xref: /openbmc/qemu/hw/block/virtio-blk.c (revision 679cb8e1)
1 /*
2  * Virtio Block Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu-common.h"
17 #include "qemu/iov.h"
18 #include "qemu/error-report.h"
19 #include "trace.h"
20 #include "hw/block/block.h"
21 #include "sysemu/blockdev.h"
22 #include "hw/virtio/virtio-blk.h"
23 #include "dataplane/virtio-blk.h"
24 #include "scsi/constants.h"
25 #ifdef __linux__
26 # include <scsi/sg.h>
27 #endif
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/virtio/virtio-access.h"
30 
31 /* Config size before the discard support (hide associated config fields) */
32 #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \
33                                      max_discard_sectors)
34 /*
35  * Starting from the discard feature, we can use this array to properly
36  * set the config size depending on the features enabled.
37  */
38 static VirtIOFeature feature_sizes[] = {
39     {.flags = 1ULL << VIRTIO_BLK_F_DISCARD,
40      .end = virtio_endof(struct virtio_blk_config, discard_sector_alignment)},
41     {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES,
42      .end = virtio_endof(struct virtio_blk_config, write_zeroes_may_unmap)},
43     {}
44 };
45 
46 static void virtio_blk_set_config_size(VirtIOBlock *s, uint64_t host_features)
47 {
48     s->config_size = MAX(VIRTIO_BLK_CFG_SIZE,
49         virtio_feature_get_config_size(feature_sizes, host_features));
50 
51     assert(s->config_size <= sizeof(struct virtio_blk_config));
52 }
53 
54 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
55                                     VirtIOBlockReq *req)
56 {
57     req->dev = s;
58     req->vq = vq;
59     req->qiov.size = 0;
60     req->in_len = 0;
61     req->next = NULL;
62     req->mr_next = NULL;
63 }
64 
65 static void virtio_blk_free_request(VirtIOBlockReq *req)
66 {
67     g_free(req);
68 }
69 
70 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
71 {
72     VirtIOBlock *s = req->dev;
73     VirtIODevice *vdev = VIRTIO_DEVICE(s);
74 
75     trace_virtio_blk_req_complete(vdev, req, status);
76 
77     stb_p(&req->in->status, status);
78     virtqueue_push(req->vq, &req->elem, req->in_len);
79     if (s->dataplane_started && !s->dataplane_disabled) {
80         virtio_blk_data_plane_notify(s->dataplane, req->vq);
81     } else {
82         virtio_notify(vdev, req->vq);
83     }
84 }
85 
86 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
87     bool is_read, bool acct_failed)
88 {
89     VirtIOBlock *s = req->dev;
90     BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
91 
92     if (action == BLOCK_ERROR_ACTION_STOP) {
93         /* Break the link as the next request is going to be parsed from the
94          * ring again. Otherwise we may end up doing a double completion! */
95         req->mr_next = NULL;
96         req->next = s->rq;
97         s->rq = req;
98     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
99         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
100         if (acct_failed) {
101             block_acct_failed(blk_get_stats(s->blk), &req->acct);
102         }
103         virtio_blk_free_request(req);
104     }
105 
106     blk_error_action(s->blk, action, is_read, error);
107     return action != BLOCK_ERROR_ACTION_IGNORE;
108 }
109 
110 static void virtio_blk_rw_complete(void *opaque, int ret)
111 {
112     VirtIOBlockReq *next = opaque;
113     VirtIOBlock *s = next->dev;
114     VirtIODevice *vdev = VIRTIO_DEVICE(s);
115 
116     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
117     while (next) {
118         VirtIOBlockReq *req = next;
119         next = req->mr_next;
120         trace_virtio_blk_rw_complete(vdev, req, ret);
121 
122         if (req->qiov.nalloc != -1) {
123             /* If nalloc is != -1 req->qiov is a local copy of the original
124              * external iovec. It was allocated in submit_requests to be
125              * able to merge requests. */
126             qemu_iovec_destroy(&req->qiov);
127         }
128 
129         if (ret) {
130             int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type);
131             bool is_read = !(p & VIRTIO_BLK_T_OUT);
132             /* Note that memory may be dirtied on read failure.  If the
133              * virtio request is not completed here, as is the case for
134              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
135              * correctly during live migration.  While this is ugly,
136              * it is acceptable because the device is free to write to
137              * the memory until the request is completed (which will
138              * happen on the other side of the migration).
139              */
140             if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
141                 continue;
142             }
143         }
144 
145         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
146         block_acct_done(blk_get_stats(s->blk), &req->acct);
147         virtio_blk_free_request(req);
148     }
149     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
150 }
151 
152 static void virtio_blk_flush_complete(void *opaque, int ret)
153 {
154     VirtIOBlockReq *req = opaque;
155     VirtIOBlock *s = req->dev;
156 
157     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
158     if (ret) {
159         if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
160             goto out;
161         }
162     }
163 
164     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
165     block_acct_done(blk_get_stats(s->blk), &req->acct);
166     virtio_blk_free_request(req);
167 
168 out:
169     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
170 }
171 
172 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
173 {
174     VirtIOBlockReq *req = opaque;
175     VirtIOBlock *s = req->dev;
176     bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
177                             ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
178 
179     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
180     if (ret) {
181         if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
182             goto out;
183         }
184     }
185 
186     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
187     if (is_write_zeroes) {
188         block_acct_done(blk_get_stats(s->blk), &req->acct);
189     }
190     virtio_blk_free_request(req);
191 
192 out:
193     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
194 }
195 
196 #ifdef __linux__
197 
198 typedef struct {
199     VirtIOBlockReq *req;
200     struct sg_io_hdr hdr;
201 } VirtIOBlockIoctlReq;
202 
203 static void virtio_blk_ioctl_complete(void *opaque, int status)
204 {
205     VirtIOBlockIoctlReq *ioctl_req = opaque;
206     VirtIOBlockReq *req = ioctl_req->req;
207     VirtIOBlock *s = req->dev;
208     VirtIODevice *vdev = VIRTIO_DEVICE(s);
209     struct virtio_scsi_inhdr *scsi;
210     struct sg_io_hdr *hdr;
211 
212     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
213 
214     if (status) {
215         status = VIRTIO_BLK_S_UNSUPP;
216         virtio_stl_p(vdev, &scsi->errors, 255);
217         goto out;
218     }
219 
220     hdr = &ioctl_req->hdr;
221     /*
222      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
223      * clear the masked_status field [hence status gets cleared too, see
224      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
225      * status has occurred.  However they do set DRIVER_SENSE in driver_status
226      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
227      */
228     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
229         hdr->status = CHECK_CONDITION;
230     }
231 
232     virtio_stl_p(vdev, &scsi->errors,
233                  hdr->status | (hdr->msg_status << 8) |
234                  (hdr->host_status << 16) | (hdr->driver_status << 24));
235     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
236     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
237     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
238 
239 out:
240     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
241     virtio_blk_req_complete(req, status);
242     virtio_blk_free_request(req);
243     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
244     g_free(ioctl_req);
245 }
246 
247 #endif
248 
249 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
250 {
251     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
252 
253     if (req) {
254         virtio_blk_init_request(s, vq, req);
255     }
256     return req;
257 }
258 
259 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
260 {
261     int status = VIRTIO_BLK_S_OK;
262     struct virtio_scsi_inhdr *scsi = NULL;
263     VirtIOBlock *blk = req->dev;
264     VirtIODevice *vdev = VIRTIO_DEVICE(blk);
265     VirtQueueElement *elem = &req->elem;
266 
267 #ifdef __linux__
268     int i;
269     VirtIOBlockIoctlReq *ioctl_req;
270     BlockAIOCB *acb;
271 #endif
272 
273     /*
274      * We require at least one output segment each for the virtio_blk_outhdr
275      * and the SCSI command block.
276      *
277      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
278      * and the sense buffer pointer in the input segments.
279      */
280     if (elem->out_num < 2 || elem->in_num < 3) {
281         status = VIRTIO_BLK_S_IOERR;
282         goto fail;
283     }
284 
285     /*
286      * The scsi inhdr is placed in the second-to-last input segment, just
287      * before the regular inhdr.
288      */
289     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
290 
291     if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
292         status = VIRTIO_BLK_S_UNSUPP;
293         goto fail;
294     }
295 
296     /*
297      * No support for bidirection commands yet.
298      */
299     if (elem->out_num > 2 && elem->in_num > 3) {
300         status = VIRTIO_BLK_S_UNSUPP;
301         goto fail;
302     }
303 
304 #ifdef __linux__
305     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
306     ioctl_req->req = req;
307     ioctl_req->hdr.interface_id = 'S';
308     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
309     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
310     ioctl_req->hdr.dxfer_len = 0;
311 
312     if (elem->out_num > 2) {
313         /*
314          * If there are more than the minimally required 2 output segments
315          * there is write payload starting from the third iovec.
316          */
317         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
318         ioctl_req->hdr.iovec_count = elem->out_num - 2;
319 
320         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
321             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
322         }
323 
324         ioctl_req->hdr.dxferp = elem->out_sg + 2;
325 
326     } else if (elem->in_num > 3) {
327         /*
328          * If we have more than 3 input segments the guest wants to actually
329          * read data.
330          */
331         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
332         ioctl_req->hdr.iovec_count = elem->in_num - 3;
333         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
334             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
335         }
336 
337         ioctl_req->hdr.dxferp = elem->in_sg;
338     } else {
339         /*
340          * Some SCSI commands don't actually transfer any data.
341          */
342         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
343     }
344 
345     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
346     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
347 
348     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
349                         virtio_blk_ioctl_complete, ioctl_req);
350     if (!acb) {
351         g_free(ioctl_req);
352         status = VIRTIO_BLK_S_UNSUPP;
353         goto fail;
354     }
355     return -EINPROGRESS;
356 #else
357     abort();
358 #endif
359 
360 fail:
361     /* Just put anything nonzero so that the ioctl fails in the guest.  */
362     if (scsi) {
363         virtio_stl_p(vdev, &scsi->errors, 255);
364     }
365     return status;
366 }
367 
368 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
369 {
370     int status;
371 
372     status = virtio_blk_handle_scsi_req(req);
373     if (status != -EINPROGRESS) {
374         virtio_blk_req_complete(req, status);
375         virtio_blk_free_request(req);
376     }
377 }
378 
379 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
380                                    int start, int num_reqs, int niov)
381 {
382     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
383     int64_t sector_num = mrb->reqs[start]->sector_num;
384     bool is_write = mrb->is_write;
385 
386     if (num_reqs > 1) {
387         int i;
388         struct iovec *tmp_iov = qiov->iov;
389         int tmp_niov = qiov->niov;
390 
391         /* mrb->reqs[start]->qiov was initialized from external so we can't
392          * modify it here. We need to initialize it locally and then add the
393          * external iovecs. */
394         qemu_iovec_init(qiov, niov);
395 
396         for (i = 0; i < tmp_niov; i++) {
397             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
398         }
399 
400         for (i = start + 1; i < start + num_reqs; i++) {
401             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
402                               mrb->reqs[i]->qiov.size);
403             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
404         }
405 
406         trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
407                                          mrb, start, num_reqs,
408                                          sector_num << BDRV_SECTOR_BITS,
409                                          qiov->size, is_write);
410         block_acct_merge_done(blk_get_stats(blk),
411                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
412                               num_reqs - 1);
413     }
414 
415     if (is_write) {
416         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
417                         virtio_blk_rw_complete, mrb->reqs[start]);
418     } else {
419         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
420                        virtio_blk_rw_complete, mrb->reqs[start]);
421     }
422 }
423 
424 static int multireq_compare(const void *a, const void *b)
425 {
426     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
427                          *req2 = *(VirtIOBlockReq **)b;
428 
429     /*
430      * Note that we can't simply subtract sector_num1 from sector_num2
431      * here as that could overflow the return value.
432      */
433     if (req1->sector_num > req2->sector_num) {
434         return 1;
435     } else if (req1->sector_num < req2->sector_num) {
436         return -1;
437     } else {
438         return 0;
439     }
440 }
441 
442 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
443 {
444     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
445     uint32_t max_transfer;
446     int64_t sector_num = 0;
447 
448     if (mrb->num_reqs == 1) {
449         submit_requests(blk, mrb, 0, 1, -1);
450         mrb->num_reqs = 0;
451         return;
452     }
453 
454     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
455 
456     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
457           &multireq_compare);
458 
459     for (i = 0; i < mrb->num_reqs; i++) {
460         VirtIOBlockReq *req = mrb->reqs[i];
461         if (num_reqs > 0) {
462             /*
463              * NOTE: We cannot merge the requests in below situations:
464              * 1. requests are not sequential
465              * 2. merge would exceed maximum number of IOVs
466              * 3. merge would exceed maximum transfer length of backend device
467              */
468             if (sector_num + nb_sectors != req->sector_num ||
469                 niov > blk_get_max_iov(blk) - req->qiov.niov ||
470                 req->qiov.size > max_transfer ||
471                 nb_sectors > (max_transfer -
472                               req->qiov.size) / BDRV_SECTOR_SIZE) {
473                 submit_requests(blk, mrb, start, num_reqs, niov);
474                 num_reqs = 0;
475             }
476         }
477 
478         if (num_reqs == 0) {
479             sector_num = req->sector_num;
480             nb_sectors = niov = 0;
481             start = i;
482         }
483 
484         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
485         niov += req->qiov.niov;
486         num_reqs++;
487     }
488 
489     submit_requests(blk, mrb, start, num_reqs, niov);
490     mrb->num_reqs = 0;
491 }
492 
493 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
494 {
495     VirtIOBlock *s = req->dev;
496 
497     block_acct_start(blk_get_stats(s->blk), &req->acct, 0,
498                      BLOCK_ACCT_FLUSH);
499 
500     /*
501      * Make sure all outstanding writes are posted to the backing device.
502      */
503     if (mrb->is_write && mrb->num_reqs > 0) {
504         virtio_blk_submit_multireq(s->blk, mrb);
505     }
506     blk_aio_flush(s->blk, virtio_blk_flush_complete, req);
507 }
508 
509 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
510                                      uint64_t sector, size_t size)
511 {
512     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
513     uint64_t total_sectors;
514 
515     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
516         return false;
517     }
518     if (sector & dev->sector_mask) {
519         return false;
520     }
521     if (size % dev->conf.conf.logical_block_size) {
522         return false;
523     }
524     blk_get_geometry(dev->blk, &total_sectors);
525     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
526         return false;
527     }
528     return true;
529 }
530 
531 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
532     struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
533 {
534     VirtIOBlock *s = req->dev;
535     VirtIODevice *vdev = VIRTIO_DEVICE(s);
536     uint64_t sector;
537     uint32_t num_sectors, flags, max_sectors;
538     uint8_t err_status;
539     int bytes;
540 
541     sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
542     num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
543     flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
544     max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
545                   s->conf.max_discard_sectors;
546 
547     /*
548      * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
549      * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
550      * the integer variable.
551      */
552     if (unlikely(num_sectors > max_sectors)) {
553         err_status = VIRTIO_BLK_S_IOERR;
554         goto err;
555     }
556 
557     bytes = num_sectors << BDRV_SECTOR_BITS;
558 
559     if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
560         err_status = VIRTIO_BLK_S_IOERR;
561         goto err;
562     }
563 
564     /*
565      * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
566      * and write zeroes commands if any unknown flag is set.
567      */
568     if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
569         err_status = VIRTIO_BLK_S_UNSUPP;
570         goto err;
571     }
572 
573     if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
574         int blk_aio_flags = 0;
575 
576         if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
577             blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
578         }
579 
580         block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
581                          BLOCK_ACCT_WRITE);
582 
583         blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
584                               bytes, blk_aio_flags,
585                               virtio_blk_discard_write_zeroes_complete, req);
586     } else { /* VIRTIO_BLK_T_DISCARD */
587         /*
588          * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
589          * discard commands if the unmap flag is set.
590          */
591         if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
592             err_status = VIRTIO_BLK_S_UNSUPP;
593             goto err;
594         }
595 
596         blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
597                          virtio_blk_discard_write_zeroes_complete, req);
598     }
599 
600     return VIRTIO_BLK_S_OK;
601 
602 err:
603     if (is_write_zeroes) {
604         block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
605     }
606     return err_status;
607 }
608 
609 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
610 {
611     uint32_t type;
612     struct iovec *in_iov = req->elem.in_sg;
613     struct iovec *out_iov = req->elem.out_sg;
614     unsigned in_num = req->elem.in_num;
615     unsigned out_num = req->elem.out_num;
616     VirtIOBlock *s = req->dev;
617     VirtIODevice *vdev = VIRTIO_DEVICE(s);
618 
619     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
620         virtio_error(vdev, "virtio-blk missing headers");
621         return -1;
622     }
623 
624     if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
625                             sizeof(req->out)) != sizeof(req->out))) {
626         virtio_error(vdev, "virtio-blk request outhdr too short");
627         return -1;
628     }
629 
630     iov_discard_front(&out_iov, &out_num, sizeof(req->out));
631 
632     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
633         virtio_error(vdev, "virtio-blk request inhdr too short");
634         return -1;
635     }
636 
637     /* We always touch the last byte, so just see how big in_iov is.  */
638     req->in_len = iov_size(in_iov, in_num);
639     req->in = (void *)in_iov[in_num - 1].iov_base
640               + in_iov[in_num - 1].iov_len
641               - sizeof(struct virtio_blk_inhdr);
642     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
643 
644     type = virtio_ldl_p(vdev, &req->out.type);
645 
646     /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
647      * is an optional flag. Although a guest should not send this flag if
648      * not negotiated we ignored it in the past. So keep ignoring it. */
649     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
650     case VIRTIO_BLK_T_IN:
651     {
652         bool is_write = type & VIRTIO_BLK_T_OUT;
653         req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
654 
655         if (is_write) {
656             qemu_iovec_init_external(&req->qiov, out_iov, out_num);
657             trace_virtio_blk_handle_write(vdev, req, req->sector_num,
658                                           req->qiov.size / BDRV_SECTOR_SIZE);
659         } else {
660             qemu_iovec_init_external(&req->qiov, in_iov, in_num);
661             trace_virtio_blk_handle_read(vdev, req, req->sector_num,
662                                          req->qiov.size / BDRV_SECTOR_SIZE);
663         }
664 
665         if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
666             virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
667             block_acct_invalid(blk_get_stats(s->blk),
668                                is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
669             virtio_blk_free_request(req);
670             return 0;
671         }
672 
673         block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
674                          is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
675 
676         /* merge would exceed maximum number of requests or IO direction
677          * changes */
678         if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
679                                   is_write != mrb->is_write ||
680                                   !s->conf.request_merging)) {
681             virtio_blk_submit_multireq(s->blk, mrb);
682         }
683 
684         assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
685         mrb->reqs[mrb->num_reqs++] = req;
686         mrb->is_write = is_write;
687         break;
688     }
689     case VIRTIO_BLK_T_FLUSH:
690         virtio_blk_handle_flush(req, mrb);
691         break;
692     case VIRTIO_BLK_T_SCSI_CMD:
693         virtio_blk_handle_scsi(req);
694         break;
695     case VIRTIO_BLK_T_GET_ID:
696     {
697         /*
698          * NB: per existing s/n string convention the string is
699          * terminated by '\0' only when shorter than buffer.
700          */
701         const char *serial = s->conf.serial ? s->conf.serial : "";
702         size_t size = MIN(strlen(serial) + 1,
703                           MIN(iov_size(in_iov, in_num),
704                               VIRTIO_BLK_ID_BYTES));
705         iov_from_buf(in_iov, in_num, 0, serial, size);
706         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
707         virtio_blk_free_request(req);
708         break;
709     }
710     /*
711      * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
712      * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement,
713      * so we must mask it for these requests, then we will check if it is set.
714      */
715     case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT:
716     case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT:
717     {
718         struct virtio_blk_discard_write_zeroes dwz_hdr;
719         size_t out_len = iov_size(out_iov, out_num);
720         bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) ==
721                                VIRTIO_BLK_T_WRITE_ZEROES;
722         uint8_t err_status;
723 
724         /*
725          * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains
726          * more than one segment.
727          */
728         if (unlikely(!(type & VIRTIO_BLK_T_OUT) ||
729                      out_len > sizeof(dwz_hdr))) {
730             virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
731             virtio_blk_free_request(req);
732             return 0;
733         }
734 
735         if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr,
736                                 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) {
737             virtio_error(vdev, "virtio-blk discard/write_zeroes header"
738                          " too short");
739             return -1;
740         }
741 
742         err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
743                                                             is_write_zeroes);
744         if (err_status != VIRTIO_BLK_S_OK) {
745             virtio_blk_req_complete(req, err_status);
746             virtio_blk_free_request(req);
747         }
748 
749         break;
750     }
751     default:
752         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
753         virtio_blk_free_request(req);
754     }
755     return 0;
756 }
757 
758 bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
759 {
760     VirtIOBlockReq *req;
761     MultiReqBuffer mrb = {};
762     bool progress = false;
763 
764     aio_context_acquire(blk_get_aio_context(s->blk));
765     blk_io_plug(s->blk);
766 
767     do {
768         virtio_queue_set_notification(vq, 0);
769 
770         while ((req = virtio_blk_get_request(s, vq))) {
771             progress = true;
772             if (virtio_blk_handle_request(req, &mrb)) {
773                 virtqueue_detach_element(req->vq, &req->elem, 0);
774                 virtio_blk_free_request(req);
775                 break;
776             }
777         }
778 
779         virtio_queue_set_notification(vq, 1);
780     } while (!virtio_queue_empty(vq));
781 
782     if (mrb.num_reqs) {
783         virtio_blk_submit_multireq(s->blk, &mrb);
784     }
785 
786     blk_io_unplug(s->blk);
787     aio_context_release(blk_get_aio_context(s->blk));
788     return progress;
789 }
790 
791 static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq)
792 {
793     virtio_blk_handle_vq(s, vq);
794 }
795 
796 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
797 {
798     VirtIOBlock *s = (VirtIOBlock *)vdev;
799 
800     if (s->dataplane) {
801         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
802          * dataplane here instead of waiting for .set_status().
803          */
804         virtio_device_start_ioeventfd(vdev);
805         if (!s->dataplane_disabled) {
806             return;
807         }
808     }
809     virtio_blk_handle_output_do(s, vq);
810 }
811 
812 static void virtio_blk_dma_restart_bh(void *opaque)
813 {
814     VirtIOBlock *s = opaque;
815     VirtIOBlockReq *req = s->rq;
816     MultiReqBuffer mrb = {};
817 
818     qemu_bh_delete(s->bh);
819     s->bh = NULL;
820 
821     s->rq = NULL;
822 
823     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
824     while (req) {
825         VirtIOBlockReq *next = req->next;
826         if (virtio_blk_handle_request(req, &mrb)) {
827             /* Device is now broken and won't do any processing until it gets
828              * reset. Already queued requests will be lost: let's purge them.
829              */
830             while (req) {
831                 next = req->next;
832                 virtqueue_detach_element(req->vq, &req->elem, 0);
833                 virtio_blk_free_request(req);
834                 req = next;
835             }
836             break;
837         }
838         req = next;
839     }
840 
841     if (mrb.num_reqs) {
842         virtio_blk_submit_multireq(s->blk, &mrb);
843     }
844     blk_dec_in_flight(s->conf.conf.blk);
845     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
846 }
847 
848 static void virtio_blk_dma_restart_cb(void *opaque, int running,
849                                       RunState state)
850 {
851     VirtIOBlock *s = opaque;
852 
853     if (!running) {
854         return;
855     }
856 
857     if (!s->bh) {
858         /* FIXME The data plane is not started yet, so these requests are
859          * processed in the main thread. */
860         s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
861                            virtio_blk_dma_restart_bh, s);
862         blk_inc_in_flight(s->conf.conf.blk);
863         qemu_bh_schedule(s->bh);
864     }
865 }
866 
867 static void virtio_blk_reset(VirtIODevice *vdev)
868 {
869     VirtIOBlock *s = VIRTIO_BLK(vdev);
870     AioContext *ctx;
871     VirtIOBlockReq *req;
872 
873     ctx = blk_get_aio_context(s->blk);
874     aio_context_acquire(ctx);
875     blk_drain(s->blk);
876 
877     /* We drop queued requests after blk_drain() because blk_drain() itself can
878      * produce them. */
879     while (s->rq) {
880         req = s->rq;
881         s->rq = req->next;
882         virtqueue_detach_element(req->vq, &req->elem, 0);
883         virtio_blk_free_request(req);
884     }
885 
886     aio_context_release(ctx);
887 
888     assert(!s->dataplane_started);
889     blk_set_enable_write_cache(s->blk, s->original_wce);
890 }
891 
892 /* coalesce internal state, copy to pci i/o region 0
893  */
894 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
895 {
896     VirtIOBlock *s = VIRTIO_BLK(vdev);
897     BlockConf *conf = &s->conf.conf;
898     struct virtio_blk_config blkcfg;
899     uint64_t capacity;
900     int64_t length;
901     int blk_size = conf->logical_block_size;
902 
903     blk_get_geometry(s->blk, &capacity);
904     memset(&blkcfg, 0, sizeof(blkcfg));
905     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
906     virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2);
907     virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
908     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
909     virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
910     virtio_stw_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
911     blkcfg.geometry.heads = conf->heads;
912     /*
913      * We must ensure that the block device capacity is a multiple of
914      * the logical block size. If that is not the case, let's use
915      * sector_mask to adopt the geometry to have a correct picture.
916      * For those devices where the capacity is ok for the given geometry
917      * we don't touch the sector value of the geometry, since some devices
918      * (like s390 dasd) need a specific value. Here the capacity is already
919      * cyls*heads*secs*blk_size and the sector value is not block size
920      * divided by 512 - instead it is the amount of blk_size blocks
921      * per track (cylinder).
922      */
923     length = blk_getlength(s->blk);
924     if (length > 0 && length / conf->heads / conf->secs % blk_size) {
925         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
926     } else {
927         blkcfg.geometry.sectors = conf->secs;
928     }
929     blkcfg.size_max = 0;
930     blkcfg.physical_block_exp = get_physical_block_exp(conf);
931     blkcfg.alignment_offset = 0;
932     blkcfg.wce = blk_enable_write_cache(s->blk);
933     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
934     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
935         virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
936                      s->conf.max_discard_sectors);
937         virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
938                      blk_size >> BDRV_SECTOR_BITS);
939         /*
940          * We support only one segment per request since multiple segments
941          * are not widely used and there are no userspace APIs that allow
942          * applications to submit multiple segments in a single call.
943          */
944         virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
945     }
946     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
947         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
948                      s->conf.max_write_zeroes_sectors);
949         blkcfg.write_zeroes_may_unmap = 1;
950         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
951     }
952     memcpy(config, &blkcfg, s->config_size);
953 }
954 
955 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
956 {
957     VirtIOBlock *s = VIRTIO_BLK(vdev);
958     struct virtio_blk_config blkcfg;
959 
960     memcpy(&blkcfg, config, s->config_size);
961 
962     aio_context_acquire(blk_get_aio_context(s->blk));
963     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
964     aio_context_release(blk_get_aio_context(s->blk));
965 }
966 
967 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
968                                         Error **errp)
969 {
970     VirtIOBlock *s = VIRTIO_BLK(vdev);
971 
972     /* Firstly sync all virtio-blk possible supported features */
973     features |= s->host_features;
974 
975     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
976     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
977     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
978     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
979     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
980         if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
981             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
982             return 0;
983         }
984     } else {
985         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
986         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
987     }
988 
989     if (blk_enable_write_cache(s->blk)) {
990         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
991     }
992     if (blk_is_read_only(s->blk)) {
993         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
994     }
995     if (s->conf.num_queues > 1) {
996         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
997     }
998 
999     return features;
1000 }
1001 
1002 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1003 {
1004     VirtIOBlock *s = VIRTIO_BLK(vdev);
1005 
1006     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1007         assert(!s->dataplane_started);
1008     }
1009 
1010     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1011         return;
1012     }
1013 
1014     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1015      * cache flushes.  Thus, the "auto writethrough" behavior is never
1016      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1017      * Leaving it enabled would break the following sequence:
1018      *
1019      *     Guest started with "-drive cache=writethrough"
1020      *     Guest sets status to 0
1021      *     Guest sets DRIVER bit in status field
1022      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
1023      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
1024      *     Guest writes 1 to the WCE configuration field (writeback mode)
1025      *     Guest sets DRIVER_OK bit in status field
1026      *
1027      * s->blk would erroneously be placed in writethrough mode.
1028      */
1029     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
1030         aio_context_acquire(blk_get_aio_context(s->blk));
1031         blk_set_enable_write_cache(s->blk,
1032                                    virtio_vdev_has_feature(vdev,
1033                                                            VIRTIO_BLK_F_WCE));
1034         aio_context_release(blk_get_aio_context(s->blk));
1035     }
1036 }
1037 
1038 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1039 {
1040     VirtIOBlock *s = VIRTIO_BLK(vdev);
1041     VirtIOBlockReq *req = s->rq;
1042 
1043     while (req) {
1044         qemu_put_sbyte(f, 1);
1045 
1046         if (s->conf.num_queues > 1) {
1047             qemu_put_be32(f, virtio_get_queue_index(req->vq));
1048         }
1049 
1050         qemu_put_virtqueue_element(f, &req->elem);
1051         req = req->next;
1052     }
1053     qemu_put_sbyte(f, 0);
1054 }
1055 
1056 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1057                                   int version_id)
1058 {
1059     VirtIOBlock *s = VIRTIO_BLK(vdev);
1060 
1061     while (qemu_get_sbyte(f)) {
1062         unsigned nvqs = s->conf.num_queues;
1063         unsigned vq_idx = 0;
1064         VirtIOBlockReq *req;
1065 
1066         if (nvqs > 1) {
1067             vq_idx = qemu_get_be32(f);
1068 
1069             if (vq_idx >= nvqs) {
1070                 error_report("Invalid virtqueue index in request list: %#x",
1071                              vq_idx);
1072                 return -EINVAL;
1073             }
1074         }
1075 
1076         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
1077         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
1078         req->next = s->rq;
1079         s->rq = req;
1080     }
1081 
1082     return 0;
1083 }
1084 
1085 static void virtio_blk_resize(void *opaque)
1086 {
1087     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1088 
1089     virtio_notify_config(vdev);
1090 }
1091 
1092 static const BlockDevOps virtio_block_ops = {
1093     .resize_cb = virtio_blk_resize,
1094 };
1095 
1096 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
1097 {
1098     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1099     VirtIOBlock *s = VIRTIO_BLK(dev);
1100     VirtIOBlkConf *conf = &s->conf;
1101     Error *err = NULL;
1102     unsigned i;
1103 
1104     if (!conf->conf.blk) {
1105         error_setg(errp, "drive property not set");
1106         return;
1107     }
1108     if (!blk_is_inserted(conf->conf.blk)) {
1109         error_setg(errp, "Device needs media, but drive is empty");
1110         return;
1111     }
1112     if (!conf->num_queues) {
1113         error_setg(errp, "num-queues property must be larger than 0");
1114         return;
1115     }
1116     if (!is_power_of_2(conf->queue_size) ||
1117         conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1118         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1119                    "must be a power of 2 (max %d)",
1120                    conf->queue_size, VIRTQUEUE_MAX_SIZE);
1121         return;
1122     }
1123 
1124     if (!blkconf_apply_backend_options(&conf->conf,
1125                                        blk_is_read_only(conf->conf.blk), true,
1126                                        errp)) {
1127         return;
1128     }
1129     s->original_wce = blk_enable_write_cache(conf->conf.blk);
1130     if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
1131         return;
1132     }
1133 
1134     blkconf_blocksizes(&conf->conf);
1135 
1136     if (conf->conf.logical_block_size >
1137         conf->conf.physical_block_size) {
1138         error_setg(errp,
1139                    "logical_block_size > physical_block_size not supported");
1140         return;
1141     }
1142 
1143     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1144         (!conf->max_discard_sectors ||
1145          conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1146         error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1147                    ", must be between 1 and %d",
1148                    conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1149         return;
1150     }
1151 
1152     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1153         (!conf->max_write_zeroes_sectors ||
1154          conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1155         error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1156                    "), must be between 1 and %d",
1157                    conf->max_write_zeroes_sectors,
1158                    (int)BDRV_REQUEST_MAX_SECTORS);
1159         return;
1160     }
1161 
1162     virtio_blk_set_config_size(s, s->host_features);
1163 
1164     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, s->config_size);
1165 
1166     s->blk = conf->conf.blk;
1167     s->rq = NULL;
1168     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
1169 
1170     for (i = 0; i < conf->num_queues; i++) {
1171         virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
1172     }
1173     virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
1174     if (err != NULL) {
1175         error_propagate(errp, err);
1176         virtio_cleanup(vdev);
1177         return;
1178     }
1179 
1180     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
1181     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
1182     blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
1183 
1184     blk_iostatus_enable(s->blk);
1185 }
1186 
1187 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
1188 {
1189     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1190     VirtIOBlock *s = VIRTIO_BLK(dev);
1191 
1192     virtio_blk_data_plane_destroy(s->dataplane);
1193     s->dataplane = NULL;
1194     qemu_del_vm_change_state_handler(s->change);
1195     blockdev_mark_auto_del(s->blk);
1196     virtio_cleanup(vdev);
1197 }
1198 
1199 static void virtio_blk_instance_init(Object *obj)
1200 {
1201     VirtIOBlock *s = VIRTIO_BLK(obj);
1202 
1203     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
1204                                   "bootindex", "/disk@0,0",
1205                                   DEVICE(obj), NULL);
1206 }
1207 
1208 static const VMStateDescription vmstate_virtio_blk = {
1209     .name = "virtio-blk",
1210     .minimum_version_id = 2,
1211     .version_id = 2,
1212     .fields = (VMStateField[]) {
1213         VMSTATE_VIRTIO_DEVICE,
1214         VMSTATE_END_OF_LIST()
1215     },
1216 };
1217 
1218 static Property virtio_blk_properties[] = {
1219     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1220     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1221     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1222     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1223     DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1224                       VIRTIO_BLK_F_CONFIG_WCE, true),
1225 #ifdef __linux__
1226     DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1227                       VIRTIO_BLK_F_SCSI, false),
1228 #endif
1229     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1230                     true),
1231     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
1232     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 128),
1233     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1234                      IOThread *),
1235     DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1236                       VIRTIO_BLK_F_DISCARD, true),
1237     DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1238                       VIRTIO_BLK_F_WRITE_ZEROES, true),
1239     DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
1240                        conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
1241     DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
1242                        conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
1243     DEFINE_PROP_END_OF_LIST(),
1244 };
1245 
1246 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1247 {
1248     DeviceClass *dc = DEVICE_CLASS(klass);
1249     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1250 
1251     dc->props = virtio_blk_properties;
1252     dc->vmsd = &vmstate_virtio_blk;
1253     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1254     vdc->realize = virtio_blk_device_realize;
1255     vdc->unrealize = virtio_blk_device_unrealize;
1256     vdc->get_config = virtio_blk_update_config;
1257     vdc->set_config = virtio_blk_set_config;
1258     vdc->get_features = virtio_blk_get_features;
1259     vdc->set_status = virtio_blk_set_status;
1260     vdc->reset = virtio_blk_reset;
1261     vdc->save = virtio_blk_save_device;
1262     vdc->load = virtio_blk_load_device;
1263     vdc->start_ioeventfd = virtio_blk_data_plane_start;
1264     vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1265 }
1266 
1267 static const TypeInfo virtio_blk_info = {
1268     .name = TYPE_VIRTIO_BLK,
1269     .parent = TYPE_VIRTIO_DEVICE,
1270     .instance_size = sizeof(VirtIOBlock),
1271     .instance_init = virtio_blk_instance_init,
1272     .class_init = virtio_blk_class_init,
1273 };
1274 
1275 static void virtio_register_types(void)
1276 {
1277     type_register_static(&virtio_blk_info);
1278 }
1279 
1280 type_init(virtio_register_types)
1281