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"
18 #include "virtio-blk.h"
19 #include "scsi-defs.h"
24 typedef struct VirtIOBlock
33 unsigned short sector_mask;
37 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
39 return (VirtIOBlock *)vdev;
42 typedef struct VirtIOBlockReq
45 VirtQueueElement elem;
46 struct virtio_blk_inhdr *in;
47 struct virtio_blk_outhdr *out;
48 struct virtio_scsi_inhdr *scsi;
50 struct VirtIOBlockReq *next;
54 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
56 VirtIOBlock *s = req->dev;
58 trace_virtio_blk_req_complete(req, status);
60 stb_p(&req->in->status, status);
61 virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
62 virtio_notify(&s->vdev, s->vq);
65 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
68 BlockErrorAction action = bdrv_get_on_error(req->dev->bs, is_read);
69 VirtIOBlock *s = req->dev;
71 if (action == BLOCK_ERR_IGNORE) {
72 bdrv_mon_event(s->bs, BDRV_ACTION_IGNORE, is_read);
76 if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
77 || action == BLOCK_ERR_STOP_ANY) {
80 bdrv_mon_event(s->bs, BDRV_ACTION_STOP, is_read);
81 vm_stop(RUN_STATE_IO_ERROR);
82 bdrv_iostatus_set_err(s->bs, error);
84 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
85 bdrv_acct_done(s->bs, &req->acct);
87 bdrv_mon_event(s->bs, BDRV_ACTION_REPORT, is_read);
93 static void virtio_blk_rw_complete(void *opaque, int ret)
95 VirtIOBlockReq *req = opaque;
97 trace_virtio_blk_rw_complete(req, ret);
100 int is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
101 if (virtio_blk_handle_rw_error(req, -ret, is_read))
105 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
106 bdrv_acct_done(req->dev->bs, &req->acct);
110 static void virtio_blk_flush_complete(void *opaque, int ret)
112 VirtIOBlockReq *req = opaque;
115 if (virtio_blk_handle_rw_error(req, -ret, 0)) {
120 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
121 bdrv_acct_done(req->dev->bs, &req->acct);
125 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
127 VirtIOBlockReq *req = g_malloc(sizeof(*req));
134 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
136 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
139 if (!virtqueue_pop(s->vq, &req->elem)) {
149 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
151 struct sg_io_hdr hdr;
157 * We require at least one output segment each for the virtio_blk_outhdr
158 * and the SCSI command block.
160 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
161 * and the sense buffer pointer in the input segments.
163 if (req->elem.out_num < 2 || req->elem.in_num < 3) {
164 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
170 * No support for bidirection commands yet.
172 if (req->elem.out_num > 2 && req->elem.in_num > 3) {
173 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
179 * The scsi inhdr is placed in the second-to-last input segment, just
180 * before the regular inhdr.
182 req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
184 memset(&hdr, 0, sizeof(struct sg_io_hdr));
185 hdr.interface_id = 'S';
186 hdr.cmd_len = req->elem.out_sg[1].iov_len;
187 hdr.cmdp = req->elem.out_sg[1].iov_base;
190 if (req->elem.out_num > 2) {
192 * If there are more than the minimally required 2 output segments
193 * there is write payload starting from the third iovec.
195 hdr.dxfer_direction = SG_DXFER_TO_DEV;
196 hdr.iovec_count = req->elem.out_num - 2;
198 for (i = 0; i < hdr.iovec_count; i++)
199 hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
201 hdr.dxferp = req->elem.out_sg + 2;
203 } else if (req->elem.in_num > 3) {
205 * If we have more than 3 input segments the guest wants to actually
208 hdr.dxfer_direction = SG_DXFER_FROM_DEV;
209 hdr.iovec_count = req->elem.in_num - 3;
210 for (i = 0; i < hdr.iovec_count; i++)
211 hdr.dxfer_len += req->elem.in_sg[i].iov_len;
213 hdr.dxferp = req->elem.in_sg;
216 * Some SCSI commands don't actually transfer any data.
218 hdr.dxfer_direction = SG_DXFER_NONE;
221 hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
222 hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
224 ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
226 status = VIRTIO_BLK_S_UNSUPP;
228 hdr.resid = hdr.dxfer_len;
229 } else if (hdr.status) {
230 status = VIRTIO_BLK_S_IOERR;
232 status = VIRTIO_BLK_S_OK;
236 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
237 * clear the masked_status field [hence status gets cleared too, see
238 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
239 * status has occurred. However they do set DRIVER_SENSE in driver_status
240 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
242 if (hdr.status == 0 && hdr.sb_len_wr > 0) {
243 hdr.status = CHECK_CONDITION;
246 stl_p(&req->scsi->errors,
247 hdr.status | (hdr.msg_status << 8) |
248 (hdr.host_status << 16) | (hdr.driver_status << 24));
249 stl_p(&req->scsi->residual, hdr.resid);
250 stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
251 stl_p(&req->scsi->data_len, hdr.dxfer_len);
253 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);
262 #endif /* __linux__ */
264 typedef struct MultiReqBuffer {
265 BlockRequest blkreq[32];
266 unsigned int num_writes;
269 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
273 if (!mrb->num_writes) {
277 ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
279 for (i = 0; i < mrb->num_writes; i++) {
280 if (mrb->blkreq[i].error) {
281 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
289 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
291 BlockDriverAIOCB *acb;
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);
300 acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
302 virtio_blk_flush_complete(req, -EIO);
306 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
308 BlockRequest *blkreq;
311 sector = ldq_p(&req->out->sector);
313 bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
315 trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
317 if (sector & req->dev->sector_mask) {
318 virtio_blk_rw_complete(req, -EIO);
321 if (req->qiov.size % req->dev->conf->logical_block_size) {
322 virtio_blk_rw_complete(req, -EIO);
326 if (mrb->num_writes == 32) {
327 virtio_submit_multiwrite(req->dev->bs, mrb);
330 blkreq = &mrb->blkreq[mrb->num_writes];
331 blkreq->sector = sector;
332 blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
333 blkreq->qiov = &req->qiov;
334 blkreq->cb = virtio_blk_rw_complete;
335 blkreq->opaque = req;
341 static void virtio_blk_handle_read(VirtIOBlockReq *req)
343 BlockDriverAIOCB *acb;
346 sector = ldq_p(&req->out->sector);
348 bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
350 if (sector & req->dev->sector_mask) {
351 virtio_blk_rw_complete(req, -EIO);
354 if (req->qiov.size % req->dev->conf->logical_block_size) {
355 virtio_blk_rw_complete(req, -EIO);
359 acb = bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
360 req->qiov.size / BDRV_SECTOR_SIZE,
361 virtio_blk_rw_complete, req);
363 virtio_blk_rw_complete(req, -EIO);
367 static void virtio_blk_handle_request(VirtIOBlockReq *req,
372 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
373 error_report("virtio-blk missing headers");
377 if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
378 req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
379 error_report("virtio-blk header not in correct element");
383 req->out = (void *)req->elem.out_sg[0].iov_base;
384 req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
386 type = ldl_p(&req->out->type);
388 if (type & VIRTIO_BLK_T_FLUSH) {
389 virtio_blk_handle_flush(req, mrb);
390 } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
391 virtio_blk_handle_scsi(req);
392 } else if (type & VIRTIO_BLK_T_GET_ID) {
393 VirtIOBlock *s = req->dev;
396 * NB: per existing s/n string convention the string is
397 * terminated by '\0' only when shorter than buffer.
399 strncpy(req->elem.in_sg[0].iov_base,
400 s->serial ? s->serial : "",
401 MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
402 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
404 } else if (type & VIRTIO_BLK_T_OUT) {
405 qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
406 req->elem.out_num - 1);
407 virtio_blk_handle_write(req, mrb);
409 qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
410 req->elem.in_num - 1);
411 virtio_blk_handle_read(req);
415 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
417 VirtIOBlock *s = to_virtio_blk(vdev);
419 MultiReqBuffer mrb = {
423 while ((req = virtio_blk_get_request(s))) {
424 virtio_blk_handle_request(req, &mrb);
427 virtio_submit_multiwrite(s->bs, &mrb);
430 * FIXME: Want to check for completions before returning to guest mode,
431 * so cached reads and writes are reported as quickly as possible. But
432 * that should be done in the generic block layer.
436 static void virtio_blk_dma_restart_bh(void *opaque)
438 VirtIOBlock *s = opaque;
439 VirtIOBlockReq *req = s->rq;
440 MultiReqBuffer mrb = {
444 qemu_bh_delete(s->bh);
450 virtio_blk_handle_request(req, &mrb);
454 virtio_submit_multiwrite(s->bs, &mrb);
457 static void virtio_blk_dma_restart_cb(void *opaque, int running,
460 VirtIOBlock *s = opaque;
466 s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
467 qemu_bh_schedule(s->bh);
471 static void virtio_blk_reset(VirtIODevice *vdev)
474 * This should cancel pending requests, but can't do nicely until there
475 * are per-device request lists.
480 /* coalesce internal state, copy to pci i/o region 0
482 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
484 VirtIOBlock *s = to_virtio_blk(vdev);
485 struct virtio_blk_config blkcfg;
487 int cylinders, heads, secs;
488 int blk_size = s->conf->logical_block_size;
490 bdrv_get_geometry(s->bs, &capacity);
491 bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
492 memset(&blkcfg, 0, sizeof(blkcfg));
493 stq_raw(&blkcfg.capacity, capacity);
494 stl_raw(&blkcfg.seg_max, 128 - 2);
495 stw_raw(&blkcfg.cylinders, cylinders);
496 stl_raw(&blkcfg.blk_size, blk_size);
497 stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
498 stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
499 blkcfg.heads = heads;
500 blkcfg.sectors = secs & ~s->sector_mask;
502 blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
503 blkcfg.alignment_offset = 0;
504 memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
507 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
509 VirtIOBlock *s = to_virtio_blk(vdev);
511 features |= (1 << VIRTIO_BLK_F_SEG_MAX);
512 features |= (1 << VIRTIO_BLK_F_GEOMETRY);
513 features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
514 features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
516 if (bdrv_enable_write_cache(s->bs))
517 features |= (1 << VIRTIO_BLK_F_WCACHE);
519 if (bdrv_is_read_only(s->bs))
520 features |= 1 << VIRTIO_BLK_F_RO;
525 static void virtio_blk_save(QEMUFile *f, void *opaque)
527 VirtIOBlock *s = opaque;
528 VirtIOBlockReq *req = s->rq;
530 virtio_save(&s->vdev, f);
533 qemu_put_sbyte(f, 1);
534 qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
537 qemu_put_sbyte(f, 0);
540 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
542 VirtIOBlock *s = opaque;
547 virtio_load(&s->vdev, f);
548 while (qemu_get_sbyte(f)) {
549 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
550 qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
554 virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
555 req->elem.in_num, 1);
556 virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
557 req->elem.out_num, 0);
563 static void virtio_blk_resize(void *opaque)
565 VirtIOBlock *s = opaque;
567 virtio_notify_config(&s->vdev);
570 static const BlockDevOps virtio_block_ops = {
571 .resize_cb = virtio_blk_resize,
574 VirtIODevice *virtio_blk_init(DeviceState *dev, BlockConf *conf,
578 int cylinders, heads, secs;
579 static int virtio_blk_id;
583 error_report("virtio-blk-pci: drive property not set");
586 if (!bdrv_is_inserted(conf->bs)) {
587 error_report("Device needs media, but drive is empty");
592 /* try to fall back to value set with legacy -drive serial=... */
593 dinfo = drive_get_by_blockdev(conf->bs);
594 if (*dinfo->serial) {
595 *serial = strdup(dinfo->serial);
599 s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
600 sizeof(struct virtio_blk_config),
601 sizeof(VirtIOBlock));
603 s->vdev.get_config = virtio_blk_update_config;
604 s->vdev.get_features = virtio_blk_get_features;
605 s->vdev.reset = virtio_blk_reset;
610 s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
611 bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
613 s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
615 qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
617 register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
618 virtio_blk_save, virtio_blk_load, s);
619 bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
620 bdrv_set_buffer_alignment(s->bs, conf->logical_block_size);
622 bdrv_iostatus_enable(s->bs);
623 add_boot_device_path(conf->bootindex, dev, "/disk@0,0");
628 void virtio_blk_exit(VirtIODevice *vdev)
630 VirtIOBlock *s = to_virtio_blk(vdev);
631 unregister_savevm(s->qdev, "virtio-blk", s);
632 virtio_cleanup(vdev);