.status | (hdr.msg_status << 8) |
252 (hdr.host_status << 16) | (hdr.driver_status << 24));
253 stl_p(&req->scsi->residual, hdr.resid);
254 stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
255 stl_p(&req->scsi->data_len, hdr.dxfer_len);
257 virtio_blk_req_complete(req, status);
265 /* Just put anything nonzero so that the ioctl fails in the guest. */
266 stl_p(&req->scsi->errors, 255);
267 virtio_blk_req_complete(req, status);
271 typedef struct MultiReqBuffer {
272 BlockRequest blkreq[32];
273 unsigned int num_writes;
276 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
280 if (!mrb->num_writes) {
284 ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
286 for (i = 0; i < mrb->num_writes; i++) {
287 if (mrb->blkreq[i].error) {
288 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
296 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
298 bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
301 * Make sure all outstanding writes are posted to the backing device.
303 virtio_submit_multiwrite(req->dev->bs, mrb);
304 bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
307 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
309 BlockRequest *blkreq;
312 sector = ldq_p(&req->out->sector);
314 bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
316 trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
318 if (sector & req->dev->sector_mask) {
319 virtio_blk_rw_complete(req, -EIO);
322 if (req->qiov.size % req->dev->conf->logical_block_size) {
323 virtio_blk_rw_complete(req, -EIO);
327 if (mrb->num_writes == 32) {
328 virtio_submit_multiwrite(req->dev->bs, mrb);
331 blkreq = &mrb->blkreq[mrb->num_writes];
332 blkreq->sector = sector;
333 blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
334 blkreq->qiov = &req->qiov;
335 blkreq->cb = virtio_blk_rw_complete;
336 blkreq->opaque = req;
342 static void virtio_blk_handle_read(VirtIOBlockReq *req)
346 sector = ldq_p(&req->out->sector);
348 bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
350 trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
352 if (sector & req->dev->sector_mask) {
353 virtio_blk_rw_complete(req, -EIO);
356 if (req->qiov.size % req->dev->conf->logical_block_size) {
357 virtio_blk_rw_complete(req, -EIO);
360 bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
361 req->qiov.size / BDRV_SECTOR_SIZE,
362 virtio_blk_rw_complete, req);
365 static void virtio_blk_handle_request(VirtIOBlockReq *req,
370 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
371 error_report("virtio-blk missing headers");
375 if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
376 req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
377 error_report("virtio-blk header not in correct element");
381 req->out = (void *)req->elem.out_sg[0].iov_base;
382 req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
384 type = ldl_p(&req->out->type);
386 if (type & VIRTIO_BLK_T_FLUSH) {
387 virtio_blk_handle_flush(req, mrb);
388 } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
389 virtio_blk_handle_scsi(req);
390 } else if (type & VIRTIO_BLK_T_GET_ID) {
391 VirtIOBlock *s = req->dev;
394 * NB: per existing s/n string convention the string is
395 * terminated by '\0' only when shorter than buffer.
397 strncpy(req->elem.in_sg[0].iov_base,
398 s->blk->serial ? s->blk->serial : "",
399 MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
400 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
402 } else if (type & VIRTIO_BLK_T_OUT) {
403 qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
404 req->elem.out_num - 1);
405 virtio_blk_handle_write(req, mrb);
407 qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
408 req->elem.in_num - 1);
409 virtio_blk_handle_read(req);
413 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
415 VirtIOBlock *s = to_virtio_blk(vdev);
417 MultiReqBuffer mrb = {
421 while ((req = virtio_blk_get_request(s))) {
422 virtio_blk_handle_request(req, &mrb);
425 virtio_submit_multiwrite(s->bs, &mrb);
428 * FIXME: Want to check for completions before returning to guest mode,
429 * so cached reads and writes are reported as quickly as possible. But
430 * that should be done in the generic block layer.
434 static void virtio_blk_dma_restart_bh(void *opaque)
436 VirtIOBlock *s = opaque;
437 VirtIOBlockReq *req = s->rq;
438 MultiReqBuffer mrb = {
442 qemu_bh_delete(s->bh);
448 virtio_blk_handle_request(req, &mrb);
452 virtio_submit_multiwrite(s->bs, &mrb);
455 static void virtio_blk_dma_restart_cb(void *opaque, int running,
458 VirtIOBlock *s = opaque;
464 s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
465 qemu_bh_schedule(s->bh);
469 static void virtio_blk_reset(VirtIODevice *vdev)
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 = to_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, lets 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 dont 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 = to_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 = to_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 (bdrv_enable_write_cache(s->bs))
539 features |= (1 << VIRTIO_BLK_F_WCE);
541 if (bdrv_is_read_only(s->bs))
542 features |= 1 << VIRTIO_BLK_F_RO;
547 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
549 VirtIOBlock *s = to_virtio_blk(vdev);
552 if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
556 features = vdev->guest_features;
557 bdrv_set_enable_write_cache(s->bs, !!(features & (1 << VIRTIO_BLK_F_WCE)));
560 static void virtio_blk_save(QEMUFile *f, void *opaque)
562 VirtIOBlock *s = opaque;
563 VirtIOBlockReq *req = s->rq;
565 virtio_save(&s->vdev, f);
568 qemu_put_sbyte(f, 1);
569 qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
572 qemu_put_sbyte(f, 0);
575 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
577 VirtIOBlock *s = opaque;
583 ret = virtio_load(&s->vdev, f);
588 while (qemu_get_sbyte(f)) {
589 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
590 qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
594 virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
595 req->elem.in_num, 1);
596 virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
597 req->elem.out_num, 0);
603 static void virtio_blk_resize(void *opaque)
605 VirtIOBlock *s = opaque;
607 virtio_notify_config(&s->vdev);
610 static const BlockDevOps virtio_block_ops = {
611 .resize_cb = virtio_blk_resize,
614 VirtIODevice *virtio_blk_init(DeviceState *dev, VirtIOBlkConf *blk)
617 static int virtio_blk_id;
620 error_report("drive property not set");
623 if (!bdrv_is_inserted(blk->conf.bs)) {
624 error_report("Device needs media, but drive is empty");
628 blkconf_serial(&blk->conf, &blk->serial);
629 if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
633 s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
634 sizeof(struct virtio_blk_config),
635 sizeof(VirtIOBlock));
637 s->vdev.get_config = virtio_blk_update_config;
638 s->vdev.set_config = virtio_blk_set_config;
639 s->vdev.get_features = virtio_blk_get_features;
640 s->vdev.set_status = virtio_blk_set_status;
641 s->vdev.reset = virtio_blk_reset;
642 s->bs = blk->conf.bs;
643 s->conf = &blk->conf;
646 s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
648 s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
650 qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
652 register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
653 virtio_blk_save, virtio_blk_load, s);
654 bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
655 bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size);
657 bdrv_iostatus_enable(s->bs);
658 add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
663 void virtio_blk_exit(VirtIODevice *vdev)
665 VirtIOBlock *s = to_virtio_blk(vdev);
666 unregister_savevm(s->qdev, "virtio-blk", s);
667 blockdev_mark_auto_del(s->bs);
668 virtio_cleanup(vdev);
671 /******************** VirtIOBlk Device **********************/
673 static int virtio_blkdev_init(DeviceState *dev)
676 VirtIOBlockState *s = VIRTIO_BLK_FROM_QDEV(dev);
678 assert(s->trl != NULL);
680 vdev = virtio_blk_init(dev, &s->blk);
685 /* Pass default host_features to transport */
686 s->trl->host_features = s->host_features;
688 if (virtio_call_backend_init_cb(dev, s->trl, vdev) != 0) {
692 /* Binding should be ready here, let's get final features */
693 if (vdev->binding->get_features) {
694 s->host_features = vdev->binding->get_features(vdev->binding_opaque);
699 static Property virtio_blkdev_properties[] = {
700 DEFINE_BLOCK_PROPERTIES(VirtIOBlockState, blk.conf),
701 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlockState, blk.conf),
702 DEFINE_PROP_STRING("serial", VirtIOBlockState, blk.serial),
704 DEFINE_PROP_BIT("scsi", VirtIOBlockState, blk.scsi, 0, true),
706 DEFINE_PROP_BIT("config-wce", VirtIOBlockState, blk.config_wce, 0, true),
707 DEFINE_VIRTIO_BLK_FEATURES(VirtIOBlockState, host_features),
709 DEFINE_PROP_TRANSPORT("transport", VirtIOBlockState, trl),
710 DEFINE_PROP_END_OF_LIST(),
713 static void virtio_blkdev_class_init(ObjectClass *klass, void *data)
715 DeviceClass *dc = DEVICE_CLASS(klass);
716 dc->init = virtio_blkdev_init;
717 dc->props = virtio_blkdev_properties;
720 static TypeInfo virtio_blkdev_info = {
721 .name = "virtio-blk",
722 .parent = TYPE_DEVICE,
723 .instance_size = sizeof(VirtIOBlockState),
724 .class_init = virtio_blkdev_class_init,
727 static void virtio_blk_register_types(void)
729 type_register_static(&virtio_blkdev_info);
732 type_init(virtio_blk_register_types)